Genome Profiling for Aflatoxin B1 Resistance in Saccharomyces cerevisiae Reveals a Role for the CSM2/SHU Complex in Tolerance of Aflatoxin B1-Associated DNA Damage

Exposure to the mycotoxin aflatoxin B1 (AFB1) strongly correlates with hepatocellular carcinoma (HCC). P450 enzymes convert AFB1 into a highly reactive epoxide that forms unstable 8,9-dihydro-8-(N7-guanyl)-9-hydroxyaflatoxin B1 (AFB1-N7-Gua) DNA adducts, which convert to stable mutagenic AFB1 formamidopyrimidine (FAPY) DNA adducts. In CYP1A2-expressing budding yeast, AFB1 is a weak mutagen but a potent recombinagen. However, few genes have been identified that confer AFB1 resistance. Here, we profiled the yeast genome for AFB1 resistance. We introduced the human CYP1A2 into ∼90% of the diploid deletion library, and pooled samples from CYP1A2-expressing libraries and the original library were exposed to 50 μM AFB1 for 20 hs. By using next generation sequencing (NGS) to count molecular barcodes, we initially identified 86 genes from the CYP1A2-expressing libraries, of which 79 were confirmed to confer AFB1 resistance. While functionally diverse genes, including those that function in proteolysis, actin reorganization, and tRNA modification, were identified, those that function in postreplication DNA repair and encode proteins that bind to DNA damage were over-represented, compared to the yeast genome, at large. DNA metabolism genes also included those functioning in checkpoint recovery and replication fork maintenance, emphasizing the potency of the mycotoxin to trigger replication stress. Among genes involved in postreplication repair, we observed that CSM2, a member of the CSM2(SHU) complex, functioned in AFB1-associated sister chromatid recombination while suppressing AFB1-associated mutations. These studies thus broaden the number of AFB1 resistance genes and have elucidated a mechanism of error-free bypass of AFB1-associated DNA adducts.

into stable AFB 1 -formamidopyrimidine (FAPY) adducts (Essigmann et al. 1977;Lin et al. 1977;Croy and Wogan 1981). The anomers of the AFB 1 -FAPY-DNA adduct block DNA replication or cause mutations in Escherichia coli (Smela et al. 2002;Brown et al. 2006) and in vitro (Lin et al. 2014). Metabolically active AFB 1 can also indirectly damage DNA through oxidative stress (Shen et al. 1995;Beddard and Masey 2006;Bernabucci et al. 2011;Singh et al. 2015). Identifying genes that repair AFB 1 -associated DNA damage could help identify which individuals are at elevated risk for HCC. However, epidemiological data has been inconsistent, and only a few candidate DNA repair genes have been proposed, such as XRCC1 (Pan et al. 2011, Xu et al. 2015, XRCC3 (Long et al. 2008;De Mattia et al. 2017) and XRCC4 (Long et al. 2013).
AFB 1 resistance genes have been identified from model organisms, revealing mechanisms by which AFB 1 -associated DNA adducts can be both tolerated and excised. Both prokaryotic and eukaryotic nucleotide excision repair (NER) pathways function to remove AFB 1 -associated DNA adducts (Leadon et al. 1981;Alekseyev et al. 2004;Bedard and Massey 2006). Recently, the base excision repair gene (BER) NEIL1 has been implicated in direct repair AFB 1 -associated DNA adducts (Vartanian et al. 2017). To tolerate persistent AFB 1 -associated DNA lesions, translesion polymerases in yeast, such as those encoded by REV1 and REV7, and in mouse, such as Polz, confer resistance and may promote genome stability (Lin et al. 2014).
While yeast does not contain endogenous P450 genes that can metabolically activate AFB 1 into an active genotoxin (Sengstag et al. 1996;Van Leeuwen et al. 2012, Fasullo et al. 2014, human CYP1A2 can be expressed in yeast cells (Sengstag et al. 1996;Fasullo et al. 2014). Interestingly, metabolically activated AFB 1 is a potent recombinagen but weak mutagen (Sengstag et al. 1996). CYP1A2-activated AFB 1 reacts to form both the unstable AFB 1 -N 7 -Gua adducts and the stable AFB 1 -FAPY DNA adducts (Fasullo et al. 2008). The AFB 1 -associated DNA damage, in turn, triggers a robust DNA damage response that includes checkpoint activation (Fasullo et al. 2008), cell cycle delay (Fasullo et al. 2010), and the transcriptional induction of stress-induced genes (Keller-Seitz et al. 2004;Guo et al. 2006). Profiles of the transcriptional response to AFB 1 exposure reveals induction of genes in growth and checkpoint signaling pathways, DNA and RNA metabolism, and protein trafficking (Keller-Seitz et al. 2004;Guo et al. 2006). While genes involved in recombinational repair and postreplication repair confer AFB 1 resistance (Keller-Seitz et al. 2004;Guo et al. 2005;Fasullo et al. 2010), it is unclear the functional significance of many genes in the stress induced pathways in conferring resistance since transcriptional induction is not synonymous with conferring resistance (Birrell et al. 2002).
In this study, we profiled the yeast genome for AFB 1 resistance. We asked which genes confer AFB 1 resistance in the presence or absence of human CYP1A2 expression by screening the non-essential diploid collection by high throughput sequencing of molecular barcodes (Pierce et al. 2006;St Onge et al. 2007;Smith et al. 2010). While we expected to identify NER, recombinational repair, and postreplication repair, which had previously been identified (Keller-Seitz et al. 2004;Guo et al. 2005;Fasullo et al. 2010), our high throughput screen identified novel genes involved in AFB 1 resistance. These included genes involved in Rad51 assembly, cell cycle progression, RNA metabolism, and oxidative stress. Our results thus underscore the importance of recombination in both mutation avoidance and in conferring AFB 1 resistance.
Using LiAc-mediated gene transformation we introduced human CYP1A2 into BY4741, csm2, rad4, rad51, csm2 rad51, and csm2 rad4 strains. The CYP1A2-expression plasmid, pCS316, was obtained by CsCl centrifugation (Ausubel et al. 1995) and the restriction map was verified based on the nucleotide sequence of the entire plasmid. An alternative CYP1A2-expression plasmid, pCYP1A2_NAT2 was constructed by removing the hOR sequence from pCS316 and replacing it with a Not1 fragment containing the human NAT2.

Media and chemicals
Standard media were used for the culture of yeast and bacterial strains (Burke et al. 2000). LB-AMP (Luria broth containing 100 mg/ml ampicillin) was used for the culture of the bacterial strain DH1 strain containing the vector pCS316. Media used for the culture of yeast cells included YPD (yeast extract, peptone, dextrose), SC (synthetic complete, dextrose), SC-HIS (SC lacking histidine), SC-URA (SC lacking uracil), and SC-ARG (SC-lacking arginine). Media to select for canavanine resistance contained SC-ARG (synthetic complete lacking arginine) and 60 mg/mL canavanine (CAN) sulfate, and media to select for 5-fluoroorotic acid (FOA) resistance contained SC-URA supplemented with 4x uracil and FOA (750 mg/ml), as described by Burke et al. (2000). FOA plates contained 2.2% agar; all other plates contained 2% agar. AFB 1 was purchased from Sigma Co., and a 10 mM solution was made in dimethyl sulfoxide (DMSO).

Measuring DNA Damage-Associated recombination and mutation events
To measure AFB 1 -associated genotoxic events, log phase yeast cells (A 600 = 0.5-1) were exposed to indicated doses of AFB 1 , previously dissolved in DMSO. Cells were maintained in synthetic medium (SC-URA) during the carcinogen exposure. After the exposure, cells were washed twice in H 2 O, and then plated on SC-HIS or SC-ARG CAN to measure unequal SCE or mutation frequency, respectively. An appropriate dilution was inoculated on YPD to measure viability (Fasullo et al. 2008).

Construction of CYP1A2-expression library
To introduce CYP1A2 (pCS316, Sengstag et al. 1996, and pCYP1A2_NAT2) into the yeast diploid deletion collection, we used a modified protocol for high throughput yeast transformation in 96-well plates (Gietz and Schiestl 2007). In brief, FOA R isolates were isolated from each individual strain in the diploid collection and inoculated in 96-well plates, each containing 100 ml of YPD medium. After incubation over-night at 30°, plates were centrifuged, washed in sterile H 2 O, and resuspended in one-step buffer (0.2 N LiAc, 100 mM DTT, 50% PEG, MW 3300, 500 mg/ml denatured salmon sperm DNA). After addition of 1 mg pCS316 and incubation for 30 min at 30°, 10 ml were directly inoculated on duplicate SC-URA plates. Two Ura + transformants were chosen corresponding to each well and frozen in SC-URA 0.75% DMSO. We introduced the CYP1A2-containing plasmids into approximately 90% of the deletion collection.
Functional profiling of the yeast genome The CYP1A2-expressing libraries were pooled and frozen in SC-URA medium containing 0.75% DMSO (n = 4150). The pooled cells (100 ml) were added to 2 ml of SC-URA and allowed to recover for two hours. Cell were then diluted to A 600 = 0.85 in 2 ml of SC-URA and exposed to either 50 mM AFB 1 in 0.5% DMSO, and 0.5% DMSO alone. Cells were then incubated with agitation at 30°for 20 hs. Similarly, the pooled BY4743 library (n = 4607) was directly diluted to A 600 = 0.85 in YPD and also exposed to 50 mM AFB 1 and DMSO for 20 hs. Independent triplicate experiments were performed for each library and each chemical treatment.
After AFB 1 exposure, cells were washed twice in sterile H 2 O and frozen at -80°. Cells were resuspended in 10 mM Tris-HCl, 1 mM EDTA, 100 mM NaCl, 2% Triton X-100, 1% SDS, pH 8 and DNA was isolated by "smash and grab (Hoffman and Winston 1987)." Barcode sequences, which are unique for each strain in the deletion collection Giaever et al. 2004), were amplified by PCR using a protocol described by Smith et al. (2010). The primers used for amplification are listed in the Table S2. 125 bp PCR products were then isolated from 10% polyacrylamide gels by diffusion in 0.5M NH 4 Ac 1 mM EDTA for 24 hs (30°) followed by ethanol precipitation. The DNA was quantified after being resuspended in Tris EDTA pH 7.5 and the integrity of the DNA was verified by electrophoresis on 10% polyacrylamide. Equal amounts of DNA were pooled from treated and untreated samples. The uptags were then sequenced using the Illumina Platform at the University Buffalo Genomics and Bioinformatics Core (Buffalo, New York). Sequence information was then uploaded to an accessible computer server for further analysis. The software to demultiplex the sequence information, match the uptag sequences with the publish ORFs, and calculate the statistical significance of the differences in log2N ratios was provided by F. Doyle. Tag counts were analyzed with the TCC Bioconductor package (Sun et al. 2013) using TMM normalization (Robinson and Oshlack 2010) and the edgeR test method . Statistical testing was performed with edgeR TCC package for tag count comparison in the R programming language; an R script with invocation details is provided in the File S1. Data files have been deposited in the Gene Expression Omnibus database, GSE129699.
Over-enrichment analysis Gene Ontology (GO) categories were identified by a hypergeometric distribution with freely available software from Princeton University using the Generic Gene Ontology Term Finder (http://go.princeton.edu/cgibin/GOTermFinder) and Bonferroni correction for P values. Enriched GO terms were further refined using ReViGO (Supek et al. 2011). Enrichment analysis was analyzed using Panther (http:// pantherdb.org/tools/) with a P value cutoff of , 0.05 (Cherry et al. 2012, Mi et al. 2016. The AFB 1 sensitivity of mutants corresponding to GO groups was verified by growth curves and trypan blue assays. Growth assays in 96 well plate to measure AFB 1 sensitivity In brief, individual saturated cultures were prepared for each yeast strain. Cell density was adjusted to $0.8 · 10 7 cells/ml for all cultures. We maintained the cells in selective medium (SC-URA). In each microtiter well, 95 ml of media and 5 ml of cells (8 · 10 4 cells) were aliquoted in duplicate for blank, control and experimental samples. For experimental samples, we added AFB 1 , dissolved in DMSO, for a final concentration of 50 mM and 100 mM. The microtiter dish was placed in a plate reader that is capable of both agitating and incubating the plate at 30°, as previously described (Fasullo et al. 2010;Fasullo et al. 2014). We measured the A 600 at 10 min intervals, for a total period for 24 hs, 145 readings. Data at 1h intervals was then plotted. To avoid evaporation during the incubation, the microtiter dishes were sealed with clear optical tape (Fasullo et al. 2010). To calculate area under the curve (AUC), we used a free graphing application (https://www.padowan.dk/download/), and measured the time interval between 0-20 hs, as performed in previous publications (O'Connor et al. 2012). After cells were exposed to AFB 1 and DMSO, we calculated the ratio (AUC AFB1 /AUC DMSO ) · 100% to determine the percent yeast growth obtained in the presence of the toxin. For the wild-type diploid BY4743 expressing CYP1A2 (YB556), the percent of yeast growth after exposure to 50 mM is 89.7 6 2.5. Statistical significance of differences between growth percentages for diploid strains and BY4743 were determined by the Student's t-test, assuming constant variance between samples.
To determine epistasis of AFB 1 -resistance genes, we calculated the deviation e according to e = W xy -W x x W y , where W x and W y are the fitness coefficients determined for each single mutant exposed to AFB 1 and W xy is the product. Fitness was calculated by determining the generation time of both the single and double mutants over three doubling times. Zero and negative values are indicative of genes that do not interact or participate in the same pathway to confer fitness Trypan blue exclusion assay to monitor cell viability after acute AFB 1 exposure To measure cell viability after AFB 1 exposure, we performed a trypan blue exclusion assay. Selected strains expressing CYP1A2 were inoculated in SC-URA until cultures reached an A 600 $0.1-0.5, and then exposed to either 50 mM in AFB 1 or DMSO (solvent) alone. After incubating for 3 hs, cells were washed twice in sterile phosphate buffered saline (PBS) and stained with trypan blue at a final concentration $10 mg/ml (Liesche et al. 2015). Cells were counted in a Nexcelom cellometer T4, according to the manufacturer's instructions. A minimum of 10 4 cells were counted and all strains were tested at least twice. Statistical significance was determined by the Student's t-test.

Western blot analysis
Expression of CYP1A2 was determined by Western blots and MROD assays. Cells were inoculated in SC-URA medium. Cells in log growth phase (A 600 = 0.5-1) were concentrated and protein extracts were prepared as previously described by Foiani et al. (1994). Proteins were separated on 10% acrylamide/0.266% bis-acrylamide gels and transferred to nitrocellulose membranes. Human CYP1A2 was detected by Western blots using goat anti-CYP1A2 (Abcam), and a secondary bovine anti-goat antibody. For a loading control on Western blots, b-actin was detected using a mouse anti-b-actin antibody (Abcam 8224) and a secondary goat anti-mouse antibody. Signal was detected by chemiluminescence, (Fasullo et al. 2014).

Measuring CYP1A2 enzymatic activity
We measured CYP1A2 enzymatic activity using a modified protocol described by Pompon et al. (1996). In brief, cells obtained from 100 ml of selective media were pelleted and resuspended in 5 ml Tris EDTA KCl (pH 7.5, TEK) buffer. After five minute incubation at room temperature, cells were pelleted, resuspended in 1 ml 0.6 M Sorbitol Tris pH 7.5, and glass beads were added. Cells were lysed by agitation. The debris was pelleted at 10,000 · g at 4°, and the supernatant was diluted in 0.6 M Sorbitol Tris pH 7.5 and made 0.15 M in NaCl and 1% in polyethylene glycol (MW 3350) in a total volume of 5 ml. After incubation on ice for 1 hr. and centrifugation at 10,000 rpm for 20 min, the precipitate was resuspended in Tris 10% glycerol pH 7.5, and stored at 2 80°.
CYP1A2 enzymatic activity was measured in cell lysates by quantifying 7-methoxyresorufin O-demethylase (MROD) activities (Fasullo et al. 2014), using a protocol similar to that quantifying ethoxyresorufin O-deethylase (EROD) activity (Eugster et al. 1992;Sengstag et al. 1994). The buffer contained 10 mM Tris pH 7.4, 5mM methoxyresorufin (Sigma) and 500 mM NADPH. The production of resorufin was measured in real-time by fluorescence in a Tecan plate reader, calibrated at 535 nm for excitation and 580 nm for absorption, and standardized using serial dilutions of resorufin. The reaction was started by the addition of NADPH and resorufin was measured at one minute intervals during the one hour incubation at 37°; rat liver microsomes (S9) were used as a positive control while the reaction without NADPH served as the negative control. Enzyme activities were measured in duplicate for at least two independent lysates from each strain and expressed in pmol/min/mg protein.

Data availability
All yeast strains and plasmids are available upon request and are detailed in Table S1 Figure S1 describes the percentage of growth obtained after 86 strains were exposed to 50 mM AFB 1 and a subset of the more resistant strains that were exposed to both 50 mM and 100 mM AFB1. Figure S2 is a bar graph detailing the percent viability of strains after an acute exposure 50 mM AFB 1 . Figure S3 contains the growth curves for BY4743, shu1, shu2, and psy3 after exposure to 50 mM AFB 1 . File S1 describes edge R script with invocation details. Supplemental material available at figshare: https://doi.org/10.25387/g3.12895313.

RESULTS
We used three BY4743-derived libraries to profile the yeast genome for AFB 1 resistance. The first was a pooled library of 4607 yeast strains, each strain containing a single deletion in a non-essential gene (Jo et al. 2009). The second was a pooled library of approximately 4900 strains each containing individual deletions in non-essential genes and was made by introducing pCS316 into each strain by yeast transformation. The third was a pooled library of approximately 5000 strains expressing both CYP1A2 and NAT2; this pooled library can be used to screen polyaromatic and heterocyclic compounds that require CYP1A2 and NAT2 for metabolic activation. By calculating area under the growth curves (AUC) for cells exposed to AFB 1 and solvent (DMSO) alone and measuring the ratio (AUC AFB1 /AUC DMSO ), we determined that the AFB 1 concentration to elicit $10% decrease in growth (D 10, ) for BY4743 expressing CYP1A2 (YB556) was 50 mM, while the dose to elicit a 16% decrease in growth (D 16 ) was 100 mM. The number of AFB 1 -associated DNA adducts formed in vivo after exposure to 100 mM AFB 1 is less than twice of that detected after exposure to 50 mM AFB 1 (Fasullo et al. 2008), suggesting that metabolic activation is more efficient after exposure to 50 mM AFB 1 ; we therefore chose 50 mM AFB 1 exposure to identify genes that confer resistance to AFB 1 -associated metabolites ( Figure 1A). The D 10 for BY4743 expressing CYP1A2 and the human oxidoreductase (hOR) was the same as in BY4743 cells expressing CYP1A2 and NAT2. To confirm metabolic activation of AFB 1 into a potent genotoxin, we showed that growth of the rad52 diploid mutant was significantly impaired ( Figure 1D). Cells that did not express CYP1A2 showed slight growth delay after cells were exposed to 100 mM AFB 1 ( Figure 1C).

Confirmation of CYP1A2 activity
To confirm that CYP1A2 was both present and functional in the library, we performed Western blots ( Figure 1B) and MROD assays, as in previous studies (Fasullo et al. 2014). Two independent assays were performed for four different ORFs (RAD2, RAD18, RAD55, and OGG1); the range of average MROD activity was 5-10 units pmol/sec/ mg protein (see Table S3). These results are similar to what was observed for the wild type BY4743 expressing CYP1A2 (Fasullo et al. 2014) and for various haploid mutants (Guo et al. 2005). These studies indicate that CYP1A2 is active in diploid strains and can be detected by Western blots in BY4743-derived strains containing pCS316, in agreement with previous studies (Guo et al. 2005). a Genes in "bold" are those that are responsive to replication stress. Ã Appears twice among screens (q , 0.1). ÃÃ Appears twice among screens (q , 0.1 and P , 0.05). # Transcription induced by AFB 1 exposure.

b.
m.value is the numeric vector of fold change on a log2 scale, rounded to three significant digits. c Gene function descriptions obtained from www.yeastgenome.org. d q value is the numeric vector calculated based on the p-value using the p.adjust function with default parameter settings, rounded to three significant digits.

Identification of genes by barcode analysis
We classified genes that confer AFB 1 resistance (q , 0.1) genes into: 1) those that confer resistance to AFB 1 without CYP activation, and 2) those that confer resistance to P450-activated AFB 1 . After exposing cells to 50 mM AFB 1 , we identified barcodes for approximately 51% and 89% of the genes from pooled deletion library with and without CYP1A2, respectively (for complete listing, see https:// www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE129699).
One gene, CTR1, which functions in high-affinity copper and iron transport (Dancis et al. 1994), was identified as statistically different from the pooled library that did not express CYP1A2; this gene was identified twice in the screen (YPR124W and completely overlapping YPR123C). The human homolog confers drug resistance, suggesting human CTR1 is involved in xenobiotic transport (Furukawa et al. 2008). No DNA repair genes were identified in screening the library lacking CYP1A2, consistent with observations that cytochrome P450metabolic activation AFB 1 is required to form AFB 1 -associated DNA adducts (Fasullo et al. 2014).
Using the same stringent assessment (q , 0.1), in three independent screens, we identified 96 ORFs that confer AFB 1 resistance in cells expressing CYP1A2, of which 86 genes have been ascribed a function, and one ORF, YBR099C, is completely internal to MMS4.
Genes that confer resistance were associated with a negative m value (Table 1), as defined by fold change on a log2 scale; this indicates that corresponding deletion strains would be relatively depleted after AFB 1 exposure, compared to exposure to solvent (DMSO) alone. Alternatively, genes associated with a positive m value (Table S4) imply that corresponding mutants would be relatively enriched after AFB 1 exposure. 43 ORFs were only associated with a positive m value, and nine ORFs were associated with both a negative and positive m values in independent screens (Table S4).
To confirm the AFB 1 -sensitivity for strains associated with negative m values, we determined percent growth inhibition for individual strains and compared calculations with that obtained for BY4743 expressing CYP1A2 (YB556). The percent growth inhibition for 86 strains after exposure to 50 mM AFB 1 was determined are listed in Figure S1; significance was determined by comparison with YB556. Among strains that indicated both positive and negative m values, growth curves indicated that these deletion strains were actually more AFB 1 -sensitive compared to the wild-type BY4743, consistent with the negative m values. We confirmed that of 79 out of 86 strains (92%) are AFB 1 -sensitive ( Figure S1); growth curves are shown for a subset of these strains ( Figure 2) and trypan blue staining indicated that viability is lost Figure 2 Growth curves for selected diploid mutants identified in the high throughput screen. All strains contain pCS316, expressing CYP1A2. A 600 is plotted against time (Hs). Standard deviations are indicated at 1 h time points. The growth curves (Panel A) are indicated for wild type (YB556) and the csm2, alk1, ssm4, cue1, and trm9 diploid mutants. Red lines indicate strains both expressing CYP1A2 and exposed to 50 mM AFB 1 . The bar graph (Panel B) indicates area percent growth of AFB 1 -exposed strains as determined by the ratio of the area under curves (area under the curve for treated strain/area under the curve for strain exposed to DMSO x 100%).
n■ among representative strains after acute AFB 1 exposure ( Figure  S2). Strains which have not been confirmed by any criteria are listed in Table S5. AUCs were calculated for a few strains (gtt1, ies2, sip18, tpo4) that were only associated with positive m values; none of these were AFB 1 -sensitive.
Of the 79 AFB 1 -sensitive strains, 15 (19%) are deleted in welldocumented DNA repair genes (Table 1). Another 64 are deleted for genes that have diverse functions (Table 2) but also include an additional five genes (BLM10, FUM1, PSY2, DPB3, NUP60) noted to confer resistance to diverse DNA damaging agents and placed in n■ a Genes in "bold" are those that are responsive to replication stress. Ã Appears twice among screens (q , 0.1). ÃÃ Appears twice among screens (q , 0.1 and P , 0.05).

b.
m.value is the numeric vector of fold change on a log2 scale, rounded to three significant digits. c Gene function descriptions obtained from www.yeastgenome.org. d q value is the numeric vector calculated based on the p-value using the p.adjust function with default parameter settings, rounded to three significant digits. the DNA repair gene ontology group (Table 3). Five different genes, participating in nucleotide excision repair, postreplication repair and ribosome assembly were twice found among the 79 strains. An additional five genes were found that that were highly statistically different (q , 0.1) in one screen and statistically different in another screen (P , 0.05). Among these were those involved in proteolysis (CUE1), vacuolar acidification (VOA1), cell cycle progression (FKH2), DNA recombinational repair (RAD55) and postreplication repair (RAD18).
According to GO process enrichment (https://go.princeton.edu/ cgi-bin/GOTermFinder), resistance genes included those that function in the DNA damage response, DNA repair, postreplication repair, DNA damage stimulus, and meiotic cell cycle progression; 13 GO groups are shown in Table 3 (for full list of complete  86 putative ORFs, see Table S6). Of the 79 AFB 1 resistance genes, 42 genes belong to the top 13 GO groups. One GO group that was unexpected was meiotic cell cycle process, which includes meioticspecific genes SPO1 and DIT1. Among the stress responsive genes are those that function in cell wall maintenance and glycogen metabolism, which were previously identified to confer resistance to a variety of toxins, such as benzopyrene and mycophenolic acid (O'Connor et al. 2012). Other genes involved in carbohydrate metabolism, such as GRE3 that encodes aldose reductase, could have a direct role in detoxification and is induced by cell stress (Barski et al. 2008). Genes involved in rearrangement of the cellular architecture include BIT2, AKL1 and PPG1; these genes function to rearrange the cellular architecture when cells are stressed (Schmidt et al. 1996). Thus, among AFB 1 resistance genes are those that function to maintain structural integrity by affecting the cytoskeletal and cell wall architecture.
Other gene ontology groups encompass functions involved in mitochondrial maintenance and response to oxidative stress, and RNA metabolism (Table S6). Genes involved in mitochondrial function and response to oxidative stress include TRX3, MRPL35, MIX23, MIS1; TRX3 (thioredoxin reductase) functions to reduce oxidative stress in the mitochondria (Greetham and Grant 2009). RNA metabolism genes include those involved in chemical modification of tRNA, including MIS1, TRM9, DUS1, and RIT1, and those involved in n■ Table 3 Aflatoxin resistant genes categorized by gene ontology groups involved in biological process and function RNA translation, such as TMA20 and YIH1. TRM9 confers resistance to alkylated DNA damage, and links translation with the DNA damage response (Begley et al. 2007). These genes are consistent with the notion that AFB 1 causes oxidative damage and that mitochondria are targets of AFB 1 -induced DNA damage.
In grouping genes according to protein function and cellular components (Table 3 and Table 4), DNA repair complexes were readily identified. Among these were the Shu complex, and the NER complexes I and II (Table 4). However, other interesting complexes that were identified included the glycosidase II complex, and the CK2 complex. Because DNA repair and DNA damage response genes were the most prominent of the GO groups, we focused on the function of these genes in conferring AFB 1 resistance. As expected, these genes included those that participate in DNA recombination (RAD54, RAD55), nucleotide excision repair (RAD1, RAD4, RAD1,RAD10, RAD23), and postreplication repair (RAD5, RAD18, REV1, REV3). Many of these genes function in cell cycle progression. For example, PSY2 and CKB2 (Toczyski et al. 1997) promote cell cycle progression after cell cycle delay or arrest caused by stalled forks or double-strand breaks, respectively.
To determine which GO biological processes and protein functions were most enriched among the AFB 1 resistance genes we used the Panther software (Mi et al. 2016). A larger proportion of the AFB 1 resistance genes are involved in DNA repair and metabolism, compared with the genome at large (Figure 3). A fraction of these genes also participate in postreplication repair, including DNA tolerance pathways that are error-free and error-prone replication (Table 5). In classifying protein functions, we analyzed whether hydrolases, nucleases, phosphatases, DNA damage binding were more enriched among the 79 AFB 1 resistance genes, compared to the genome, at large (Figure 3). Of these groups, DNA damage binding was enriched among resistance genes (P , 0.05).
To further determine the strength of the interactions among the confirmed 79 AFB 1 resistance genes, we performed interactome mapping, using STRING software (https://string-db.org/, Szklarczyk et al. 2019), which associates proteins according to binding, catalysis, literature-based, and unspecified interactions (Figure 4). The interactome complex in yeast included 78 nodes and 152 edges with a 3.77 average node degree. Besides the NER complexes, individual complexes included the Shu complex, the Glucosidase II complex, and the protein kinase CK2 complex (Table 4); the glucosidase II complex is conserved in mammalian cells (Figure 4). While the strength and number of these interactions was particularly strong among the DNA repair genes, other interactions were elucidated, such as the interactions of protease proteins with cell cycle transcription factors and cyclins (Figure 4). Since many known AFB 1 resistance genes, which function in DNA repair and postreplication repair pathways, were not present among highly statistically different genes (q , 0.1), we also used a less stringent (P , 0.05) qualifier to identify potential AFB 1 resistance genes. Among genes identified were additional members of the SHU complex, including SHU1 and SHU2. These genes were confirmed by additional growth curves ( Figure S3) and the percent growth was determined ( Figure S1).
The SHU complex was previously identified as participating in error-free DNA damage tolerance and mutation avoidance (Shor et al. 2005, Xu et al. 2013. The complex confers resistance to alkylating agents, such as methyl methanesulfonate (MMS), and cross-linking agents, such as cisplatin, but not to UV and X-ray (Godin et al. 2016). We previously showed that while X-ray associated unequal SCE (SCE) was RAD5-independent (Fasullo and Sun 2017), MMS and 4NQO-associated unequal SCE occurs by wellconserved RAD5-dependent mechanisms (Unk et al. 2010;Fasullo and Sun 2017). We therefore postulated that the SHU complex suppresses AFB 1 -associated mutagenesis while promoting AFB 1associated template switching. We introduced pCS316 (CYP1A2) into both the haploid wild-type strain (YB204) and a csm2 mutant (YB558, see Table S1) to measure frequencies of AFB 1 -associated unequal SCE and can1 mutations. Our results showed that while we observed a threefold increase in SCE after exposure to AFB 1 in wild type strains, we observed less than a twofold increase in sister chromatid exchange in the csm2 mutant ( Figure 5). However, we observed a net increase in AFB 1 -associated Can R mutations in the in csm2 mutant, compared to wild type (P , 0.05). Average survival was only slightly higher in the wild type (51%) than in the csm2 mutant (49%), but not statistically different (P = 0.8, N =4). We suggest that similar to MMS-associated DNA lesions, CSM2 functions to suppress AFB 1 -associated mutagenesis while promoting template switching of AFB 1 -associated DNA adducts.
If CSM2 participates in a RAD51-dependent recombinational repair pathway to tolerate AFB1-associated DNA lesions, then we would expect that RAD51 would be epistatic to CSM2 for AFB 1 resistance (Glassner and Mortimer 1994). We measured AFB 1 sensitivity in the csm2, rad51 and csm2 rad51 haploid mutants compared to wild type, using growth curves ( Figure 6). Our data indicate the csm2 rad51 double mutant is no more AFB 1 sensitive compared to either rad51 single mutants indicating that CSM2 and RAD51 are in the same epistasis group for AFB 1 sensitivity. In contrast, csm2 rad4 double mutants are more sensitive to AFB 1 than either the csm2 and rad4 single mutants; the fitness measurement of the double mutant (0.071) is also less than the product of the csm2 (0.34) and rad4 (0.28) single mutants. These data indicate that CSM2 participates in a n■ Represents the number present among 79 genes. c See Table 1 and Table 2 for full description.
Human orthologs for many essential yeast genes can directly complement the corresponding yeast genes (Kachroo et al. 2015). Human homologs are listed for 46 yeast AFB 1 -resistance genes ( Table 6). These homologs include those for DNA repair, DNA damage tolerance, cell cycle, and cell maintenance genes. Several of these genes, such as the human CTR1 (Zhou and Gitschier 1997), can directly complement the yeast gene. Other DNA human DNA repair genes, such as those that encode RAD54 (Kanaar et al. 1996), RAD5 (Unk et al. 2010), and RAD10 orthologs, can partially complement sensitivity to DNA damaging agents (Aggarwal and Brosh 2012).

DISCUSSION
Human CYP1A2-mediated activation of the mycotoxin AFB 1 generates a highly reactive epoxide that interacts with DNA, RNA, and protein, forming adducts which interfere in replication, transcription, and protein function. Previous experiments have documented the role of checkpoint genes, RAD genes, and BER genes in conferring AFB 1 resistance in budding yeast (Keller-Seitz et al. 2004;Guo et al. 2005;Fasullo et al. 2010). The goal of this project was to identify additional AFB 1 resistance genes that may elucidate why AFB 1 is a potent yeast recombinagen but a weak mutagen (Sengstag et al. 1996).
Here, we profiled the yeast genome for AFB 1 resistance using three yeast non-essential diploid deletion libraries; one was the original library and the other two expressed human CYP1A2. We identified 96 ORFs, of which 86 have been ascribed a function and 79 were confirmed to be AFB 1 sensitive, relative to the wild type. These resistance genes reflect the broad range of functions, including cellular and metabolic processes, actin reorganization, mitochondrial responses, and DNA repair. Many of the DNA repair genes and checkpoint genes have been previously identified in screens for resistance to other toxins (Lee et al. 2005;Giaever and Nislow 2014;De La Rosa et al. 2017). While individual resistance genes are shared Figure 3 GO enrichment 79 AFB 1 resistance genes, according to biological process and protein function. The top circles represent the number of genes of that are grouped according to Process, using the Generic Gene Ontology Term Finder, http://go.princeton.edu/cgibin/GOTermFinder. The bottom circles are those which are grouped according to protein function, using GO term finder according to function, https://www.yeastgenome.org/goSlimMapper. GO process groups included translesion synthesis, DNA synthesis involved in DNA repair, postreplication repair, DNA biosynthetic process, cellular response to DNA damage, DNA repair, meiotic cell cycle process, response to stress, and cellular response to stimulus. GO function groups include ion binding, hydrolase, nucleic acid binding, oxidoreductase, and transcription factors. Number of genes belonging to each GO is indicated within or just outside the pie slice.
among diverse toxins, such as doxorubicin, nystatin, cycloheximide, rapamycin, and amphotericin, the top ten AFB 1 -associated GO enrichments are not represented among these diverse toxins. However, AFB 1 -associated GO enrichments, including postreplication repair and translesion synthesis, are shared among genes that confer resistance to cross-linking agents, such as trichloroethylene (De La Rosa et al. 2017) and cisplatin (Lee et al. 2005). One similarity between cross-linking agents and metabolically activated AFB 1 is that they can form DNA adducts that impede DNA replication.
While mitochondrial maintenance genes and oxidative stress genes (Amici et al. 2007;Mary et al. 2012) are expected AFB 1 resistance genes based on studies of individual mutants (Guo et al. n■  b Number of expected genes among initial set of 79 ORFs based on reference list of 6026 genes; fractional values less than 1 were designated as ,1. c Fold enrichment is the ratio of the number of AFB1 resistant genes identified to the expected number. d Displaying only results for Bonferroni-corrected probabilities, P , 0.05. e Significance indicated by "+" Figure 4 The protein interactome encoded by AFB 1 resistance genes in budding yeast (left, A) and protein interactome encoded by their associated human homologs (right, B). The interactome was curated using String V11 (https://string-db.org, Szklarczyk et al. 2019), using a high confidence level of 0.7 and MCL cluster factor of 1.1. Proteins are represented by colored circles (nodes); different colors represent distinct interacting clusters. A core group, in red seen in both images, includes proteins that function in DNA repair pathways, and interact with proteases, transcription and cell cycle factors. Lines represent the edges; a solid blue line indicates a binding event, a dark line indicates a reaction, and a purple line indicates catalysis. The lighter lines indicate a strong connection, as deduced from the literature. Lines that terminate with a dot indicate an unspecified interaction, whether positive or negative.
2005; Fasullo et al. 2010), we also identified novel AFB 1 resistance genes that participate in DNA postreplication repair, both by modulating checkpoint responses and by recombination-mediated mechanisms. Of key importance, the CSM2/SHU complex (Shor et al. 2005) was required for AFB 1 -associated sister chromatid recombination, underscoring the role of recombination-mediated template switch mechanisms for tolerating AFB 1 -associated DNA damage. Since many yeast genes are conserved in mammalian organisms (Bernstein et al. 2011), we suggest similar mechanisms for tolerating AFB 1 -associated DNA damage may be present in mammalian cells.
We used a novel reagent consisting of a pooled yeast library expressing human CYP1A2, on a multi-copied expression vector. Because CYP1A2 activates AFB 1 when toxin concentration is low (Eaton and Gallagher 1994), our modified yeast library mimicked AFB 1 activation when low hepatic AFB 1 concentrations generate DNA adducts. One limitation of the screen in the CYP1A2-expressing library is that AFB 1 -associated toxicity is not directly proportional to AFB 1 concentration (Fasullo et al. 2010); we speculate that CYP1A2 activity is the limiting factor. Although individual yeast strains expressed similar amounts of CYP1A2 activity from among the subset of deletion strains tested, it is still possible that profiling resistance among individual deletion strains is influenced by the stability or variable expression of the membrane-associated human CYP1A2 (Murray and Correia 2001). Second, many DNA repair and checkpoint genes that were previously documented to confer resistance, such as RAD52, were not identified in the screen (Guo et al. 2006;Fasullo et al. 2010). One possible reason is that some, such as rad52, grow poorly (Figure 1, Fasullo et al. 2008), and we suspect that other slow-growing strains dropped out early in the time course of exposure. Future experiments will more carefully assess generation times needed to detect known resistance genes in the library expressing CYP1A2.
Because metabolically activated AFB 1 causes protein, RNA, and lipid damage (Weng et al. 2017), besides DNA damage, we expected to find a functionally diverse set of AFB 1 resistance genes. Among AFB 1 resistance genes were those involved in protein degradation and ammonia transport, actin reorganization, tRNA modifications, ribosome biogenesis, RNA translation, mitochondrial, and metabolic functions. Some genes encoding these functions, such as BIT2 and TRM9, have important roles in maintaining genetic stability and in double-strand break repair (Schmidt et al. 1996;Begley et al. 2007;Schonbrun et al. 2013;Shimada et al. 2013). A more direct role in DNA repair mechanisms have been noted for FUM1 (Leshets et al. 2018;Saatchi and Kirchmaier 2019), DPB3 (Gallo et al. 2019), and BLM10 (Qian et al. 2013). Several genes, such as PPG1, are involved in glycogen accumulation; these genes are also required to enter the quiescent state . Other genes are involved in cell wall synthesis, including MNN10, SCW10 and ROT2; we speculate that cell wall synthesis genes confer resistance by impeding AFB 1 entrance into the cell, while genes involved in protein degradation in the ER may stabilize CYP1A2 and thus enhance AFB 1 -conferred genotoxicity. Indeed, two resistance genes, CUE1 and SSM4, are associated with degradation of mammalian P450 proteins in yeast (Murray and Correia 2001). Glucan and other cell wall constituents have also been speculated to directly inactivate AFB1, and yeast fermentative products are supplemented in cattle feed to prophylactically reduce AFB1 toxicity (Pereyra et al. 2013).
Although AFB 1 -associated cellular damage is associated with oxidative stress (Amici et al. 2007;Mary et al. 2012;Liu and Wang 2016) only a few yeast genes that confer resistance to reactive oxygen species (ROS) were identified in our screens. These genes included TRX3, YND1, VPS13, BIT2, GTP1, FKH2, SHH3, NRP1, and BUD20, which have a wide variety of functions (Schmidt et al. 1996;Greetham and Grant 2009;Bassler et al. 2012). While known genes associated Figure 5 AFB 1 -associated sister chromatid recombination and mutagenesis frequencies in the wild type and csm2 haploid mutant. The top part of each panel shows the assays for sister chromatid exchange and mutagenesis; for both assays the oval represents the centromere and the single line represents duplex DNA. For simplicity, the left arm of chromosomes IV and V are not shown. (Top left, A) Unequal sister chromatid recombination is monitored by selecting for His + prototrophs that result from recombination between the juxtaposed, truncated his3 fragments. The his3-D39 lacks the 39 sequences (arrow head), while the his3-D59 lacks to promoter sequences (feathers). Both his3 fragments are located with the amino acid reading frames oriented to the centromere. The his3 fragments share a total of 450 bp sequence homology. Bottom of panel A shows the frequencies of unequal sister chromatid exchange (SCE) obtained from the wild type (YB204 pCS316) and the haploid csm2 mutant (YB559) after exposure to 0, 15, and 25 mM AFB 1 . (Top right, B) The arrow notes the occurrence of point, missense, or deletion mutations that can occur in the CAN1 gene and result in canavanine resistance (Can R ) mutations. Bottom of panel B shows the frequencies of (Can R ) mutants in wild type (YB204 pCS316) and csm2 (YB559) strain after exposure to 0 and 50 mM AFB 1 . For complete genotype, see Table S1.
with oxidative stress and oxidative-associated DNA damage, such as YAP1, SOD1, and APN1, were not identified, other mitochondrial genes, such as TRX3, were identified. Guo et al. (2005) also showed that the haploid apn1 mutant was not AFB 1 sensitive. We offer two different explanations: first, the AFB 1 -associated oxidative damage is largely localized to the mitochondria, and second, there may be redundant pathways for conferring resistance to AFB 1 -associated oxidative damage, and therefore single genes were not identified. It is most likely the later as screens with oxidants (like t-BuOOH) also fail to identify expected antioxidant enzymes (Said et al. 2004).
A majority of the AFB 1 resistance genes belong to GO groups that include postreplication repair, DNA damage-inducible genes, DNA repair, response to stimulus, or response to replication stress (Table  3). Proteins encoding functionally diverse genes, such as RAD54, RAD5, GFD1, TMA20, SKG3, GRE3, and ATG29, are repositioned in the yeast cells during DNA replication stress (Tkach et al. 2012). The requirement for unscheduled DNA synthesis was illustrated by identifying genes involved in the DNA damage-induced expression of ribonucleotide reductase; these included RNR3 and TRM9. TRM9, involved in tRNA modification, functions to selectively translate DNA damage-inducible genes, such as RNR1 (Begley et al. 2007).
One unifying theme was that cell cycle progression and recovery from checkpoint-mediated arrest is a prominent role in mediating toxin resistance. FKH2 functions as a transcription factor that promotes cell cycle progression and G 2 -M progression. Other genes are involved in the modulation of the checkpoint response, such as PSY2. While CKB1 and CKB2 have broad functions, including histone phosphorylation and chromatin remodeling (Barz et al. 2003;Cheung et al. 2005), CKB2 is also required for toleration of double-strand breaks (Guillemain et al. 2007;Toczyski et al. 1997), and thus may function for the toleration of AFB 1 -associated damage. These genes support the notion that some of the AFB 1 -associated DNA adducts are well tolerated and can be actively replicated.
While we expected to identify individual genes involved in DNA damage tolerance, such as RAD5, REV1, REV3, the CSM2/PSY3 complex is novel. Absence of CSM2 confers deficient AFB 1 -associated SCE but higher frequencies of AFB 1 -associated mutations, suggesting that CSM2 functions to suppress AFB 1 -associated mutations by RAD51-mediated template switch mechanisms (Guo et al. 2006;Fasullo et al. 2008). Consistent with this idea, RAD51 is epistatic to CSM2 in conferring AFB 1 sensitivity, while rad4 cms2 double mutant exhibits synergistic AFB 1 sensitivity with respect to the single rad4 and csm2 single mutants. However, rad4 rad51 mutants still exhibit more AFB 1 sensitivity than rad4 csm2, suggesting that RAD51 may be involved in conferring resistance to other AFB 1 -associated DNA lesions, such as double-strand breaks. We also expect that the RAD51 paralogs, RAD55 and RAD57, share similar AFB 1 -associated functions with RAD51. Considering that RAD57 is the XRCC3 ortholog, determining Figure 6 Growth curves of wild type (YB155), csm2 (YB557), rad4 (YB661), rad51 (YB177), csm2 rad51 (YB663), and csm2 rad4 (YB660) haploid cells after exposure to 50 and 100 mM AFB 1 . (Left) Growth of cells containing pCS316 and expressing CYP1A2 after chronic exposure to 0.5% and 1.0% DMSO (black), 50 mM (red), and 100 mM (red) AFB 1 . The relevant genotype is given above the panel (see Table S1, for complete genotype). Approximately 10 5 log-phase cells were inoculated in each well, n = 2. A 600 is plotted against time (hs). Bars indicate the standard deviations of measurements, n = 2.
whether yeast RAD51 paralogs suppress AFB 1 -associated mutation will aid in identifying similar complexes in mammalian cells. Such complexes may elucidate why XRCC3 polymorphisms are risk factors in AFB 1 -associated liver cancer (Long et al. 2008;Ji et al. 2015).
Human homologs of several of the identified yeast genes (Table 6) are, hyper-methylated, mutated, or over-expressed in liver disease and cancer. For example, mutations and promoter methylations of the human RAD5 ortholog, Helicase-Like Transcription Factor (HLTF), are observed in hepatocellular carcinoma (Zhang et al. 2013;Dhont et al. 2016). Mutations in PRKCSH and GANAB, human orthologs of GTB1 and ROT2, are linked to polycystic liver disease (Porath et al. 2016;Perugorria and Banales 2017). The CKB2 ortholog, CSNK2B (Zhang et al. 2015;Chua et al. 2017;Dotan et al. 2001), is over-expressed in several liver cancers and therapeutics are currently in clinical trial (Gray et al. 2014;Li et al. 2017;Trembley et al. 2017). It is tempting to speculate that over-expression of CSNK2B also confers AFB 1 resistance.
Human homologs of other yeast genes have been correlated to the etiology and progression of other cancers, including colon and renal cancer. These include the human homolog for TRM9, ALKB8, and the human homolog for TMA20, MCT1, which can complement translation defects of tma20 mutants and has been implicated in n■ a For full description, see Table 1 and Table 2.
modulating stress (Herbert et al. 2001) and double-strand break repair (Hsu et al. 2007). MCT-1 overexpression and p53 is noted to lead to synergistic increases in chromosomal instability (Kasiappan et al. 2009). Heterozygous germline mutations of fumarate hydratase (FH) predispose for hereditary leiomyomatosis and renal cell carcinoma (Lehtonen et al. 2006). In both mammalian and yeast cells, FH participates in double-strand break repair (Leshets et al. 2018) and may thus suppress genetic instability. In summary, we profiled the yeast genome for AFB 1 resistance and identified novel genes that confer resistance. The novel genes included those involved in tRNA modifications, RNA translation, DNA repair, protein degradation, and actin reorganization. Genes that function in DNA damage response and DNA damage tolerance were over-represented, compared to the yeast genome. We suggest that the CSM2 (SHU complex) functions to promote error-free replication of AFB 1 -asociated DNA damage, and it will be interesting to determine whether mammalian orthologs of the SHU complex function similarly.

ACKNOWLEDGMENTS
We acknowledge William Burhans for his initial encouragement and support, and grants from the National Institutes of Health: R21ES1954, F33ES021133, and R15ES023685-03. We would like to thank Chris Vulpe for his gift of the pooled BY4743 library and guidance, Mingseng Sun for initiating work on AFB 1 -induced recombination, Jonathan Bard for bioinformatics expertise and valuable assistance in data processing, and Adam Smith and Akaash Kannan for technical contributions. Additional help was provided by Michael Dolan and Kevin Lin with bioinformatics.