Natural variation in a single amino acid underlies cellular responses to topoisomerase II poisons

Many medications, including chemotherapeutics, are differentially effective from one patient to the next. Understanding the causes of these population-wide differences is a critical step towards the development of personalized treatments and improvements to existing medications. Here, we investigate natural differences in sensitivity to anti-neoplastic drugs that target topoisomerase II, using the model organism Caenorhabditis elegans. We show that wild isolates of C. elegans vary in their sensitivity to these drugs, and we use an unbiased statistical and molecular genetics approach to demonstrate that this variation is explained by a methionine-to-glutamine substitution in topoisomerase II (TOP-2). The presence of a non-polar methionine at this residue increases hydrophobic interactions between TOP-2 and the poison etoposide, as compared to a polar glutamine. We hypothesize that this stabilizing interaction results in increased genomic instability in strains that contain a methionine residue. The residue affected by this substitution is conserved from yeast to humans and is one of the few differences between the two human topoisomerase II isoforms (methionine in hTOPIIα and glutamine in hTOPIIβ). We go on to show that this substitution influences binding and cytotoxicity of etoposide and two additional topoisomerase II poisons in human cell lines. These results explain why hTOPIIα and hTOPIIβ are differentially affected by various poisons and demonstrate the utility of C. elegans in understanding the genetics of drug responses.


Introduction
Antineoplastic regimens used to treat cancer are often associated with poor prognoses and severe side effects. Ideally, antineoplastic regimens can be tailored to an individual patient based on various genetic markers known to be associated with drug response to maximize therapeutic effectiveness and minimize unwanted side effects. Advances in sequencing technologies over the course of the past decade promised the discovery of many genetic variants that contribute to human health. Though large-scale sequencing projects have lead to the identification of many genetic variants associated with disease risk (Visscher et al. 2012), relatively few variants have been identified that contribute to clinically relevant traits such as response to antineoplastic compounds. In fact, only 71 of over 500 FDA-approved antineoplastic compounds use genetic information to affect treatment efficacy (www.fda.gov).
Unfortunately, the predictive power of these identified genetic variants can be inconsistent due to biases in the sampled population (Boddy 2013) and other key limitations of clinical genomewide association (GWA) studies that attempt to link genetic variants with treatment outcomes.
The major factor limiting the efficacy of these studies is sample size because it is difficult to identify large numbers of individuals exposed to the same antineoplastic regimens. This limitation is compounded when considering environmental (Liu et al. 2013;Hunter 2005) and tumor heterogeneity (Koboldt et al. 2013). As a result, most variants discovered to be associated with outcomes in clinical GWA studies offer low predictive power for patient responses to treatment (Park et al. 2012). These limitations and others emphasize the need for novel approaches to identify variants that predict patient outcomes to antineoplastic compounds.
Studies of model organisms have greatly facilitated our understanding of basic cellular processes. In recent years, Saccharomyces cerevisiae and Drosophila melanogaster have been used to understand the physiological effects of small molecules and repurposed as screening platforms to identify new antineoplastic compounds (Willoughby et al. 2013;Perlstein et al. 2007; King et al. 2014). The ability to generate extremely large numbers of recombinant yeast facilitates the identification of genomic regions that are predictive of drug response (Ehrenreich et al. 2010;Bloom et al. 2013). Furthermore, the specific genes and variants within regions can be identified and functionally validated in yeast (Demogines et al. 2008;Liti & Louis 2012;Stern 2014). By contrast, D. melanogaster studies offer the ability to study the physiological responses to drugs in the context of multiple tissue types, but functional validation of specific genes and variants associated with drug responses has been more limited (King et al. 2014).
The roundworm Caenorhabditis elegans has the advantages of both S. cerevisiae and D. melanogaster because large cross populations can be generated to study the physiological responses to drugs in a metazoan. These attributes have made C. elegans an important model for connecting differential drug responses with genetic variants present in the species (Ghosh et al. 2012;Andersen et al. 2015).
Here, we take advantage of natural genetic variation present in C. elegans to identify the genetic basis underlying susceptibility to a panel of clinically relevant antineoplastic compounds that poison the activity of topoisomerase II enzymes. The inhibition of these enzymes by topoisomerase II poisons results in the accumulation of double-stranded breaks and genome instability (Pommier et al. 2010;Pommier et al. 1984;Gómez-Herreros et al. 2013).
Topoisomerase II enzymes are targeted by antineoplastic regimens because proliferative cell populations require their enzymatic activity to relieve topological stress ahead of the replication fork (Nitiss 2009). Using two unbiased genetic mapping approaches, we show that divergent physiological responses to the topoisomerase II poison etoposide are determined by natural genetic variation in a C. elegans topoisomerase II enzyme. Furthermore, we show using CRISPR/Cas9-mediated genome editing that variation in a specific amino acid (Q797M) underlies the cytotoxic effects of etoposide. This residue is conserved in humans and is one of the few differences between the putative drug-binding pockets of the two topoisomerase II isoforms (M762 in hTOPIIα and Q778 in hTOPIIβ). Previous structural studies on hTOPIIβ implicated this glutamine residue in etoposide binding because of its proximity to the drugbinding pocket (Wu et al. 2011;Wu et al. 2013). However, a study on hTOPIIα suggested that the corresponding methionine residue has no functional role in drug binding (Wendorff et al. 2012). We present a mechanistic model to explain how variation at this residue underlies differential responses to etoposide and other topoisomerase II poisons. Finally, we use genomeedited human cell lines to show that this residue in hTOPIIα contributes to differential toxicity of various topoisomerase II poisons. These results demonstrate the power of using C. elegans natural genetic variation to identify mechanisms of drug susceptibility in human cells that could inform human health decisions based on genetic information.

A single major-effect locus explains variation in response to etoposide
We investigated etoposide sensitivity in C. elegans using a high-throughput fitness assay. In brief, animals were grown in liquid culture in presence of etoposide, and body lengths of progeny and offspring production were measured using a COPAS BIOSORT (Supplemental figure 1). In this assay, shorter body lengths are indicative of developmental delay. To identify an appropriate dose of etoposide for this assay, we performed dose-response experiments on four genetically diverged isolates of C. elegans: N2 (Bristol), CB4856 (Hawaii), JU258, and DL238. We chose 250 µM etoposide for further experiments because it was the lowest concentration at which we observed an etoposide-specific effect in all four strains tested, trait differences between the laboratory Bristol strain (N2) and a wild strain from Hawaii (CB4856) strains were maximized, and the median animal length was highly heritable (Supplemental figure 2).
When grown in etoposide, progeny of the Hawaii strain are on average 75 µm shorter than progeny of the Bristol strain. To map the genetic variants underlying this difference, we performed our high-throughput fitness assay on a panel of 265 recombinant inbred advanced intercross lines (RIAILs), generated between a Bristol derivative (QX1430) and Hawaii (Andersen et al. 2015). We measured median animal length for each RIAIL strain grown in etoposide, and we corrected for assay-to-assay variability and effects of the drug carrier (DMSO) using a linear model. We used the resulting regressed median animal length trait (referred to as animal length) for quantitative trait locus (QTL) mapping. This mapping identified a major-effect QTL for etoposide resistance on chromosome II at 11.83 Mb ( Figure 1A). This QTL explained 27% of the phenotypic variance among the recombinant lines. The QTL confidence interval spans from 11.67 to 11.91 Mb on chromosome II and contains 90 genes, 68 of which contain variation between the parental strains.
We next sought to validate this QTL using homozygous reciprocal near-isogenic lines (NILs), which contain either the QTL confidence interval from the Bristol strain introgressed into the Hawaii strain or the interval from the Hawaii strain introgressed into the Bristol strain. NILs with the genomic interval derived from the Bristol strain have increased resistance to etoposide compared to the Hawaii strain (Supplemental figure 3). Similarly, NILs with the genomic interval derived from the Hawaii strain exhibited decreased resistance to etoposide. These results confirmed that genetic variation located on the right arm of chromosome II contributes to differential etoposide susceptibility.

The same locus on chromosome II explains variation in response to etoposide in a panel of wild C. elegans isolates
In the initial dose response experiments, we found that JU258 and DL238 had different responses to etoposide than the Bristol and Hawaii strains, suggesting that additional genetic variation present in the wild C. elegans population could also contribute to etoposide response.
To identify this additional variation, we performed a genome-wide association (GWA) mapping of etoposide resistance in 138 wild C. elegans isolates. This analysis led to the identification of a QTL on the right arm of chromosome II with a peak position at 11.88 Mb ( Figure 1B). This QTL has a genomic region of interest that spans from 11.70 to 12.15 Mb for which we found no evidence of selection (Supplemental figure 4) or geographic clustering of the peak QTL allele (Supplemental figure 5). In addition, this QTL overlaps with the QTL identified through linkage mapping described above. Of the 138 wild isolates assayed, including the Hawaiian strain, 46 have the alternate (non-Bristol) genotype at the peak position on chromosome II ( Figure 1C). Similar to our observations using the recombinant lines, the 46 strains that contain the alternate genotype are more sensitive to etoposide than strains containing the Bristol genotype at the QTL peak marker. We hypothesized that variation shared between the Hawaiian strain and the other 45 alternate-genotype strains contributes to etoposide sensitivity because we detected overlapping QTL, with the same direction of effect, between GWA and linkage mapping experiments. This hypothesis suggested that we could condition a fine-mapping approach on variants found in the Hawaiian strain and shared across these 45 strains.
To fine-map the QTL, we focused on variants shared among wild isolates. Using data from the C. elegans whole-genome variation dataset (Cook, Zdraljevic, Tanny, et al. 2016) we calculated Spearman's rho correlations between animal length and each single-nucleotide variant (SNV) in the QTL confidence interval ( Figure 1D). SNVs in only three genes, npp-3, top-2, and ZK930.5, were highly correlated with the etoposide response (rho > 0.45). Of these genes, the top-2 gene encodes a topoisomerase II enzyme that is homologous to the two human isoforms of topoisomerase II. We prioritized top-2 because topoisomerase II enzymes are the cellular targets for etoposide (Pommier et al. 2010).

Genetic variation in top-2 contributes to differential etoposide sensitivity
To determine if genetic variation present in the top-2 gene contributes to differential etoposide sensitivity, we performed a reciprocal hemizygosity test (Stern 2014). Prior to this test, we determined that resistance to etoposide is dominant by measuring the lengths of F1 heterozygotes from a cross between the Bristol and Hawaii strains in the presence of etoposide (Supplemental figure 6). Additionally, we tested npp-3 and top-2 deletion alleles from the Bristol genetic background and found that only loss of top-2 contributes to etoposide sensitivity (Supplemental figure 7). To more definitively show a causal connection of top-2 variation to etoposide sensitivity, we used a reciprocal hemizygosity test. First, we introgressed the top-2(ok1930) deletion allele into the Hawaiian genetic background. The  heterozygote that contains the Bristol top-2 allele is more resistant to etoposide treatment than the Hawaii/Bristol(∆top-2) heterozygote, which suggests that the Bristol top-2 allele underlies etoposide resistance (Figure 2A, Supplemental figure 8). The observed differences between the Hawaii/Bristol(∆top-2) and Bristol/Hawaii(∆top-2) heterozygotes confirmed that top-2 variation underlies differential susceptibility to etoposide.

A glutamine-to-methionine variant in TOP-2 contributes to etoposide response
To identify genetic variants in top-2 that contribute to etoposide resistance in the Bristol strain, we focused on genomic differences between the Bristol and Hawaii strains. Based on gene expression data between the Bristol and Hawaii strains (Rockman et al. 2010), top-2 is expressed at similar levels. Therefore, we concluded that etoposide resistance in the Bristol strain is likely caused by coding variation. The C. elegans top-2 gene contains 31 SNVs across the population-wide sample of 138 wild isolates. We narrowed our search to 16 variants present in the Hawaiian strain. Two of these variants are in the 3' UTR, three are in introns, and six are synonymous variants that likely do not contribute to etoposide resistance. The remaining five variants encode for amino acid changes in the TOP-2 enzyme. Of these five variants, four were highly correlated with etoposide sensitivity in the wild isolate panel: Q797M, I1206L, Q1217A, and D1387N. Multiple-sequence alignment of TOP-2 peptides across yeast, D. melanogaster, mice, and humans revealed that I1206L, Q1217A, and D1387N are in the variable C-terminal domain (Supplemental file 1). By contrast, the Q797M variant is located in the conserved DNA binding and cleavage domain (Schmidt et al. 2012). Structural data suggest that the TOP-2 Q797 residue lies within the putative etoposide-binding pocket (Wu et al. 2011), and the corresponding residue is a methionine (M762) in the hTOPOIIα and a glutamine (Q778) in hTOPOIIβ (Wendorff et al. 2012). Additionally, the two human isoforms differ in one other residue within the putative etoposide-binding pocket (S800(α)/A816(β)). Therefore, the C. elegans glutamine-to-methionine TOP-2 variant mirrors one of two differences within the etoposide-binding pocket of the two human topoisomerase II enzyme isoforms. Crucially, hTOPOIIα forms a more stable DNA-TOPOII cleavage complex with etoposide than hTOPOIIβ (Bandele & Osheroff 2008). We hypothesized that etoposide sensitivity in both C. elegans and the human isoforms is affected by this residue.
To test the effects of the Q797M variant on C. elegans response to etoposide, we used CRISPR/Cas9-mediated genome editing to change this residue. We replaced the glutamine residue in the Bristol strain with a methionine and the methionine residue in the Hawaii strain with a glutamine. We exposed the allele-replacement strains to etoposide and found that the methionine-containing Bristol animals were more sensitive than glutamine-containing Bristol animals ( Figure 2B). Conversely, the glutamine-containing Hawaii animals were more resistant to etoposide than the methionine-containing Hawaii animals ( Figure 2B). These results confirm that this variant contributes to differential etoposide sensitivity between the Bristol and Hawaii strains.

Methionine mediates stronger hydrophobic interactions with etoposide than glutamine
We hypothesized that the non-polar functional group attached to the glycosidic bond of etoposide contributes to increased stability of the drug-enzyme complex by forming a more stable interaction with the methionine residue than with the glutamine residue. To test this hypothesis, we simulated etoposide docking into the putative drug-binding pocket of the TOP-2 homology model generated by threading the C. elegans peptide sequence into the hTOPOIIβ structure (RMSD = 1.564 Å, PDB:3QX3; (Wu et al. 2011)). Upon etoposide binding, the free energy (∆G) of the drug-binding pocket was -10.09 Kcal/mol for TOP-2 Q797 ( Figure 3A) and -12.67 Kcal/mol for TOP-2 M797 ( Figure 3B). Therefore, we expect C. elegans strains that contain a methionine at this residue to accumulate more genomic damage when exposed to etoposide. The resulting physiological effect of increased genomic damage likely delays development and causes the progeny of exposed individuals to be shorter.

TOP-2 variation causes allele-specific interactions with an expanded set of topoisomerase II poisons
Because the molecular docking simulations explain the observed physiological effects of etoposide exposure, we hypothesized that the 797 residue of TOP-2 would mediate differential interactions with additional topoisomerase II poisons based on their chemical structures. Like etoposide, teniposide, dactinomycin, and amsacrine each contain core cyclic rings that are thought to interfere with the re-ligation step of the topoisomerase II catalytic cycle through DNA interactions (Pommier et al. 2010). However, the functional groups attached to the core cyclic rings of each poison vary in their polarity and size, which could affect interactions with topoisomerase II enzymes. For example, the only difference between teniposide and etoposide is the presence of a thienyl or methyl group attached to the D-glucose derivative, respectively, but they share a similarly sized and hydrophobic functional group. We predicted that these two drugs would have comparable interactions with the TOP-2 alleles and elicit a similar physiological response. By contrast, the polar functional groups of dactinomycin likely have stronger interactions with the glutamine variant and induce increased cytotoxicity in animals that contain this allele. We quantified the physiological responses of the TOP-2 allele-replacement strains exposed to these two drugs and found that each response matched our predictions ( Figure 4). Specifically, strains harboring the TOP-2 methionine allele were more sensitive to teniposide than those strains that contain the glutamine allele. Conversely, strains with the TOP-2 glutamine allele were more sensitive to dactinomycin than those strains with the methionine allele. Unlike etoposide, teniposide, or dactinomycin, the core cyclic rings of amsacrine do not have an equivalent functional group to interact with the TOP-2 797 residue, suggesting that variation at TOP-2 residue 797 will have no impact on amsacrine sensitivity. Although the Bristol and Hawaiian strains differed, we found that the allele status of TOP-2 had no quantifiable effect on amsacrine response ( Figure 4) and different genomic loci control response to this drug (Supplemental figure 9). These results support the hypothesis that the polarity of the putative drug-binding pocket determines the cytotoxic effects of multiple, but not all, topoisomerase II poisons. To further explore this hypothesis, we tested a drug (XK469) that has preferential hTOPOIIβ specificity (Gao et al. 1999). Surprisingly, we found that the strains that contain the methionine allele (like hTOPOIIα) were more sensitive to XK469 (Supplemental figure 10). This result indicates that an additional mechanism might contribute to XK469 specificity in human cells and underscores the importance of functional validation of specific residues that are thought to be involved in targeted drug binding.

Variation in the equivalent site in topoisomerase II alpha causes differential susceptibility to diverse poisons in human cells
To determine if differences in the hydrophobicities of the two human topoisomerase II putative drug-binding pockets underlie etoposide sensitivity, we used CRISPR/Cas9 genome editing and a pooled-sequencing approach to create human embryonic kidney 293 cells (293T) that encode hTOPOIIα enzymes with a hTOPOIIβ-like drug-binding pocket. Cells were incubated with genome-editing machinery for six hours, allowed to recover for five days, and then split into two populations for etoposide exposure or no etoposide exposure. Etoposide treatment provided a selective pressure that upon further passaging led to a greater than 160fold enrichment of cells that contain the glutamine-edited hTOPOIIα allele as compared to populations of cells exposed to no drug ( Figure 5). These results show that cells with the glutamine-edited hTOPOIIα allele are more resistant to etoposide treatment than cells with the non-edited methionine hTOPOIIα allele. Notably, the rarity of genome editing events makes it unlikely that every copy of the hTOPOIIα gene in this diploid/polyploid cell line is edited.
Because we see etoposide resistance in these incompletely edited cells, hTOPOIIα dimeric complexes likely contain one edited and one wild-type copy of hTOPOIIα and do not bind etoposide as well as causing less cytotoxicity. These data confirm both our dominance test (Supplemental figure 6) and the two-drug model of etoposide binding (Bromberg et al. 2003) in which both enzymes of the homodimer must be bound by poison to be completely inhibited.
Additionally, we performed the reciprocal experiment to edit the glutamine-encoding hTOPOIIβ gene to a version that encodes methionine. If the methionine hTOPOIIβ allele is more sensitive to etoposide than the glutamine hTOPOIIβ, we would expect to observe a depletion of methionine-edited cells upon etoposide treatment. However, because glutamine-to-methionine editing occurred in less than 1% of the cells, it was difficult to detect further reductions in methionine allele frequencies (Supplemental table 1). Overall, we demonstrate that this residue underlies variation in etoposide response in both C. elegans and human cell lines.
Our C. elegans results using the genome-edited top-2 strains show that variation at this residue underlies differences in some but not all topoisomerase II poisons. We exposed the edited human cell lines to these different poisons to test this hypothesis. We found that cells containing the glutamine-edited hTOPOIIα allele are less affected by both teniposide and XK469, as indicated by the respective 8.2-and 2.8-fold increase in edited cell frequency upon drug exposure as compared to cell populations with no drug added. These results mirror our observations that C. elegans strains with the methionine TOP-2 allele are more sensitive to these drugs. We observed moderate-to-no change in edited allele frequencies between cell populations exposed to amsacrine (~1.6-fold increase) or dactinomycin (~0.93-fold decrease) and those cells exposed to no-drug control conditions. These results indicate that the hTOPOIIα M762 residue does not interact with amsacrine. Our expectation was that dactinomycin would be more cytotoxic to cells that contain the glutamine hTOPOIIα allele, and therefore result in a depletion of edited cells. As mentioned above, we do not have the power to detect this depletion in response because of low CRISPR-editing efficiency. However, we did expect to see enrichment of hTOPOIIβ Q778M edited cells upon exposure to dactinomycin in the reciprocal experiment. Despite this expectation, we saw no change in hTOPOIIβ Q778M edited cell frequency between dactinomycin and the no drug control. Our inability to detect any enrichment of allele frequencies might result from dactinomycin having multiple cellular targets in addition to topoisomerase II (Koba & Konopa 2005). Taken together, our findings testing a variety of poisons on human cell lines recapitulated the results from C. elegans.

Discussion
Few genetic markers have been identified that predict patient responses to chemotherapeutic regimens (Moen et al. 2012;Giacomini et al. 2007). The goal of this study was to introduce new methods for the rapid and cost-effective identification of genetic variants that explain differences in chemotherapeutic response. Our approach leveraged genetic and phenotypic variation present in the model organism C. elegans to identify a single amino acid variant (Q797M) in the topoisomerase II enzyme that underlies differences in etoposide response. Mechanistic insights into differential etoposide binding between the glutamine or methionine alleles gave us the power to predict the physiological responses to an expanded panel of topoisomerase II poisons. These results highlight how the combination of a highly sensitive phenotyping assay with classical and quantitative genetics approaches in C. elegans can rapidly identify the mechanistic underpinnings of phenotypic variability in response to a key class of antineoplastic compounds.
Our approach stands in stark contrast to previous underpowered human cell line (Huang et al. 2007) and clinical studies (Low et al. 2013) that failed to identify any statistically significant associations between etoposide-induced cytotoxicity and genetic variation in the human population. However, the residue we identified in C. elegans does not vary in the human population (Lek et al. 2016), suggesting that GWA studies would not have identified this variant as a marker for etoposide sensitivity. Nevertheless, this residue is one of the few differences between the putative drug-binding pockets of the two human topoisomerase II isoforms (M762 in hTOPIIα and Q778 in hTOPIIβ), which allowed us to investigate the molecular underpinnings of drug binding. We verified that this single amino acid change in the human topoisomerase II isoforms results in profound differences in topoisomerase II poison-induced cytotoxicity using 293T cells. Though previous hTOPOIIβ structural studies have implicated this glutaminemethionine difference as functionally important for etoposide binding (Wu et al. 2011;Wu et al. 2013), studies involving hTOPOIIα have argued that this residue is not involved (Wendorff et al. 2012). The results presented here unequivocally show that this residue contributes to differential topoisomerase II poison-induced cytotoxicity and have important implications for targeted drug design.
Although topoisomerase II poisons can bind and inhibit both hTOPOIIα and hTOPOIIβ, hTOPOIIα is the cellular target of poisons in most cancers because it is expressed in proliferating cells (Pommier et al. 2010). However, recent evidence suggests that side effects associated with these treatments are caused by inhibition of hTOPOIIβ in differentiated cells (Chen et al. 2013). For example, antineoplastic treatment regimens that contain the epipodophyllotoxins (e.g. etoposide or teniposide) are hypothesized to increase the risk of developing secondary malignancies caused by hTOPOIIβ-dependent 11q23 translocations (Felix et al. 2006;Cowell et al. 2012;Azarova et al. 2007;Ratain et al. 1987 figure 11), which could affect drug response. However, the extent to which these variants impact responses to topoisomerase II poisons is unknown, so functional validation is required. The approach of editing human cells and following allele frequencies via sequencing represents a scalable method to assess the functional role of these variants and avoids single-cell cloning.
Importantly, differences in responses to topoisomerase II poisons might not be affected by variation in the topoisomerase II isoforms but instead mediated by variation in cellular import, metabolism, or export. Pharmacogenomic data available for many antineoplastic compounds (Yang et al. 2009;Sim et al. 2011), in combination with human variation data (Lek et al. 2016), can be used to prioritize and test variants in highly conserved regions of proteins known to be involved in these alternative processes. This biased approach focused on candidate variants is necessitated by the lack of power in clinical GWA studies and is not guaranteed to successfully connect variants to differences in drug response. For this reason, unbiased mapping approaches in model organisms combined with functional validation in genome-edited human cells will greatly expand our current understanding of how human genetic variation affects drug responses.

Strains
Animals were cultured at 20ºC with the bacterial strain OP50 on modified nematode growth medium (NGM), containing 1% agar and 0.7% agarose to prevent burrowing of the wild isolates. For each assay, strains were grown at least five generations with no strain entering starvation or encountering dauer-inducing conditions ). Wild C. elegans isolates used for genome-wide association are described previously (Cook, Zdraljevic, Tanny, et al. 2016;Cook, Zdraljevic, Roberts, et al. 2016). Recombinant inbred advanced intercross lines (RIAILs) used for linkage mapping were constructed previously (Andersen et al. 2015). Strains constructed for this manuscript are listed in Supplemental Information. Construction of individual strains is detailed in the corresponding sections below.

High-throughput fitness assay
We used a modified version (Supplemental figure 1) of the high-throughput fitness assay (HTA) described previously (Andersen et al. 2015). In short, strains are passaged for four generations to reduce transgenerational effects from starvation or other stresses. Strains are then bleach-synchronized and aliquoted to 96-well microtiter plates at approximately one embryo per microliter in K medium (Boyd et al. 2012

Calculation of fitness traits for genetic mappings
Phenotype data generated using the BIOSORT were processed using the R package easysorter, which was specifically developed for processing this type of data set (Shimko & Andersen 2014). Briefly, the function read_data, reads in raw phenotype data, runs a support vector machine to identify and eliminate bubbles. Next, the remove_contamination function eliminates any wells that were contaminated prior to scoring population parameters for further analysis. Contamination is assessed by visual inspection. The sumplate function is then used to generate summary statistics of the measured parameters for each animal in each well. These summary statistics include the 10th, 25th, 50th, 75th, and 90th quantiles for TOF. Measured brood sizes are normalized by the number of animals that were originally sorted into the well.
After summary statistics for each well are calculated, the regress(assay=TRUE) function in the easysorter package is used to fit a linear model with the formula (phenotype ~ assay) to account for any differences between assays. Next, outliers are eliminated using the bamf_prune function. This function eliminates strain values that are greater than two times the IQR plus the 75th quantile or two times the IQR minus the 25th quantile, unless at least 5% of the strains lie outside this range. Finally, drug-specific effects are calculated using the regress(assay=FALSE) function from easysorter, which fits a linear model with the formula (phenotype ~ control phenotype) to account for any differences in population parameters present in control DMSOonly conditions.

Linkage Mapping
A total of 265 RIAILs were phenotyped in the HTA described previously for control and etoposide conditions. The phenotype data and genotype data were entered into R and scaled to have a mean of zero and a variance of one for linkage analysis (Supplemental table 5).
Quantitative trait loci (QTL) were detected by calculating logarithm of odds (LOD) scores for each marker and each trait as where r is the Pearson correlation coefficient between RIAIL genotypes at the marker and phenotype trait values (Bloom et al. 2013). The maximum LOD score for each chromosome for each trait was retained from three iterations of linkage mappings (Supplemental table 6). We randomly permuted the phenotype values of each RIAIL while maintaining correlation structure among phenotypes 1000 times to estimate significance empirically. The ratio of expected peaks to observed peaks was calculated to determine the genome-wide error rate of 5% of LOD 4.61. Broad-sense heritability was calculated as the fraction of phenotypic variance explained by strain from fit of a linear mixedmodel of repeat phenotypic measures of the parents and RIAILs (Brem & Kruglyak 2005). The total variance explained by each QTL was divided by the broad-sense heritability to determine how much of the heritability is explained by each QTL. Confidence intervals were defined as the regions contained within a 1.5 LOD drop from the maximum LOD score.

Genome-wide association mapping
Genome-wide association (GWA) mapping was performed using 152 C. elegans isotypes (Supplemental table 7). We used the cegwas R package for association mapping (Cook, Zdraljevic, Roberts, et al. 2016). This package uses the EMMA algorithm for performing association mapping and correcting for population structure (Kang et al. 2008), which is implemented by the GWAS function in the rrBLUP package (Endelman 2011). The kinship matrix used for association mapping was generated using a whole-genome high-quality singlenucleotide variant (SNV) set (Cook, Zdraljevic, Tanny, et al. 2016) and the A.mat function from the rrBLUP package. SNVs previously identified using RAD-seq (Andersen et al. 2012

Fine mapping
Fine mapping was performed on variants from the whole-genome high-quality SNV set within a defined region of interest for all mappings that contained a significant QTL. Regions of interest surrounding a significant association were determined by simulating a QTL with 20% variance explained at every RAD-seq SNV present in 5% of the phenotyped population. We then identified the most correlated SNV for each mapping. Next, we determined the number of SNVs away from the simulated QTL SNV position that captured 95% of the most correlated SNVs. A range of 50 SNVs upstream or downstream of the peak marker captured 95% of the most significant SNVs in the simulated mappings. We therefore used a region 50 SNVs from the last SNV above the Bonferroni-corrected p-value on the left side of the peak marker and 50 SNVs from the last SNV above the Bonferroni-corrected p-value on the right side of the peak marker.
The snpeff function from the cegwas package was used to identify SNVs from the wholegenome SNV set with high to moderate predicted functional effects present in a given region of interest (Cingolani et al. 2012). The correlation between each variant in the region of interest and the kinship-corrected phenotype used in the GWA mapping was calculated using the variant_correlation function and processed using the process_correlations function in the cegwas package (Supplemental table 10). ClustalX was used to perform the multiple sequence alignment between various topoisomerase II orthologs (Supplemental file 1).

Near-isogenic line generation
NILs were generated by crossing N2xCB4856 RIAILs to each parental genotype. For each NIL, eight crosses were performed followed by six generations of selfing to homozygose the genome. Reagents used to generate NILs are detailed in Supplemental Information. The NILs responses to 250 µM etoposide were quantified using the HTA fitness assay described above (Supplemental table 11).

Dominance tests
Dominance experiments were performed using the fluorescent reporter strain EG7952 The phenotypes of the progeny were scored using the BIOSORT as described above (Supplemental table 12). Heterozygous progeny were computationally identified as those individuals that had fluorescence levels between the non-fluorescent and fluorescent parental strains.
To generate injection mixes, the tracrRNA and crRNAs were incubated at 95ºC for 5 minutes and 10ºC for 10 minutes. Next, Cas9 protein was added and incubated for 5 minutes at room temperature. Finally, repair templates and nuclease-free water were added to the mixtures and loaded into pulled injection needles (1B100F-4, World Precision Instruments, Sarasota, FL).
Individual injected P 0 animals were transferred to new 6 cm NGM plates approximately 18 hours after injections. Individual F 1 rollers were then transferred to new 6 cm plates and allowed to generate progeny. The region surrounding the desired Q797M (or M797Q) edit was then amplified from F 1 rollers using oECA1087 and oECA1124. The PCR products were digested using the HpyCH4III restriction enzyme (R0618L, New England Biolabs, Ipswich, MA).
Differential band patterns signified successfully edited strains because the N2 Q797, which is encoded by the CAG codon, creates an additional HpyCH4III cut site. Non-Dpy, non-Rol progeny from homozygous edited F 1 animals were propagated. If no homozygous edits were obtained, heterozygous F 1 progeny were propagated and screened for presence of the homozygous edits. F 1 and F 2 progeny were then Sanger sequenced to verify the presence of the proper edit. Allele swap strains responses to the topoisomerase II poisons were quantified using the HTA fitness ass described above.

Molecular docking simulations
The C. elegans TOP-2 three-dimensional structure homology model was built by threading the C. elegans TOP-2 peptide to the human topoisomerase II beta structure (PDB accession code 3QX3; 59% identity, 77% similarity) using the Prime3.1 module implemented in Schrodinger software (Jacobson et al. 2002;Jacobson et al. 2004). After building the model, a robust energy minimization was carried out in the Optimized Potentials for Liquid Simulations (OPLS) force field. The minimized structure was subjected to MolProbity analysis, and the MolProbity score suggested with greater than 95% confidence that the minimized structure model was a good high-resolution structure (Davis et al. 2007).
Next, the Prot-Prep wizard was used to prepare the TOP-2 homology model, which fixed the hydrogen in the hydrogen bond orientations, eliminated the irrelevant torsions, fixed the missing atoms, assigned the appropriate force field charges to the atoms (Sastry et al. 2013).
After preparing the structure, the glutamine 797 was mutated to various rotamers of methionine (Q797M), which subsequently underwent minimization in the OPLS force field. The energy-   Divergence, as measured by Tajima's D, is shown across the etoposide QTL confidence interval (II:11021073-12008179). The whole-genome SNV data set (Cook, Zdraljevic, Tanny, et al. 2016;Cook, Zdraljevic, Roberts, et al. 2016)   ).   Table S1: CRISPR allele-swap strains phenotype data: Processed phenotype data of CRISPR allele-swap strains in the presence of etoposide.              were crossed to each other to generate a top-2(ok1930)/mIn1 strain in the CB4856 genetic background and named ECA338.

Generation of top-2 allele replacement strains
All allele replacement strains were generated using CRISPR/Cas9-mediated genome engineering, using the co-CRISPR approach described in the main text methods (Kim et al. 2014) with Cas9 ribonucleoprotein delivery (Paix et al. 2015