A Chemogenomic Screen Reveals Novel Snf1p/AMPK Independent Regulators of Acetyl-CoA Carboxylase

Acetyl-CoA carboxylase (Acc1p) is a key enzyme in fatty acid biosynthesis and is essential for cell viability. To discover new regulators of its activity, we screened a Saccharomyces cerevisiae deletion library for increased sensitivity to soraphen A, a potent Acc1p inhibitor. The hits identified in the screen (118 hits) were filtered using a chemical-phenotype map to exclude those associated with pleiotropic drug resistance. This enabled the identification of 82 ORFs that are genetic interactors of Acc1p. The main functional clusters represented by these hits were “transcriptional regulation”, “protein post-translational modifications” and “lipid metabolism”. Further investigation of the “transcriptional regulation” cluster revealed that soraphen A sensitivity is poorly correlated with ACC1 transcript levels. We also studied the three top unknown ORFs that affected soraphen A sensitivity: SOR1 (YDL129W), SOR2 (YIL092W) and SOR3 (YJR039W). Since the C18/C16 ratio of lipid acyl lengths reflects Acc1p activity levels, we evaluated this ratio in the three mutants. Deletion of SOR2 and SOR3 led to reduced acyl lengths, suggesting that Acc1p is indeed down-regulated in these strains. Also, these mutants showed no differences in Snf1p/AMPK activation status and deletion of SNF1 in these backgrounds did not revert soraphen A sensitivity completely. Furthermore, plasmid maintenance was reduced in sor2Δ strain and this trait was shared with 18 other soraphen A sensitive hits. In summary, our screen uncovered novel Acc1p Snf1p/AMPK-independent regulators.


Introduction
Cytosolic acetyl-CoA carboxylase (ACCase) plays a key role in lipid metabolism. This enzyme catalyses the carboxylation of acetyl-CoA, producing malonyl-CoA, a critical intermediate in fatty acid biosynthesis. In Saccharomyces cerevisiae, two ACCase isoforms were found, Acc1p, located in the cytoplasm, and Hfa1p, located in the mitochondrial matrix and less studied. Acc1p is essential in all organisms Even the supplementation of growth medium with longchain fatty acids does not reverse the lethality of its deletion in yeast [1]. This suggests that a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 scored (the scoring method is described in the Materials and Methods section). The growth of 156 mutants was inhibited, compared with no-drug YPD controls (S1 Table). To verify these preliminary hits, we rescreened the mutants under the same growth conditions, but using a multi-spot assay, where each strain was plated at three different cellular dilutions. This allowed us to assess their sensitivity to soraphen A on a semi-quantitative scale (Fig 1A). Using this assay, 127 hits were confirmed (S2 Table), and nine strains with deleted dubious ORFs were excluded from the dataset at this point. According to the soraphen A sensitivity score, these hits were distributed from 1 to 5, and 55% of 118 confirmed hits showed moderate to high sensitivity (score from 3 to 5) (Fig 1B and S2 Table).
To verify the reliability of the data, we searched for previously reported soraphen A sensitive strains within the literature. We successfully identified four such sensitive strains: sit4Δ, sap190Δ, sur4Δ and fen1Δ [3,22,24]. We also observed some resistant mutants (data not shown), including snf1Δ and snf4Δ. During the course of our work, results of a large automated screening of heterozygous and homozygous S. cerevisiae deletion collections encompassing 1800 compounds, including soraphen A, were published [25]. Although the results of soraphen A screen were not further explored by the authors, this work identified 86 hypersensitive mutants The published approach by Hoepfner et al. is different from ours. First, the published study employed a diploid and not a haploid collection, including both homozygous and heterozygous mutants, which allowed the investigation of essential genes excluded from our collection. Second, the authors performed the screen in liquid medium using pools of mutants. Finally, soraphen A sensitivity was assessed by RT-PCR (as each mutant in the library had a unique identifying sequence code). Despite these differences, we identified 21 common hits (S2 Table, labelled as "coincident genes"), which represented 26% of our original hits. Altogether, we conclude that our data are reliable.
As our main goal was to identify Acc1p interactors, we addressed if our results could be biased by the inhibition of a second acetyl-CoA carboxylase, Hfa1p, localized in the mitochondrial matrix. To the best of our knowledge, there is not notion in the literature whether Hfa1P is also a soraphen A target. Hfa1p and Acc1p are thought to have evolved from a single gene [26]. Soraphen A interacts with 18 amino acids in the BC domain [20]. To investigate whether soraphen A binds to Hfa1p, BC domains of both isoforms were aligned (Fig 2A). We discovered 73% identity and 84% positive substitutions between the sequences. Furthermore, from the soraphen A-interacting residues, only residue F512 was not conserved between Hfa1p and Acc1p. This suggested that both enzymes could potentially be inhibited by soraphen A. To address if the possible inhibition of Hfa1p by soraphen A could interfere with our results, first we sought for hfa1Δ strain among our hits, this results was further confirmed using a multispot semi-quantitative assay. Afterwards we tested the drug's effect on growth of WT strain on glycerol, a respiratory substrate. Hfa1p is essential for respiration by providing substrates for the mitochondrial type II fatty acid biosynthetic pathway (FAS II). If both ACC1p and Hfa1p isoforms contributed to sensitivity we expected a higher IC 50 for soraphen A in glycerol but as shown in Fig 2B, the IC 50 for soraphen A was not significantly different regardless of the carbon source used. Since Hfa1p provides substrates for FAS II, its activation in cells treated with soraphen A could alleviate inhibition of Hfa1p. FAS II is a multimeric complex and the loss of any of its subunits impairs mitochondrial fatty acid synthesis. We therefore checked whether deletion of some of these subunits (Cem1p, Oar1p, Htd2p and Etr1p) affected soraphen A sensitivity, but none of the mutant strains were more sensitive to the drug than WT (S1A Fig). Finally,we searched for negative HFA1 interactors among our screen hits, i.e. genes whose deletion in an hfa1Δ background led to reduced growth under standard conditions ( Fig 2C). The screen poorly identified HFA1 negative interactors. Out of 50 HFA1 negative interactors reported thus far, just three of them, CPT1, PBS2 and TPS2, were identified in our screen; only TPS2 was retained after we filtered our data with a chemical-phenotype map (vide infra). Also the results of the screening of Hoefner et al., did not found any HFA1 interactor. Under the conditions tested, the evidences point out that the possible inhibition of Hfa1p does not interfere with our results. Therefore we considered that our hits refer to Acc1p regulation.

Chemical-phenotype analysis reveals 82 novel putative regulators of Acc1p activity
We considered as ACC1p interactors ORFs that fit at least one of the following three possibilities: i) ORFs whose deletion constitutively reduce Acc1p activity, ii) ORFs that are important upon inhibition of Acc1p; or iii) ORFs whose deletion increases cellular demand for Acc1p activity. To identify these interactors, we applied a more stringent filter to the 118 confirmed hits, to exclude genes associated with non-specific responses to drugs, such as drug extrusion or detoxifying systems. We also excluded strains where soraphen A sensitivity was potentially linked to a general inhibition of lipid metabolism. We took advantage of a previously published chemical-phenotype network (CP network), which is a bipartite hybrid network comprising  [20]. The residues (18) that were previously shown to interact with soraphen A in Acc1p BC domain [20] are marked in yellow. The regulatory serine-77 is marked in red. Note that Hfa1p translation begins at a non-canonical (Ile) start codon in position -71 relative to the annotated initial methionine, marked in black [26,27]. (B) Soraphen A sensitivity in fermentative and respiratory substrates. WT strain was grown for 48 h in liquid YP medium containing 2% glucose (YPD) or 2% glycerol (YPGly) in the presence of increasing concentrations of soraphen A. Soraphen A IC 50 values were determined as described in Methods section. Results are the means of four independent experiments ± standard deviations (non-paired t-test; ns, non-significant).(C) The screen does not select negative interactors of HFA1. hfa1Δ strain shows no growth defects when grown under standard conditions in YPD; however, cell growth is impaired upon deletion of gene A, a negative interactor. If Hfa1p would be inhibited by soraphen A, a strain with a deletion of gene A (or any negative interactor of HFA1) would be more sensitive to the drug. That was not the case as only 3 out of 50 negative interactors were identified in the screen (CPT1, PBS2 and TPS2), however they were filtered in subsequent analysis. 426 chemicals and 5233 ORFs whose deletion leads to sensitivity to those compounds [15]. Using these data, we constructed a subnetwork, the Lipid Chemical Phenotype (LipCP) network (S3 Table), consisting only of drugs that are known to affect lipid metabolism (e.g. imidazoles, triazoles and polyenes; S4 Table) ( Fig 3A). Then, we evaluated the relative connectivity of our hits in both CP and LipCP networks using a rank plot analysis [28] (the connectivity scores obtained in each network are presented in S5 Table). The rank plot presents ORFs in four quadrants, highlighting their relative importance (based on degrees) in either network ( Fig 3B).
We assumed that the Acc1p interactors must show low connectivity in both CP and LipCP networks, and would be localised in the third quadrant of the rank plot. We selected the appropriate hits and adopted yet another, even more stringent filter; we arbitrarily discarded hits with scores higher than 0.4 in both the CP and LipCP networks. We thus obtained a final group of 82 Acc1p interactors (Table 1). Strikingly, 40% (37) of the ORFs were not present in the LipCP network (score 0), which suggested that indeed they were highly specific to Acc1p inhibition.
ORFs with high scores in both networks (second quadrant) are systematically identified in chemogenomic screens. They are associated with general drug response mechanisms, such as PDR5, the main ABC transporter responsible for soraphen A extrusion [29], and WHI2, involved in general stress response. ORFs poorly connected to general drugs (low score in CP network) but scoring high in the LipCP network (fourth quadrant) were probably identified in our screen because they could be involved in mechanisms that globally affect lipid metabolism. Thus, the effect of their deletion on soraphen A sensitivity might not be restricted to Acc1p activity regulation. For example, IRA2 from the Ras-cAMP signalling pathway was previously thought to participate in the regulation of carbon fluxes into storage lipids [30], which might affect lipid metabolism in general. Thus, we are confident that the hits in these remaining quadrants are not Acc1p modulators.

Functional analysis of novel Acc1p regulators
The final Acc1p interactors (82) were clustered according to their physiological roles (Fig 4). The major groups were "transcriptional regulators", "post-translational modification of proteins", "lipid metabolism", "ORFs with unknown function" and "vesicle trafficking". Because of the considerable amount of information generated by our screen, we decided to focus on "transcriptional regulators" and "ORFs with unknown function" clusters.
The transcriptional cluster contained 17 genes. Interestingly, INO2 and INO4 were not identified in our screen nor in the screen by Hoepfner et al. (2014), even though their deletion down-regulates ACC1 transcription levels [1]. Surprisingly, a more detailed analysis revealed that out of 20 transcription factors (TFs) reported to affect ACC1 transcription [31,32], just two, DAT1 and SPT20, were identified in our screen and none of them was identified by Hoepfner et al. (2014). Because of the high discrepancy between literature and our results, we re-tested all 20 TFs plus TF-deleted strains found only in our screen (spt2Δ and pho23Δ) ( Fig  5) and compared their soraphen A sensitivities with the reported levels of ACC1 transcription [32]. Indeed, our screen missed just 3 of 20 re-tested strains (SFP1, TUP1 and SPT4) which suggests that soraphen A sensitivity and ACC1 transcription are poorly correlated, These three missing mutants were then submitted to the same chemical-phenotype analysis as before ( Fig 3B, blue circles), and SFP1 and SPT4 were included in our final set of candidate ORFs (Fig  4, diamond-shaped nodes). In conclusion, six hits (sfp1Δ, spt4Δ, dat1Δ, spt20Δ, spt2Δ and pho23Δ) are good TF candidates for further studies focused on transcriptional regulation of ACC1.

Three ORFs with unknown function affect Acc1p activity
The main challenge following genome sequencing comprises the characterization of proteins with unknown function that are often divergent or lineage-specific. In the present study, we  [15]. An ORF is connected to a drug when its deletion increases sensitivity to that drug. LipCP network was extracted by choosing drugs that target the lipid metabolism (S3 Table) and their connected ORFs from the CP network (S4 Table). (B) ORFs identified among our final hits (S2 Table), plotted according to their relative connectivity in both CP and LipCP networks. The score represents the number of drug connections of an ORF in each network, normalised so that the most connected ORF scores 1 (S5 Table). Quadrants are numbered as shown. ORFs that are located in the third quadrant and with degree lower than 0.4 in both networks, represented by dotted lines, were considered to be primarily associated with soraphen A treatment. Final hits are marked as black dots. Blue circles represent ORFs that were missed by the screen but were added in later after individual re-testing (SPT4, SFP1 and TUP1; please refer to the text). focused on the top three most-sensitive candidates with unknown function: YDL129W, YIL092W and YJR039W, which we called SOR1, SOR2 and SOR3, respectively, as in "SORaphen sensitive mutants". SOR1, SOR2 and SOR3 encode proteins with molecular weights of 32.92 kDa, 71.02 kDa and 127.42 kDa, respectively. All of them were found to be unique to fungi, as verified by BLAST (data not shown). To test whether these hits were direct Acc1p modulators, we evaluated their triacylglycerol and steryl ester levels, and compared these to levels in soraphen A-treated cells. Treatment with a sub-lethal concentration of soraphen A (0.1 μg/ml) reduced the triacylglycerol, but not steryl ester, levels in WT cells (Fig 6A and 6B). Interestingly, in sor1Δ sor2Δ and sor3Δstrains, no significant differences in triacylglycerol or steryl ester levels were observed. This result was unanticipated; however, we next investigated the length of the acyl groups in the total cellular lipid fractions by GC/MS ( Fig 6C, Table 2), as it has been recently shown that such parameter is affected by the activity of Acc1p [10,33]. Indeed, inhibition of Acc1p by soraphen A reduced the C18/C16 ratio by~40% in WT cells.
We also observed a reduction in the C18/C16 ratio in sor2Δ and sor3Δ cells (~20% in both) but  Table) were filtered by rank plot analysis employing the chemical phenotype network (Fig 3). The ORFs were manually clustered according to their description in Saccharomyces Genome Database (SGD, www.yeastgenome.org). Node size represents the inhibition of growth score in the presence of soraphen A (from 5 to 1). Diamond-shaped nodes represent ORFs that were missed by the screen but added in later after individual re-testing ( Fig 5). In these cases, no sensitivity score was assigned. Thicker nodes represent ORFs that have not been previously associated with lipid metabolism inhibition (LipCP network score equal to zero).
doi:10.1371/journal.pone.0169682.g004 not in sor1Δ cells. Taken together, these results suggested Sor2p and Sor3p as probable Acc1p regulators. The Snf1p/AMPK pathway is the most studied Acc1p regulatory pathway. Therefore, we investigated whether Sor2p and Sor3p regulate Acc1p activity via that pathway or whether they participate in novel regulatory pathways. We deleted each SOR gene in a snf1Δ background, which is hyper-resistant to soraphen A (IC 50 = 0.95 μg/ml) when compared with WT (IC 50 = 0.19 μg/ml) [22,23] Soraphen A sensitivity of sor1Δ snf1Δ mutant was unaltered in relation to snf1Δ, while in sor2Δ snf1Δ and sor3Δ snf1Δ, it was~50% and~30% lower than that measured for snf1Δ cells (Fig 6D). These results indicated that SOR2 and SOR3 act independently of Snf1p. Furthermore, we found no difference between Snf1p phosphorylation levels in SOR knockout mutants (Fig 6E), which suggested that the Snf1p/AMPK pathway was not compromised in these mutants. In these experiments, it was striking that sor2Δ was unable to retain a plasmid containing an epitope-tagged Snf1p. This phenotype was reproduced in independent clones, was exclusive to SOR2 deletion and is further explored in the following section.

Reduced plasmid maintenance in sor2Δ mutant
The failure of sor2Δ cells (Fig 6E) to maintain a plasmid encoding hemagglutinin-tagged Snf1p (Snf1p-HA), led us to investigate whether this was caused by SNF1 overexpression. To test this, we also transformed WT and sor2Δ cells with an empty vector and compared the stability of both plasmids stabilities. The results revealed that, under selective pressure, SNF1 overexpression did not affect plasmid maintenance in WT as both plasmids were maintained by more than 95% of the cells (Fig 7A). On the other hand, in sor2Δ strain, both plasmids were maintained in only~10% of the population. We also tested whether the plasmids were lost as a consequence of reduced Acc1p activity (Fig 7B). We found that Acc1p inhibition by soraphen A did not induce plasmid loss in WT cells. We also tested if plasmid loss in sor2Δ strain was that affect ACC1 transcript levels were retested for soraphen A sensitivity as before (Fig 3). Published ACC1 transcript levels in these mutants [32] are shown. TFs mutant strains are marked in green or red, depending on whether ACC1 transcription was down-regulated or up-regulated, respectively [31]. TF mutant strains identified by our screen but with unaltered ACC1 transcription [30] are marked in black. The lys2Δ strain was employed as a control. The results are representative of four independent experiments. 1, 2 Two yjl206cΔ strains were present in our collection and both were tested. due to decreased repair of DNA, which could be a cause for genomic instability. However, sor2Δ was as sensitive as WT to the DNA damaging agents hydroxyurea (100mM) and methyl methanesulfonate (0.02%).

Discussion
We screened a yeast deletion library to identify genes whose deletion conferred sensitivity to soraphen A, a potent Acc1p inhibitor, with the aim of finding new regulators or processes that are linked with Acc1p activity. The data generated were reliable as the screen identified all  Novel Regulators of Acetyl-CoA Carboxylase Activity deletion strains previously reported as soraphen A sensitive. From the strains that were retested, we estimated a 10% rate of false negatives (data not shown). Chemogenomic screens are prone to identify mutants that affect drug sensitivity via unspecific mechanisms, as in the case of pleiotropic drug resistance. To overcome this issue, we employed a biosystematic approach. We successfully discarded 30% of the 118 initial hits, including PDR5, an ABCtransporter responsible for pumping soraphen A out of the cell [29]. By analysing the physiological functions of our final hits, we found that the major groups comprised "transcriptional regulators", "post-translational modification of proteins", "lipid metabolism", "ORFs with unknown function" and "vesicle trafficking". Further analysis of the "transcriptional regulators" cluster revealed that from 23 TFs previously reported as putative ACC1 regulators [12,34,31,32], the deletion of only six of them correlated with altered soraphen A sensitivity. ACC1 presents an upstream activation sequence recognised by the Ino2p/ Ino4p complex and it was reported that ACC1 transcription is up-regulated in an opi1Δ strain and down-regulated in ino2Δ and ino4Δ strains [1]; however, these strains did not present abnormal soraphen A sensitivity. This is in agreement with the poor correlation between mRNA and protein levels of many genes [35]. The final six TFs presented here (SFP1, SPT4, DAT1, SPT20, SPT2 and PHO23) are promising targets for Acc1p modulation.
Together with SIT4 and SAP190, in the "post-translational modification of proteins" cluster, we found putative regulatory elements for further study (Fig 5), such as catalytic or regulatory subunits of kinase complexes. None of these ORFs have been previously shown to interact physically or genetically with Acc1p and it would be interesting to investigate whether these kinases are directly involved in Acc1p phosphorylation. We also found five ORFs associated with ubiquitination (UBP3, UBR2, MUB1, SLX5 and RAD16). There is evidence that the Ubp3p/Bre5p complex participates in Acc1p regulation: i), these two proteins interact physically with Acc1p [36]; and, ii), it has been demonstrated that acetyl-CoA carboxylase is subject of ubiquitin-mediated proteolysis in mammalian cells [37], although this phenomenon remains unclear in yeast.
A major concern that arose from our screen was that the mitochondrial acetyl-CoA carboxylase, Hfa1p, could also have been inhibited by soraphen A, as it shares with Acc1p all the residues previously shown to interact with soraphen A, except for F512. Our screen and the screen by Hoepfner et al. [25] failed to identify HFA1 negative interactors. Also we observed that loss of any of the FASII subunits, that impair mitochondrial fatty acid synthesis as well as HFA1 deletion, are not more sensitive to soraphen A. These results suggested that mitochondrial fatty acid synthesis does not play a role in the cellular response to soraphen A. The different conditions tested strongly indicate that the chemical-genetic interactions observed in this work are specific for Acc1p. However, to reach the conclusion that Hfa1p is not inhibited by soraphen A, this isoform should be purified and the enzymatic activity assayed in the presence of soraphen A.
From the ten ORFs with unknown functions, deletion of three of them resulted in a moderate to high sensitivity to soraphen A. These ORFs were named as SOR1 (YDL129W), SOR2 (YIL092W) and SOR3 (YJR039W). To check whether these ORFs were constitutive Acc1p regulators, we measured the lengths of the acyl chains of cellular lipids in corresponding mutant strains. It has been shown that the reduction of mammalian acetyl-CoA carboxylase activity by soraphen A or its activation in yeast after a deletion of SNF1 lead to shorter or longer acyl was grown for 48 h in liquid SD medium in the presence of selective pressure (without uracil), with the indicated concentrations of soraphen A. Plasmid stability was assessed as in (B) and the data are presented as the means from two experiments. doi:10.1371/journal.pone.0169682.g007 Novel Regulators of Acetyl-CoA Carboxylase Activity chains, respectively [10,33]. We observed reduced C18/C16 acyl ratios in sor2Δ and sor3Δ strains, as well as in WT treated with soraphen A. This indicated that Acc1p was constitutively low in these strains, suggesting that SOR2 and SOR3 are Acc1p regulators. The exact mechanisms by which Sor2p and Sor3p participate in Acc1p regulation remain elusive. However, it is clear that they do not involve indirect regulation via the modulation of Snf1p/AMPK activity (as already shown for SIT4 and SAP190, for example [22], since Snf1p/AMPK phosphorylation levels were not altered in these strains and deletion of SOR2 and SOR3 in a snf1Δ background partially reversed soraphen A resistance. We cannot make such conclusions regarding SOR1, however, as its deletion did not affect the C18/C16 ratio. However, it is important to mention that when grown on alternative carbon sources, such as glycerol, sor1Δ sensitivity was reversed, which was not the case of sor2Δ and sor3Δ (data not shown). We hypothesized that deletion of SOR1 could either reduce cell fitness (which would increase its dependence on Acc1p activity) or impair any backup system responsible for sustaining Acc1p activity upon soraphen A treatment. Both hypotheses are still to be tested.
Finally, we observed that plasmid maintenance in sor2Δ strain was abnormal, which led us to hypothesize that genomic instability and Acc1p activity may be functionally related. Although plasmid/chromosome maintenance may be achieved by several different mechanisms, we observed a strong enrichment of mutants previously reported as genomic unstable among our hits. Interestingly, it has been previously shown that histone acetylation is associated with genomic stability [38] and increased H4K16 acetylation or loss of H4K16 deacetylase SIR2 leads to genomic instability. Accordingly, sor2Δ could inhibit ACC1 to increase the levels of acetyl CoA available for histone acetylation. However, soraphen A treatment of WT cells did not affect plasmid maintenance indicating that the problems in plasmid maintenance presented in the sor2Δ strain is not a consequence solely of the reduced ACC1p activity. On the other hand, it could be that the genomic instability could be the cause of the decrease in Acc1p activity. To test whether we could find a correlation between genes involved in genomic instability and the modulators of Acc1p, we re-analysed our screen primary hits. Among the 118 hits, we found 18 ORFs (15.3%, labelled as "genomic instability" in S2 Table) whose deletion affected plasmid/chromosome stability according to Saccharomyces Genome Database (SGD, www.yeastgenome.org) (as in August, 2016). This frequency of occurrence was higher than that observed in our knockout collection (279 out of 4860 strains, i.e. 5.7%); therefore, this 2.5 fold enrichment was highly significant (P < 0.0001, hypergeometric distribution). As "genome instability" is a phenomenon that can be triggered by dysfunction of several processes it is difficult to determine which of these directly leads to ACC1 down-regulation. Further work should be done to address this question.

Reagents
Soraphen A was a gift from Rolf Müller (Helmholtz-Zentrum für Infektionsforschung, Braunschweig, Germany) and was maintained as a 0.1 mg/ml stock solution in 10% methanol. All other reagents were from Sigma-Aldrich (St. Louis, MO, USA).

Strains and media
S. cerevisiae strain BY4741 (MATa his3Δ1 leu2Δ0 met15Δ0 ura3Δ0) and yeast deletion library containing~4860 strains where non-essential ORFs were individually replaced with a G418 (geneticin) resistance gene were obtained from Open Biosystems. The deletion collection was recovered from glycerol stocks via replica plating on rich YPD solid medium (1% yeast extract, 2% peptone, 2% glucose and 2% agar) at 30˚C for 48 h. We verified the correct deletion of YDL129W, YIL092W and YJR039W genes by PCR and confirmed these to be the correct strains (data not shown). Double mutants sor1Δ snf1Δ, sor2Δ snf1Δ and sor3Δ snf1Δ were constructed by mating sor1Δ, sor2Δ or sor3Δ from the deletion library with a snf1Δ (MATα) strain and incubating in sporulation medium (0.02% raffinose, 0.3% potassium acetate) for 7 d. Double mutants were selected on glucose-MSG medium (2% glucose, 0.1% monosodium glutamate, 0.67% yeast nitrogen base without amino acids or ammonium sulphate) containing nourseothricin (100 μg/ml and geneticin (200 μg/ml), supplemented with histidine, leucine, methionine and uracil (0.003% final concentration of each). Isolated colonies were randomly picked and the presence of SOR and SNF1 loci confirmed by PCR. Plasmid maintenance tests employed sor2Δ and WT transformed with pYC-SNF1:3HA plasmid, derived from pYC210, following the lithium acetate transformation method [39]. As means of selective pressure in these tests, uracil was removed from SD medium.

Screening and scoring
Strains were spotted in triplicate with a 96-pin replicator onto a new YPD plate in the absence (control plate) or presence of 0.2 μg/ml or 0.4 μg/ml of soraphen A, based on the IC 50 value previously reported by our group [22]. Growth was registered after 72 h of incubation at 30˚C, as all strains should have reached stationary phase by that time point. The screening was performed twice. To identify sensitive mutants, the growth of each strain in the presence of soraphen A was compared with its growth in the absence of soraphen A, and an arbitrary score was assigned. The scores ranged from 5 to 0: a score of 0 reflected no difference in growth on drug-containing and control plates; a score of 5 reflected no growth in the presence of soraphen A. The final score was the sum of the scores obtained in the individual experiments. Strains with total scores equal to or higher than 5 were selected (S1 Table) and assessed in a multi-spot assay. These strains were grown in 96-well microplates filled with 0.2 ml of liquid YPD. Plates were incubated at 30˚C under shaking for 48 h. From these pre-cultures, 10, 100 and 1000-fold dilutions were prepared in fresh 96-well microplates and were immediately spotted onto YPD solid medium supplemented with soraphen A with a replicator. Plates were incubated at 30˚C. Growth was recorded after 48 h. Each plate received a score from 5 (full growth) to 0 (no growth). Final sensitivity score was calculated as the score on control YPD plate minus the sum of the scores obtained in the presence of soraphen A (some examples were shown in Fig 1A) (S2 Table).

LipCP network
LipCP network is a subnetwork of the previously published CP network [15]. Briefly, the CP network consists of 426 chemical compounds with their chemical-genetic interactions in S. cerevisiae identified. To integrate these data with our results, we constructed a LipCP network by selecting drugs that are known to primarily affect lipid metabolic processes. We found 20 drugs that interfere with lipid biosynthesis/catabolism (e.g. cerulenin, azoles and peroxisome activators) and five sterol-binding compounds (e.g. polyenes, saponins) (S4 Table). Thus, the LipCP network encompassed these 25 drugs and their interactions with 3181 ORFs (edges and detailed information can be found in S3 Table).

Network rendering and databases
The networks that are graphically represented in this paper were rendered with Cytoscape software [40]. Interaction data were obtained from BIOGRID [41]. Transcription factor information was obtained from YEASTRACT [42].

Soraphen A sensitivity tests
For pre-cultures, cells were inoculated in liquid YPD and incubated for 24 h at 30˚C with shaking. To confirm soraphen A sensitivity, OD 600 was measured and pre-cultures were diluted in a microplate to obtain 10 7 , 10 6 and 10 5 cells/ml suspensions. Cellular suspensions were immediately spotted onto YPD solid medium supplemented and soraphen A (as indicated in the figures) with a replicator.
Plates were incubated at 30˚C and pictures were taken after 2 d. For IC 50 determinations, a 96-well microplate was filled with 100 μl/well of fresh YPD with increasing concentrations of soraphen A. Cellular suspensions were diluted to OD 600 = 0.01 and 100 μl was added to each well (final concentrations of soraphen A ranged from 4 to 0.03 μg/ml). Plates were incubated at 30˚C. Growth was recorded after 24 h by measuring absorbance at 600 nm in a plate reader (Spectramax 5M, Molecular Devices). The IC 50 value was defined as the concentration of the drug required for 50% growth inhibition compared with the growth of drug-free controls. In both experiments, it was necessary to employ lys2Δ strain as the control since we observed that strains from the MATa deletion library were slightly more tolerant to soraphen A and other stressors than WT BY4741 [43].

Lipid analysis
Yeast cells were grown for 24 h in YPD medium (soraphen A was added to 0.1 μg/ml final concentration, as indicated). Lipids were extracted and separated by thin-layer chromatography (TLC) as previously described [44,45]. Triolein, oleic acid and cholesteryl oleate were used as standards.
For acyl chain length determinations, 50 μg of tripentadecanoin, an internal standard, was added to cells prior homogenization. Total lipid extracts were dried under a stream of nitrogen and kept at -20˚C until derivatization. For fatty acid derivatization, 500 μl of NaOH/methanol (20 g of NaOH in 1 L of methanol) was added to dried extracts and the tubes were incubated in boiling water for 15 min. After cooling, 500 μl of BF 3 -methanol 10% was added and the tubes were incubated as before. After cooling, 300 μl of NaCl-saturated aqueous solution was added. Finally, 300 μl of hexane was added to the tubes and, after homogenization and separation of phases, the organic phase was transferred to gas chromatography/mass spectrometry (GC/MS) glass vials.
Samples (2 μl) were injected into a DB-5ms column in a splitless mode using an autoinjector. Helium was used as a carrier gas. The inlets and MS source were maintained at 250˚C. The oven temperature was maintained at 80˚C for 4 min, then ramped to 170˚C at a rate of 8˚C/ min and held for 10 min. The temperature was then ramped to 250˚C at a rate of 4˚C/min and held for 5 min.

Snf1p/AMPK phoshorylation status
Yeast cells were grown for 48 h in YPD liquid medium and, then, inoculated in fresh YPD (initial O.D. 600nm 0.25). Flasks were incubated under shaking (200 rpm) at 30˚C and growth was followed till O.D. 600nm~1 .0 (after 5 h). Culture samples were collected and total protein homogenates were prepared according to Yaffe and Schatz method [46] with minor modifications. NaOH and β-mercaptoethanol were added to a final concentration of 0.2 M and 1%, respectively. Tubes were incubated on ice for 2 minutes and, then, trichloroacetic acid was added to a final concentration of 6%. After incubation on ice for 10 minutes, tubes were centrifuged at maximum speed for 5 minutes. Supernatant was discarded and pellets were stored at -20˚C until use or promptly resusspended in Laemlli sample buffer to obtain a 10-O.D. suspension. Proteins were separated by electrophoresis (SDS-PAGE, 10% SDS-polyacrylamide gel, 100 V, 2 hours) and electrotransferred to Immobilon-P for 20 min at 18 V in 25 mM Tris, 192 mM glycine and 10% methanol, using a trans-blot semi-dry cell (BioRad). Membranes were treated with anti-phospho-AMPKa Thr172 (Cell Signaling) or anti-HA (Sigma) antibodies. Blots were detected using the chemiluminescence ECL Plus kit (GE Healthcare).

Plasmid maintenance test
For plasmid maintenance testing, sor2Δ or WT cells were transformed with pYC210 (empty vector) or pYC:SNF1-3HA plasmids and pre-grown under selective pressure in SD liquid medium without uracil (SD-ura) for 48 h at 30˚C with shaking. Then, the selective pressure was removed by diluting the pre-cultures 100 times in SD liquid medium containing uracil (SD +ura) and the incubation was repeated. Pre-cultures (with selective pressure) and cultures (no selective pressure) were plated on both SD +ura (for total colony-forming units (CFU) assessment) and SD-ura (for plasmid-containing CFU assessment) solid media; 200 cells were plated for each condition and plates were incubated at 30˚C. After 3 d, UFCs were manually counted. The percentage of plasmid-containing cells was calculated as the number of CFUs on SD-ura plate divided by the number of CFUs on SD +ura plate, and multiplied by 100.

S1 Fig. Deletion of mitochondrial FASII subunits does not lead to altered sensitivity to soraphen A.
Mutant strains with deletion of genes encoding FASII pathway enzymes were retested for soraphen A sensitivity as before (Fig 1). Soraphen A was added to the media at the indicated concentrations. Plates were incubated at 30˚C for 3 d and growth was registered. The lys2Δ strain was employed as a control. The results are representative of three independent experiments. (TIF) S1 Table. Sensitivity scores for null strains in the soraphen A screen, as described in the Material and Methods section. Only strains considered to be sensitive are listed. (XLS) S2 Table. Final sensitivity scores for null strains after confirmation by multi-spot assay (as described in the Material and Methods section). Only confirmed strains are presented. "Coincident genes" refers to ORFs also identified in [25]. "Genomic instability" refers to ORFs identified as affecting plasmid/chromosome maintenance according to SGD. (XLSX) S3  Table. Sensitivity scores for each identified ORF in both the original CP network and in LipCP network. Scores represent node degrees and were normalised based on the highest value of a node in each network. These results are visualised as a rank plot in Fig 4. (XLSX)