Defining the Yeast Resistome through in vitro Evolution and Whole Genome Sequencing

In vitro evolution and whole genome analysis were used to comprehensively identify the genetic determinants of chemical resistance in the model microbe, Saccharomyces cerevisiae. Analysis of 355 curated, laboratory-evolved clones, resistant to 80 different compounds, demonstrates differences in the types of mutations that are identified in selected versus neutral evolution and reveals numerous new, compound-target interactions. Through enrichment analysis we further identify a set of 137 genes strongly associated with or conferring drug resistance as indicated by CRISPR-Cas9 engineering. The set of 25 most frequently mutated genes was enriched for transcription factors and for almost 25 percent of the compounds, resistance was mediated by one of 100 independently derived, gain-of-function, single nucleotide variants found in 170-amino-acid domains in two Zn2C6 transcription factors, YRR1 and YRM1 (p < 1x 10 −100). This remarkable enrichment for transcription factors as drug resistance genes may explain why it is challenging to develop effective antifungal killing agents and highlights their important role in evolution.


Introduction
Over the past decade, decreases in sequencing costs have led to an explosion in the number of cataloged genetic variants in all fields of biology. In a recent sequencing study of 1100 yeast isolates 1,625,809 single nucleotide variants (SNVs) were identified 1 . Sequencing of thousands of mosquitoes that cause human malaria identified 57 million variants 2 . There are now 660 million cataloged human variants 3 . The challenge lies in efficiently identifying variants that change the phenotype of tumor, pathogen or agricultural pest especially when the genetic background is heterogenous and there may be tens of thousands of differences between two sequenced isolates.
Systems for studying what these SNVs do for the cell have lagged. Systematic functional genomic studies have continued to rely on strain libraries in which the entire coding region is modified.
For example, after the yeast genome was sequenced a set of homozygous and heterozygous knockout strains was constructed which bear deletions in all genes in the genome 4,5 . This set has been used to repeatedly and systematically identify knockout/knockdown lines that show sensitivity or resistance to a wide variety of different compounds 6 and remains important 7 . CRISPR-based, genome-wide knockout and knockdown studies are now being employed in many organisms to identify drug targets 8 or study processes such as the emergence of cancer drug resistance. Such studies will miss gain-offunction SNVs, which often drive natural adaptive evolution.
Although systematic CRISPR-Cas9 based analyses of SNVs are also feasible 9 with the reduction in costs of whole genome sequencing, experimental evolution, which mimics natural evolution, becomes more attractive. Here we exposed the model yeast to a large set of compounds, similar to cell-permeable small molecules or natural products that fungi might encounter in their natural environment or which might be used in agriculture or medicine. Whole genome sequencing of 355 curated, evolved, compound-resistant clones showed only a few new coding variants per clone and that statistical approaches can be used to readily identify variants that modify phenotype. We discover an enrichment for gain-of-function variants that affect transcription. These data may provide clues about why it is challenging to develop small molecule therapeutics against fungi.

Building a library of compounds that are active against a drug-sensitive yeast
To understand how yeast evolve to evade the action of small molecules, we first assembled a collection of molecules and evaluated their activity against the yeast S. cerevisiae. Specifically, we tested compound libraries comprised of (1) drugs approved for human use, (2)  Those that showed at least 70% growth inhibition were subsequently tested in dose response. To increase the probability of finding compounds with activity against yeast at physiologically-relevant concentrations, and given that compound cost, availability and resupply are major impediments to a study such as this, we used a sensitized strain of yeast, termed the "green monster (GM)" in which a variety of ABC transporters have been replaced with GFP 10 .
Overall, the compounds of the assembled collection had drug-like physiochemical properties in terms of molecular weight and the number of hydrogen bond donors and acceptors ( Figure 1A).
Maximum Common Substructure (MCS) clustering identified 307 clusters with a similarity coefficient of 0.64 ( Figure 1B, Table S1). Altogether, 286 compounds had an IC50 < 76 µM, and 98 compounds had an IC50 < 10 µM (Table S1). Of these 286 active compounds, 165 share a MCS with at least one other compound (Table S1). Cluster enrichment was observed at rates greater than expected by chance for some compounds. For example, there were 12 members of a benzothiazepine family (cluster 185, Figure 1B) in the tested group of 1,618 compounds of which six had an IC50 of less than 10 µM in the yeast model (p = 2.46 x 10 -5 ).

In vitro resistance evolution and whole genome analysis link compound structures to phenotype
Based on potency and compound availability, we selected 100 active compounds from this collection for follow-up in vitro evolution experiments. To select for resistant strains, yeast cultures (approximately 10 8 -10 9 cells) were first inoculated with sublethal compound concentrations. We subsequently ramped up the selection pressure by serial dilution of the saturated cultures into media with increased compound concentrations until resistance was observed as measured by an increase in IC50 values. Using this strategy, we successfully isolated 355 strains resistant to 80 compounds.
Cultures were considered resistant if they (1) continued to grow at compound concentrations above the IC50 value of the untreated culture, and (2) had at least a 1.5-fold shift in IC50 value compared to the drug-naïve parental line (Table S2). These resistant cultures were plated on drug-containing plates to isolate single colonies. Genomic DNA was isolated from these clones, and whole genome sequencing of both the resistant and parental strains identified mutations associated with resistance.
The IC50 values of the resistant clones increased 1.5-to 5-fold for 121 resistant strains, 5-to 10-fold for 101 resistant clones, and > 10-fold for 98 resistant strains (Table S2). In about 20% of the cases, we were unable to isolate resistant strains after many weeks of selections, often because of contamination or poor compound availability.
Next clones were sequenced to 55-fold average coverage (Table S3) using short read methodology. To detect mutations, we designed a custom whole genome analysis pipeline and filtering method (see Methods) that was automated through the computational platform, Omics Pipe 11 . Briefly, raw sequence reads were aligned to the S. cerevisiae 288C reference genome (assembly R64). SNVs and INDELs were called using GATK HaplotypeCaller 12 , filtered to retain only those of high quality and high allele fraction (appropriate for a haploid organism) and annotated with SnpEff 13 . Mutations were only considered to be potentially resistance-conferring if they were present in the evolved clone but not in the drug-sensitive parent. We discovered 1,405 high quality mutations (1,286 SNVs and 119 indels) in 731 unique genes that arose during the course of drug selection, with an average of 3.96 mutations per line (Table S4). On average we observed between 1 to 8 coding mutations per evolved clone per compound ( Figure 1C) with some variation. For example, selections with the small synthetic antifungal, tavaborole, produced four resistant clones (8-to >15-fold increase in IC50) with six total missense mutations (four in the target), while selections with the natural product, carmaphycin, resulted in three resistant clones with 28 mutations, of which 22 were coding (one in the predicted target). For the majority of compounds, we observed a strong enrichment and reproducibility for specific genes ( Figure 1D). For example, for compound MMV665852 we obtained 10 independent resistant clones with 29 independent coding mutations, of which seven were in a single gene, YRR1.
Given that yeast has roughly 6,000 genes, the Bonferonni-corrected probability of this enrichment by chance is roughly 8.38 x 10 -21 .
To further assess the likelihood that our evolved mutations would contribute to resistance we considered the types of mutations and compared these to a published set of 3,137 neutral mutations in yeast strains grown long term without compound selection 14 . We observed significantly different distributions in our drug-selected set than in the neutral set (c 2 , p < 0.0001). For example, 39% of the nucleotide base transitions for selected 1,286 single nucleotide variants (SNVs) were C to A or G to T, while for the 3,137 neutral transitions, 40% were for A to G or T to C (Figure 2A). These data provide additional evidence that the observed mutations provide a selective advantage to the evolved clones. Likewise, we observed a noteworthy difference in the coding changes. Among exonic selected mutations, 994 were nonsynonymous and 127 were synonymous (~8:1 ratio), indicating that drug treatment applied a strong positive selection, as expected ( Figure 2B). In contrast, the neutral SNVs had a strong bias toward synonymous mutations ( Figure 2C).
Copy number variants (CNVs) were also detected through a coverage-based algorithm using the output from GATK DiagnoseTargets. We observed 24 CNVs in our resistant strains (Table S5).
Unlike for studies in P. falciparum, we frequently observed aneuploidy (11 times) in addition to small, discrete, intrachromosomal amplifications (13 times). Altogether, we observed aneuploidy with eight compounds, including BMS-983970, doxorubicin, etoposide, GNF-Pf-3582, GNF-Pf-4739, hygromycin B, CBR110, and wortmannin ( Figure 2D). This is perhaps not surprising given that aneuploidies arise in S. cerevisiae as a short-term stress response 15 . We observed an amplification on chromosome XVI that involved the bZIP transcription factor, ARR1, for clones resistant to GNF-Pf-1618 and GNF-Pf-2740, as well as with four strains resistant to MMV665794. The strains, Wortmannin-13R3a and sBMH113-7R4a, both had chromosome XV CNVs that involved the transcription factors YRR1 and YRM1 (discussed below).
Our large mutational dataset also offers broad insights into the functional impacts of different variant types. Synonymous and missense variants emerged in essential genes in approximately 20% of cases. This finding agrees with the literature, which suggests that only 20% of the yeast genome encodes essential genes 16 . By this same metric, mutation types with more disruptive impacts on the resulting protein, such as premature stop codons and frameshift variants, deviate strikingly from the expected genome-wide value of 20%. These mutations occur in essential genes only 4.9% and 7.0% of the time, respectively ( Figure 2B). Of these ten disruptive mutations four were in a single gene, PAN1, involved in the regulation of endosome internalization.

Resistance-conferring intergenic mutations are rare
Although intergenic mutations are frequently found in cells not subject to selection, mutations in promoters or 3' UTRs might confer resistance by increasing or decreasing transcript levels. We mapped intergenic mutations to their nearest-neighbor coding genes (Table S6). In contrast to coding mutations, where mutations in specific genes appeared repeatedly, this analysis showed little enrichment. However, we did observe several repeated mutations in the intergenic promoter regions of a few genes. For example, we discovered three mutations upstream of the ergosterol biosynthesis and azole resistance gene, ERG9 17,18 , in addition to seven ERG9 coding region mutations. One of the intergenic mutations of ERG9 falls in the putative promoter region and was observed in selections with compound AN7973, which was also associated with two mutations in the coding region ( Figure   2E). Four coding mutations and one non-coding mutation upstream of the starting codon in the endoplasmic reticulum membrane protein and capsofungin resistance protein 19 , CSG2, were also observed. All five mutations (including the intergenic mutation) are associated with selections to compound GNF-Pf-1618 and its close analog, KAAA725 ( Figure 2F). We also identified five mutations in the coding region of PDR3 20 , a transcriptional regulator of the multidrug efflux, and an additional mutation was found downstream of the open reading frame ( Figure 2G). These data show that most mutations identified in resistant clones are coding, although intergenic mutations should not be entirely dismissed.

CRISPR/Cas9 validation shows that most genes identified more than once confer resistance, but singletons mutations may not
The presence of one or multiple SNVs in a resistant line is not proof that a specific mutation confers resistance since many mutations, so called hitchhiker mutations, can co-occur in a resistance strain and can even be non-adapative 21 . To confirm the role of these mutations in a clean genetic background, we used CRISPR/Cas9 technology to introduce 65 altered alleles from the evolved mutants back into the original (unevolved) strain. Successfully reverse-engineered strains were tested in liquid-growth assays using the same compounds from the corresponding IVIEWGA experiments.
We compared the IC50 values of these strains to those of the parental strain using a cutoff of 1.6-fold shift between edited and parental line. In total, this comparison verified that 50 genetic changes representing 39 unique genes contributed to the observed resistance ( Table S7). Mutations that were repeatedly identified tended to have a high probability of confirmation. The only exception was RPO21, a subunit of RNA polymerase, which was unconfirmed and mutated four separate times (two nonsynonymous and two synonymous mutations). For the 20 alleles that did not show a 1.6-fold change (Table S8), we noted that 11 of the resistant clones also carried additional resistance alleles in a highly represented gene such as YRR1 or YRM1 (Table S4). In addition, some of the "unconfirmed" CRISPR-Cas9 alleles resulted in a statistically-significant, gain of sensitivity. For example, a clone with an edited G454S mutation in UTP18 is 2-fold and 4-fold more sensitive to MMV665852 and CBR868, respectively, than the parent, possibly because the resistant strain has slower growth but better survival in the presence of a cytotoxic compound due to another resistance mutation. In fact, it is known that mutations in RPO21 result in transcriptional slippage, which may allow cells to better survive cytotoxic drugs that alter nucleotide pools 22 .
Using in vitro evolution for drug target and mechanism of action studies A major advantage of using yeast is that it is a model system for target discovery. To assess the functional importance of mutations we considered individual compounds and their mechanism of action. For compounds with defined targets we frequently identified mutations clustering in the active site of the proposed target molecule. For example, we isolated six strains resistant to flucytosine ( Table 1). Of the nine identified missense or stop mutations, six were in the uracil phosphoribosyltransferase domain of FUR1 (probability of enrichment by chance = 1.2 x 10 -26 using hypergeometric mean function). A homology model (Supplemental Figure 3A) reveals that they are all located near the 5-FUMP binding pocket, suggesting that these changes confer resistance by disrupting 5-FUMP binding. We obtained four tavaborole-resistant strains that were highly resistant (> 15 µM). These strains only six new high allele-fraction SNVs, four of which (R316T, V400F, V400D, and M493R), all of which were coding mutations in the 145 amino acid aminoacyl-tRNA synthetase editing domain of CDC60 (p = 1.11 x 10 -18 hypergeometric mean function ), the gene that encodes leucyl tRNA-synthetase in yeast. A LeuRS homology model (Supplemental Figure 3B) with a tavaborole ligand docked using QuickVina2 23 suggests that the CDC60 mutations confer resistance by directly interfering with tavaborole binding to Cdc60.
We also examine compounds that are used in chemotherapy. Camptothecin is a specific topoisomerase (Top1) inhibitor that binds the DNA/Top1 cleavage complex, preventing DNA religation 24 . We isolated two camptothecin-resistant yeast clones with three missense mutations, two of which were in TOP1 (G297C and E669*) ( Table 1). A homology model ( Figure 3C) 25 was constructed by aligning a partial yeast Top1 crystal structure to a crystal structure of the human TOP1 with camptothecin bound (PDB: 1T8I) 26 . This model showed that G297 is located in the core domain of the enzyme near the binding pocket, suggesting that it confers drug resistance by directly impeding drug binding, while E669* truncates the entire C-terminal domain, which contains the DNA-binding site 27 (Figure 3C), thus eliminating many protein/DNA contacts and likely impeding the formation of the drug-DNA-protein complex. Rapamycin, a macrocyclic lactone, and its analog everolimus (a socalled rapalog), potently inhibit mTOR, a protein kinase component of both the mTORC1 and mTORC2 complexes that controls cell growth and proliferation in many species. Two rapamycin resistant and three everolimus-resistant clones were identified. One carried a (S1975I) mutation in the FKBP12-rapamycin binding domain of mTOR (Table S4) and three carried a mutation in the FKBPtype peptidyl-prolyl cis-trans isomerase Pfam domain of FPR1, a small peptidylprolyl isomerase that interacts with mTOR. A model of the yeast Tor2/Fpr1/rapamycin tertiary complex shows that residue S1975 is near the bound rapamycin molecule (Figure 3D), suggesting that changes at this location might disrupt the formation of the tertiary complex. The model suggests that the two FPR1 truncation mutations (Y33* and Q61fs) (Table S4, Figure 3E) likely confer resistance by interfering with everolimus binding.
Our collection also contained compounds active against other pathogens. For example, mebendazole, a benzimidazole compound, is among the few effective drugs available for treating soiltransmitted helminths (worms) in both humans and animals. It binds to tubulin, thereby disrupting worm motility 28 . We confirmed the antifungal activity 29 of mebendazole and obtained two resistant strains in our in vitro selections. Of the nine missense mutations identified, two were in the GTPase domain of the TUB2 gene (R241S and L250F) ( Table S4), near or at the same residues (R241H and R241C) that confer resistance to the related antimitotic drug, benomyl, which also binds tubulin 30,31 .
Modeling studies ( Figure 3F) confirm that the binding mode is similar to that of benomyl, which binds with high affinity to the beta subunit of tubulin, thereby disrupting the structure and function of microtubules 32 . Despite sharing a common target with yeast, helminths and nematodes have benzimidazole-resistance mutations in codons 167, 198 and 200, suggesting some phylum specificity.
Alkylphosphocholines such as miltefosine and edelfosine were originally developed as anticancer agents, but recent work has shown that they are effective against trypanosomatid parasites such as Leishmania and Trypanosoma [33][34][35][36] . The specific target of these drugs remains uncertain.

Revealing the mechanism of action for uncharacterized compounds
To demonstrate that the yeast model can provide clues about mechanism of action we examined several poorly annotated compounds. Hectochlorin is a natural product from the marine cyanobacterium Lyngbya majuscule 40 that has strong antimalarial blood stage activity (IC50: 85.60 nM ± 0.96) as well as activity against GM yeast (IC50=0.25 µM). We identified six independent disruptive mutations, three of which were in the actin Pfam domain of Act1 (p = 1.9 x10 -13 , Table 1). We confirmed by CRISPR/Cas9 that a mutation in ACT1 confers resistance to hectochlorin in yeast (Table   S7). When mapped onto a crystal structure of Act1 (PDB: 1YAG 41 ) the altered amino acids line a distinct protein pocket ( Figure 3G) suggesting they confer resistance by directly disrupting compound binding. To assess whether hectochlorin resistance in Plasmodium occurs through a similar mechanism, we also mapped the mutations onto a synthetic construct of P. berghei Act1 protein (PDB: 4CBW (https://www.rcsb.org/structure/4CBW), which shares 97% sequence identity with PfAct1.
The altered amino acids again line a well-defined protein pocket, and the hectochlorin docked pose is also similar. This work supports published experiments that suggest actin is the target of hectochlorin 42 . To provide further support for this hypothesis, we determined if hectochlorin produces the same cell-invasion inhibition phenotype in malaria liver stage parasites as cytochalasin D, another actin polymerization inhibitor. Cytochalasin D has been shown to reduce Plasmodium sporozoite motility 43 , which is necessary for these exoerythrocytic forms to reach the host liver and begin replication. Treatment with 1 µM hectochlorin effectively blocked parasite invasion as efficiently as 10 µM cytochalasin D ( Figure 3H).

Transcriptional mechanisms are associated with multidrug resistance in yeast
Some genes appeared repeatedly across different compound sets. The set of 25 highest confidence genes (mutated five or more times across the dataset) was enriched for DNA-binding transcription factor activity (seven genes, Holm-Bonferonni-corrected p = 0.035). Altogether we observed 140 coding mutations in 24 genes affecting transcription ( Figure 4A). In addition to unique allelic exchanges in YRR1 (27x) and YRM1 (23x) (Figure 4B, C), multiple unique missense mutations Table S4) YRR1 and YRM1 are non-essential genes, and we hypothesized that our evolved resistant strains possessed YRR1 and YRM1 gain-of-function mutations that result in constitutive expression of transcriptional target genes encoding drug pumps. This hypothesis is motivated in part by the fact that others have reported resistance-conferring gain-of-function mutations in the related genes PDR1 and PDR3 48,49,50 . To expand on this previous work, we exposed a YRR1 L611F strain (generated using CRISPR/Cas9) to a set of compounds and observed cross-resistance to almost all compounds tested.
Of these, compounds that had not previously yielded YRR1 SNVs in our in vitro selections tended to have IC50 values only two-to three-fold higher in the YRR1 L611F strain. In contrast, compounds that had previously yielded YRR1 mutations in our selections tended to elicit a more than 3-fold IC50 difference (Table S9).
Others have found that deleting the YRR1 gene entirely does not significantly increase sensitivity to cytotoxic compounds 48 , providing further evidence that the YRR1 and YRM1 mutations identified in our selections-which were all single amino-acid changes-represent gain-of-function mutations. We confirmed this finding by testing a small set of cytotoxic compounds against a yrr1 deletion strain. With only one exception, we also noted no significant differences in growth inhibition over the wild-type-allele strain. Only compound MMV668507 had a dramatically lower IC50 against the yrr1 deletion strain (Table S9).
Based on Gal4 model and previous work 44 it is likely that the mutated region contains a binding site for one or more repressor proteins and that YRM1 and YRR1 SNVs are gain-of-function mutations that result in constitutive expression of drug pump genes. To test this hypothesis, we used qPCR to directly evaluate the expression of three such target genes (AZR1, FLR1, and SNG1). To this set we added the YRR1 gene itself, since YRR1 activates its own expression via an auto-feedback loop ( Figure   4E). We examined gene expression in YRR1-mutant strains, in the GM strain with a wild-type YRR1 allele, and in the yrr1 deletion strain (Table S10). Relative RNA expression levels ( Figure 4E) of the putative target genes AZR1, FLR1, SNG1 and YRR1 are 2-70-fold higher in all YRR1 evolved mutant strains tested compared to the parental GM strain or the yrr1 deletion strain. To assess whether specific YRR1 mutations confer resistance to specific compounds or show a more general resistance response, we evaluated nine different YRR1 mutant strains from our resistance selections (deletion (1), evolved (7), and CRISPR-Cas9 edited (1)) for resistance to a set of four structurally unrelated compounds (CBR113, AN7973, MMV665852, and DDD01027481). All tested strains showed strong crossresistance to all four compounds (Table S9). Taken together, these results strongly support our hypothesis that the identified YRR1 SNVs lead to constitutive transcriptional activation of their target genes involved in the pleiotropic drug response, thereby conferring general resistance to many compounds. The mutated regions may represent binding sites for repressor proteins that are removed when compounds are present.

Enrichment of transcription factors is not observed in other species.
The repeated identification of transcription factors as mediators of drug resistance in S.
cerevisiae motivated us to investigate whether similar patterns would be observed in other species.
To our knowledge, similar large-scale systematic studies of evolution in the presence of small molecules have only been performed on one other species, the malaria parasite, Plasmodium falciparum 51 . In this study, which consisted of evolutions with 37 compounds, mutations were detected in 361 genes. This study showed that many mutations were found directly in a compound's target and there was no significant enrichment for particular gene class after excluding genes involved in antigenic variation. It is remarkable that in Plasmodium only a single nonsynonymous variant in a transcription factor was identified in the entire set of 1905 mutations (although 9 frameshift or inframe indels were observed). Clearly, organisms from the two phyla use different resistance strategies. This may not be unexpected given that Plasmodium parasites spend all of their lifecycle within other organisms and are seldom exposed to environmental stresses. These data may suggest why it is relatively easy to kill intracellular parasites with small molecules and why transcriptional profiling may not be useful for drug target identification and mechanism of action studies in some species.

Discussion
To our knowledge this is one of the most comprehensive studies of the drug selected mutational landscape in fungi. Although genome-wide sets of knockout strains have been used to discover drug targets and to study drug resistance 52,4,53-55 , our approach is different in that we identify single nucleotide gain-of-function mutations in specific domains. This difference allows IVIEWGA to complement other genome-wide knockdown approaches, including those that rely on measures of haploinsufficiency in the presence of compound (HIPHOP) 52 . While S. cerevisiae is outstanding model organism, few genetic tools were needed for this study, and the approach can be applied to any organism that can be subjected to drug pressure and sequenced.
One potential disadvantage of whole genome evolutionary approaches is that background passenger mutations can accumulate during the prolonged culturing of a fast-dividing organism and some of these may not contribute to resistance 21 . But given that the overall ratio of nonsynonymous to synonymous changes was 8:1 in our study, most mutations likely do offer some advantage to the cell, even when they are not the primary driver of resistance. For these reasons, a large dataset, such as ours, can provide clarity and statistical confidence, even in the absence of CRISPR-Cas9 reconfirmation of an allele's importance. The examples that that are discussed in the manuscript did, in almost all cases, achieve strong statistical significance, with mutations appearing in the same genes, or appearing with compounds that are closely related to one another at rates not expected by chance.
The reproducibility of the results for genes and compound families indicate that the approach will be powerful in other fungal species that lack the genetic tools that are available for S. cerevisiae. It should be mentioned that many, statistically significant examples were not discussed here for the sake of brevity. For example, we find a strong association between BUL1 and BAP2 mutations and inhibitors of mitochondrial function and vacuolar ATPases mutations were associated with other scaffold families.
Little was missed in our analysis as well. A resistance allele in one of the 137 genes detected 2 or more times (55 identified more than 2 times) was discovered for 79/80 compounds and in more than 90% of clones. Some of the 30 clones, which did not have a clear, resistance-associated coding SNV had CNVs (e.g. MMV665794-R6a and ARR1). In a few cases, they bore mutations in a gene that likely interacts with the gene driving resistance in other clones derived from the same compound treatment: An SNV in YPK2 was found in one orphan rapamycin-resistant clone (Rap-4R3a) and this is involved in involved in the TORC-dependent phosphorylation of ribosomal proteins. The ortholog of yeast YAP1, hsYAP is involved in human cancer drug resistance 63 . Among other genes associated with human chemotherapy resistance include target mutations in TOR2, the target of rapamycin. We show conservation between mutations in Leishmania and yeast, Plasmodium and yeast as well.
It is noteworthy that none of the clinically approved antifungals (posaconazole, tavaborole, miconazole) are among those that elicited YRM1-or YRR1-mediated resistance mechanisms, which may indicate why these compounds are ultimately clinically effective against most fungal pathogens.
Studies on both the resistance profile and the rate at which resistance emerges are now incorporated into the drug development pipeline for eukaryotic pathogens such as Plasmodium falciparum and it may be that studies, similar to those described here, need to be performed if the objective is to create better drugs for fungal pathogens.

Yeast Strains
All yeast strains used are listed in Table S11.

S. cerevisiae susceptibility and dose-response assays
To measure compound activity against whole-cell yeast, single colonies were inoculated into

CNV Analysis
Coverage values across defined gene intervals in each alignment file were calculated using GATK DiagnoseTargets (input parameters: -max 2000 -ins 1500 - MQ 50). Coverage values were logtransformed then mean-centered across and within arrays in Cluster. Copy number variant were filtered so that they would only be retained if there was at least 2-3x fold coverage change relative to the parent strain and if they spanned four or more genes (Table S5). CNVs were visually confirmed in IGV.

Intergenic mutation analysis
A Python script was written to map the 271 intergenic mutations to known chromosomal features coordinates based on the S. cerevisiae S288C genome version R64-2-1 created by the SGD database. Each mutation was located within the chromosomal coordinate of the feature, intron, exon, and other subfeatures, as well as its proximity to those coordinates with a maximum of 500 base pair distant to both up-and downstream directions (Table S6).

Plasmodium invasion assay
The impact of hectochlorin on hepatocellular traversal and invasion by Plasmodium berghei (Pb) sporozoites was measured using a previously established flow cytometry-based assay 43 . was also utilized to mimic the treated well conditions. Sporozoites were freshly dissected and prepared 2-4 h before infection using a previously described method 68 . Immediately prior to infection, rhodamine-dextran was added to each test well (final concentration 1 mg/mL) followed by 3.5 x 10 4 Pb-GFP sporozoites. The plates were then incubated at 37°C and 5% CO2 for 2 h. Following this incubation, the cells were washed and the presence of GFP and rhodamine-dextran signals were evaluated using flow cytometry.