Cross-feeding affects the target of resistance evolution to an antifungal drug

Pathogenic fungi are a cause of growing concern. Developing an efficient and safe antifungal is challenging because of the similar biological properties of fungal and host cells. Consequently, there is an urgent need to better understand the mechanisms underlying antifungal resistance to prolong the efficacy of current molecules. A major step in this direction would be to be able to predict or even prevent the acquisition of resistance. We leverage the power of experimental evolution to quantify the diversity of paths to resistance to the antifungal 5-fluorocytosine (5-FC), commercially known as flucytosine. We generated hundreds of independent 5-FC resistant mutants derived from two genetic backgrounds from wild isolates of Saccharomyces cerevisiae. Through automated pin-spotting, whole-genome and amplicon sequencing, we identified the most likely causes of resistance for most strains. Approximately a third of all resistant mutants evolved resistance through a pleiotropic drug response, a potentially novel mechanism in response to 5-FC, marked by cross-resistance to fluconazole. These cross-resistant mutants are characterized by a loss of respiration and a strong tradeoff in drug-free media. For the majority of the remaining two thirds, resistance was acquired through loss-of-function mutations in FUR1, which encodes an important enzyme in the metabolism of 5-FC. We describe conditions in which mutations affecting this particular step of the metabolic pathway are favored over known resistance mutations affecting a step upstream, such as the well-known target cytosine deaminase encoded by FCY1. This observation suggests that ecological interactions may dictate the identity of resistance hotspots.


Introduction
The past decades have seen great advances in the medical field, with the death rate from complications of medical and surgical care steadily declining [1].Numbers have also gone down for some of the leading causes of death globally, notably deaths from HIV, which have decreased by 51% between 2000 and 2019 [2].Unfortunately, some of the treatments which help increase life expectancy also lower immune defenses, which puts patients at risk of potentially life-threatening infections caused by opportunistic pathogens [3][4][5].Among them, Candida auris, several Candida species and Aspergillus fumigatus, all drug-resistant fungal pathogens, feature on the CDC's list of antimicrobial resistance threats [6].Among other factors, the biological similarities between fungal and host cells make it difficult to design efficient antifungals with low toxicity [7].Consequently, there are only five main classes of antifungals [7].One may wonder what makes a 'good' antifungal target.
The role of most antifungals is to prevent proper synthesis of the membrane or cell wall, because they are made of components not found in mammalian cells (ergosterol, chitin-glucan, for example).If we were to look for other molecules, should the antifungal target an essential metabolic end product?An enzyme in a metabolic pathway?If the enzyme catalyzes an early step of a specific pathway, resistance could potentially arise through loss of function downstream, for instance when a toxic molecule is produced as a byproduct or through the conversion of a prodrug.Among the considerations for the development of powerful drugs targeting cell metabolism is the organization of cell communities in vivo.For instance, treatment can have drastically different outcomes depending on the microbe lifestyle, fungi in biofilms being notoriously less susceptible to antifungals [8].
Interestingly, despite the urgent need for better antifungals, the mechanisms of resistance are still poorly understood outside of the direct drug targets [3].Particularly in the medical field, it is common practice to sequence as few as a single gene to look for 'common' resistance-conferring mutations, which introduces a bias in databases such as MARDy [9].In doing so, resistance is too often studied through a keyhole.Even more challenging is deciphering if the mechanisms of resistance to a specific drug are general, or if they depend on the genetic background or the environmental conditions in which they take place.Here, we leveraged the power of experimental evolution coupled with whole-genome sequencing (WGS) to capture a large number of mutations conferring resistance to 5-fluorocytosine (5-FC, also known as flucytosine) using the model yeast Saccharomyces cerevisiae.We focused on 5-FC because it is among the oldest antifungals, it is listed by the WHO as an essential medicine [10], and its metabolism is documented enough that we can interpret this data with respect to the adaptive mechanisms and rationalize the paths to resistance [11].
5-FC is a prodrug that is not toxic to fungi before it is metabolized [12].Its metabolism involves several steps, which ultimately lead to cell toxicity (Fig 1).First, it is imported into the   are represented as blue, dark purple, light purple and light blue beads, respectively.For clarity, only two permeases known to import pyrimidines are shown: Fcy2 and Fur4.In prototrophic yeast strains, pyrimidines can be synthesized de novo from amino acids such as glutamine and aspartate [29].The steps leading to the obtention of the nucleotide precursors uridine monophosphate (UMP) and uridine diphosphate (UDP) are catalyzed by Ura3 and Ura6, respectively, in the de novo pathway [30,31].Alternatively, pyrimidines can be imported from the medium [32].Uracil can also be obtained by deamination of cytosine by Fcy1 [33].Finally, UMP can be obtained by the addition of a phosphate group by Fur1 [34].5-FC is a prodrug, which is metabolized the same way cytosine is, except its deamination leads to the obtention of the cytotoxic compound 5-FU [35].Based on work done in laboratory strains or other fungal species, resistance to 5-FC can arise when either of the following three steps are compromised: import, conversion and activation, typically through mutations inactivating Fcy2 (for example), Fcy1 and Fur1, respectively [36].We also hypothesize an alternative path to resistance in the pleiotropic drug response, wherein loss of mitochondrial function leads to derepression of the transcription factor Pdr3, which in turn overexpresses efflux pumps such as Pdr5. Figure created with BioRender, using the structures of Fcy1 (1P6O), Fur1 from C. albicans (7RH8), Ura3 (3GDL), Ura6 (1UKZ) and Pdr5 (7P04) and AlphaFold predictions of Fcy2 (P17064) and Fur4 (P05316) [37,38].https://doi.org/10.1371/journal.pgen.1011002.g001cell by several permeases, typically those involved in the import of exogenous pyrimidines (Fcy2, Fur4, Fcy21, Fcy22) [13].Next, 5-FC is deaminated by the cytosine deaminase Fcy1 into the cytotoxic compound 5-fluorouracil (5-FU).5-FU is a molecule known outside of the field of antifungals that has been used to treat cancers for several decades [14].Its main activation relies on the addition of a phosphate group by the uracil phosphoribosyltransferase Fur1.Incorporation of fluoronucleotides into RNA and DNA, as well as inhibition of the thymidylate synthase either cause growth arrest or cell death [14][15][16].Resistance to 5-FC is typically conferred by mutations in the genes involved in its metabolism, preventing its import (FCY2), its conversion into 5-FU (FCY1) or the activation of 5-FU (FUR1) (Fig 1).FCY genes were named after their phenotypes associated with FluoroCYtosine resistance.Mutations in FCY1 have regularly been reported as being associated with 5-FC resistance in clinical isolates of a broad range of pathogens such as Candida albicans, Candida dubliniensis, Nakaseomyces (Candida) glabrata and Clavispora (Candida) lusitaniae [17][18][19][20].
Few alternative paths to 5-FC resistance have been shown, with some studies pointing at a putative pleiotropic drug response resulting in efflux-mediated resistance (Fig 1) [21][22][23][24][25].This mechanism is far better characterized for azoles.Azole treatment typically favors the formation of rho 0 mutants (also known as petite mutants), which are characterized by their loss of mitochondrial DNA.This results in an alleviation of repression of the transcription factors (Pdr1, Pdr3) regulating the expression of efflux pumps [26].Some of these efflux pumps, for example Pdr5, end up overexpressed, conferring cross-resistance to several antifungals [26].
Because of the tight relationship with pyrimidine metabolism, the use of auxotrophic laboratory yeast strains to study the mechanisms of resistance may paint a distorted picture of how resistance to 5-FC and 5-FU comes to evolve outside the lab.Notably, it has been shown that the presence of uracil in the growth medium triggers a negative feedback by repressing the expression of the uracil permease Fur4 [27], potentially limiting cell entry by 5-FC.Not only does the use of specific lab strains narrow the number of detectable paths to 5-FC resistance, it also prevents the detection of potential epistatic effects.Indeed, the impact of the genetic background on the evolution of resistance is still poorly understood and has only recently become one of many interesting avenues of research in the field [28].Finally, searching a relatively small gene space for resistance-conferring mutations often limits interpretations.With the decreasing costs of WGS, we can now hope to reach saturation, in that sequencing a larger number of mutants would not substantially increase the number of relevant mutations.
For the aforementioned reasons, we set up to generate hundreds of 5-FC resistant mutants, subjecting two wild prototrophic strains of Saccharomyces cerevisiae to experimental evolution in minimal medium.Their relative fitness was estimated in several conditions to test for respiration capacity, fitness cost and cross-resistance.We show that about a third of all mutants developed generalized resistance through loss of mitochondrial function.The majority of the remaining two thirds evolved resistance to 5-FC and 5-FU.To explain their phenotype, the genomes of 276 of these mutants were analyzed by WGS.In particular, we sought out to determine if the resistance phenotype conferred by a given mutation was dependent on the genetic background, or if there were genetic interactions for a given background.For both backgrounds, virtually all sequenced mutants evolved resistance through loss-of-function mutations in the gene FUR1.Interestingly, none of the sequenced genomes showed any mutation in the well-known resistance gene FCY1, which we show is the result of negative selection imposed by the 5-FU-producing FUR1 mutants.Ultimately, sequencing a large number of genomes also led to the identification of rare alleles most likely sufficient to confer resistance to 5-FC.

Mutants evolved in 5-FC develop cross-resistance to 5-FU
Two environmental strains of S. cerevisiae, LL13-040 and NC-02 were subjected to parallel experimental evolution, which generated 682 5-FC resistant mutants.All candidate resistant strains were streaked and spotted on YPG agar medium to evaluate mitochondrial function.rho -mutants amounted to 83/296 (28%) and 140/386 (36%) for LL13-040 and NC-02, respectively.The fitness of 408 5-FC resistant mutants was estimated by measuring colony growth through time on solid media.
For both backgrounds, most mutants show an expected increase in relative fitness (relative to the parental strain) in minimal medium supplemented with 5-FC compared to control conditions (rich or minimal medium) (Fig 2).rho -mutants show a lower relative fitness in drugfree media, with LL13-040 rho -mutants displaying a larger fitness defect than NC-02 counterparts.For both backgrounds, the relative fitness in minimal medium, the same medium in which the strains were evolved, is generally lower or equivalent to that of the parental strains.Additionally, when we performed the evolution experiment in the absence of 5-FC, no resistant mutant was generated.Both observations suggest that, if mutants evolved to grow better in minimal medium, none of the evolutionary routes coincidentally conferred resistance to the antifungal.We also measured the growth on minimal medium supplemented with 5-FU, the converted form of 5-FC obtained by Fcy1-mediated deamination in the pyrimidine salvage pathway (Fig 1).rho -and rho + mutants show a similar relative fitness, which is about twice that of the parental strains in minimal medium supplemented with either 5-FC or 5-FU.Interestingly, the relative fitness in 5-FC significantly correlates with the one in 5-FU, hinting at a mechanism of resistance that would take place downstream of the conversion of 5-FC into 5-FU or that affects intake of these molecules.We hypothesized this mechanism to be either the pleiotropic drug response in the case of rho -mutants, or the inactivation of FUR1 in the case of rho + mutants.

rho -mutants are cross-resistant to fluconazole
To test if loss of respiration induces a pleiotropic drug response, we first evaluated the fitness of rho -mutants in rich medium, as well as in minimal medium, supplemented or not with several antifungals belonging to distinct classes.We tested growth in the presence of three other classes of clinical antifungals: echinocandins (micafungin and caspofungin), polyene (nystatin) and azole (fluconazole).Because of the slow growth of rho -mutants in some antifungal conditions, we incubated the plates at 37˚C.In these conditions, rho -mutants display a higher relative fitness in 5-FC than rho + mutants, while still showing a fitness defect in control media (Fig 3A).Interestingly, rho -mutants of both backgrounds show cross-resistance to fluconazole, with a relative growth more than twice that of the parental strain for NC-02 (Fig 3A).Also noteworthy is their fitness in micafungin, although this is likely the result of a very low growth rate (S1 Fig).
Since the loss of mitochondrial function in rho -mutants has been shown to lead to generalized resistance, we hypothesized that cross-resistance to 5-FC and fluconazole results from an increased expression or activity of efflux pumps.To test this hypothesis, we performed a rhodamine retention assay.Rhodamine is a known substrate of ABC transporters and its accumulation in cells has been shown to be inversely correlated with efflux capacity [39].We find that virtually all rho -mutants display a lower count of rhodamine-fluorescent cells compared to their parental strain, confirming they have gained an increased efflux capacity (Fig 3B).Notably, both parental strains accumulate much less rhodamine than a laboratory strain deleted for all ABC transporters, meaning that parental strains were already capable of rhodamine efflux in the conditions of the experiment.Even though the NC-02 strain appears to have a diminished efflux capacity compared to the LL13-040 strain, the rho -mutants of both backgrounds display comparable rhodamine accumulation levels (Fig 3B).Therefore, the net gain of efflux capacity for NC-02 mutants is higher than that of LL13-040 counterparts.This efflux capacity is correlated with the measured growth in fluconazole, however the correlation is statistically significant only for NC-02 (Fig 3C).For both backgrounds, the correlation with the measured growth in 5-FC is not significant (Fig 3D).This could be explained by a lack of statistical power, but in all likelihood, a complex regulation drives the expression of multiple transporters involved in resistance to fluconazole, as well as 5-FC.Another potential mechanism is that resistance to 5-FC could be binary, whereby enough efflux above a threshold would confer resistance, whereas it would be gradual for fluconazole, creating a correlation between efflux and growth.

rho + mutants display a large number of distinct loss-of-function mutations in FUR1
Next, we focused on the mechanisms of resistance at play in rho + mutants.To make sure we identified resistance conferring mutations, including potentially rare ones, we sequenced the genomes of most assayed rho + mutants (149/160 LL13-040 mutants and 127/134 NC-02 mutants).Overall average sequencing coverage was around 60X, with no apparent CNVs (S2 Fig) .Two computational methods were carried out to detect indels and SNPs, one with stringent criteria ("samtools"), and one with more tolerant criteria ("gatk").Because of the frequency of mutations observed in FUR1 in our preliminary analysis, we also directly amplified and Sanger sequenced FUR1 (including the promoter region) from the same genomic DNA for most of these mutants.We included WT controls as well as fcy1Δ and fcy2Δ mutants for all three analyses (samtools, gatk and Sanger).
Aggregating results from samtools, gatk and Sanger sequencing, we identified 118 distinct SNPs or indels in FUR1 (S2 Data).The use of two computational methods to call variants proved relevant, since part of the variants were detected by only one method (Fig 4A).As expected, gatk, with its set of more permissive criteria, detected variants in around 500 genes in each background, whereas samtools detected variants in only 150 genes on average (Fig 4A).Variants in other genes than FUR1 were rarely detected by both computational methods, which might be linked to the quality of their call (Fig 4A).Overall, approximately 120 genes per individual genome were found to have variants by either method (Fig 4B).This large number is a result of the filtering criteria, which were set to be more flexible to make sure rare mutations would not be discarded.Ultimately, the number of unique variants found in a given gene appeared as a good indicator to sort through the noise.Specifically, the number of unique variants found in FUR1 was much larger than in any other gene, which is strong evidence that FUR1 constitutes a hotspot for 5-FC resistance-conferring mutations (Fig 4A ).
Out of all strains for which FUR1 was Sanger sequenced (n = 268), all six controls (parental strains, fcy1Δ and fcy2Δ mutants) and only 13 mutants carried the WT allele of FUR1.Through a detailed look into our dataset, we found three candidate target genes which could alone explain the resistance phenotype of these 13 strains carrying a WT FUR1 sequence: URA6, GFA1 and ARG2.Two LL13-040 strains and one NC-02 strain carried a background-specific mutation in URA6: R52S for the former, G73S for the latter.Eight NC-02 strains carried one of four mutations in GFA1: V478F, G482R, G492S and N496K.One NC-02 strain carried a K113* mutation in ARG2.Interestingly, the corresponding mutants display a relative fitness distinct to that of FUR1 mutants (Fig 4C and S3 Data).Notably, the mutations detected in URA6 and GFA1 are harbored by strains resistant to 5-FU, whereas the ARG2 mutant is only mildly resistant to 5-FC in our experimental conditions (Fig 4C).All identified mutations in these three genes (n = 7) are predicted by mutfunc (see methods section) to impact conserved residues with five out of seven also predicted to impact protein stability.Their effect is therefore likely to be through loss of function.
Among the sequenced strains for which no clear candidate resistance mutation was identified, one appears to have been initially miscategorized as resistant to 5-FC, four display the same relative fitness in 5-FC and 5-FU as FUR1 mutants (and are therefore likely FUR1 mutants as well) and one appears to be mildly resistant to 5-FC but not 5-FU (Fig 4C).Finally, most unsequenced rho + mutants showed little to no resistance, which suggests they were false positives of the evolution experiment (Fig 4C).
To further investigate the extent and potential mechanisms of resistance, 35 resistant isolates (across both backgrounds) for which mutations were predicted in FUR1 were tested in a growth assay, alongside the parental strains, and the deletion mutants for FCY1, FCY2 and FUR1 (s_004, s_005, s_006, s_009, s_010, s_011 in S1 Data).All tested mutants showed a relative fitness similar to their null mutant counterpart (Fig 4D).This suggests most mutations inactivate the function of FUR1 and are sufficient to explain the resistance phenotypes, meaning that mutations in other genes likely play no role (Fig 4B).Importantly, the relative fitness of the fcy1Δ mutants was comparable to that of all tested FUR1 mutants.The reason why inactivating mutations in FCY1 were not picked up in our experiment is thus not because they provide lower resistance levels than FUR1 inactivating mutations in these conditions (see below).Further examination confirmed that it is shared among 39 strains which arose from the same preculture (S3 Fig) .This could be explained by a rapid fixation of the mutation that was segregating in the preculture.It is important to note however that it constitutes a rare case, since we often selected less than 20 mutants per preculture, which generally carried different mutations each (S3 Fig).
To understand why all validation mutants behaved like the null mutant, we compared their predicted impact to that of all possible substitutions.Out of 4,104 possible missense mutations in Fur1, 1,279 are predicted to impact stability.Our dataset captured a total of 76 missense mutations, 42 of which are predicted to impact stability.This is significantly more than what can be expected by chance (Fisher's exact test p-value = 1.39e-5,S4A Fig) .The same observation can be made of mutations predicted to impact function based on conservation, with 67 out of 2,667 being captured in our dataset (Fisher's exact test p-value = 4.96e-6, S4B Fig) .The results are consistent with loss-of-function mutations.We find that the substitutions in our dataset are predicted to destabilize Fur1 with generally high positive ΔΔG values, with no apparent bias in terms of position along the sequence (S4A and S5 Figs).Expression of FUR1 from a centromeric plasmid under its native promoter restored sensitivity in most tested mutants (n = 32), including all with a mutation introducing a stop codon, confirming the detected mutations lead to loss of function (Fig 5B).Interestingly, some mutations could only be partially complemented, with four strains growing at more than 75% the rate of their corresponding control (vector with no expressed ORF, Fig 5B).To better understand this dominant negative effect, all residues for which we detected a mutation in our dataset were mapped on the predicted structure of the Fur1 tetramer for which the C. albicans structure is available (PDB: 7RH8).We find that most mutated residues are not located close to the active site, but instead appear to be involved in structurally important folds and helices (Fig 5C).Additionally, some mutated residues reside at the interface between two chains, which could indicate that the mutation prevents proper assembly of the tetramer (Fig 5C).Namely, G104 and G205 appear to interact each with a cysteine residue from the adjacent chain (Fig 5C insets).This observation could explain why the mutations G104D and G205W cannot be complemented with the WT allele of FUR1 as they could also perturb the assembly of tetramers that would include WT chains.

FUR1 path to resistance prevents selection of FCY1 mutations
One of the known resistance genes to 5-FC encodes for the cytosine deaminase Fcy1, which acts upstream of Fur1 in the pyrimidine salvage pathway (Fig 1).We included fcy1Δ mutants in our validation experiments as positive controls.However, we were surprised to find no single Fcy1 mutants in our set of sequenced genomes.As loss of function in FCY1 is sufficient to lead to resistance and many mutations in the gene are destabilizing enough to cause resistance [41], one would expect FCY1 mutations to also be frequent in our experiments.
To validate the absence of detectable mutations in FCY1, we randomly selected mutants (n = 14) and performed Sanger sequencing: all carried the wild-type sequence.Next, we confirmed that FCY1 is well expressed in LL13-040, using a mEGFP protein fusion (S6A Fig) .Finally, we evaluated the functionality of Fcy1 in parental strains.A growth assay using cytosine as the sole nitrogen source in minimal medium confirmed that both strains can grow, which would not be possible without the deamination of cytosine performed by Fcy1 (S6B Fig).Relative growth corresponds to the area under the curve calculated on 25 h for the mutant with pMoBY-FUR1 divided by the same parameter for the mutant with pMoBY.Mutants were grown from single colonies in SD + 100 μg/mL 5-FC (+ G418 to maintain pMoBY).Colored dots indicate the same mutation was detected in the mutants from both backgrounds.Two mutations for which expression of the WT allele of FUR1 did not restore 5-FC sensitivity are highlighted.A closer look at the residues in the next panel suggests a putative role in assembly of the tetramer.C) Location of substitutions from our dataset on the predicted structure of Fur1, visualized using UCSF ChimeraX [40].The tetramer (C.albicans Fur1 assembly 7RH8) is represented in light gray, with 1 chain in dark gray.Residues found to be mutated at least once in our dataset are colored in pink.UTP molecules are represented in sticks colored by atom.Two insets zoom in on one interface each (the slightly opaque one depicts the interface at the back of the represented structure).The highlighted residues (each one from a different monomer) appear to interact with one another to ensure proper assembly of the tetramer.https://doi.org/10.1371/journal.pgen.1011002.g005 Next, we sought out to confirm the fitness of the fcy1Δ mutants in liquid medium compared to our 5-FC mutants as a lower fitness could explain why they were not picked up in our experiments.We performed growth assays to calculate the relative fitness in three conditions with minimal medium: with or without 5-FC and one condition where the only nitrogen source is cytosine.As expected, the fcy1Δ mutants fared well compared to our evolved mutants in 5-FC but displayed the lowest relative fitness among rho + mutants in cytosine (S7A Fig) .In comparison, virtually all rho + mutants have little to no fitness tradeoff in cytosine, indicating that FCY1 is functional.
Interestingly, when we compared the relative fitness measured in liquid medium versus the one measured on solid medium, we observed that the fcy1Δ mutant shows a significant fitness defect on solid medium when grown alongside other mutants (S7B Fig) .A recent study by our group showed that fcy1Δ could grow on plates supplemented with cytosine as the only nitrogen source as long as strains with a functional FCY1 allele (FCY1 + ) were growing in proximity [41].The auxotrophy of a fcy1Δ strain can therefore be rescued by cross-feeding uracil from FCY1 + strains.We hypothesized that this could also occur for 5-FU and, through interactions between cells of different genotypes and diffusion in the medium, it would lead to strains with a loss of function of FUR1 (fur1 -strains) taking over strains with a loss of function of FCY1 (fcy1 -strains) during experimental evolution.
Specifically, if a mutant is sensitive to a compound effluxed by its neighbor, its growth will be inhibited.The ultimate toxic compound produced from the prodrug is 5-FUMP, downstream of Fcy1 and Fur1.This means that in a complex community, a FCY1 + strain could produce 5-FU, which would inhibit its growth and that of fcy1 -neighboring cells too, provided they are FUR1 + (Fig 6A).On the other hand, a fur1 -strain would be resistant to its own production of 5-FU and to that of neighboring FCY1 + cells (Fig 6A).During experimental evolution, both fcy1 -and fur1 -mutants could arise but because they do so at low frequency at first, only the latter would increase in frequency.Pure cultures of fcy1 -and fur1 -would however be both equally resistant to 5-FC, as our previous results revealed.If this model is correct, conditioning the medium with a fcy1Δ strain should allow growth of a fur1Δ strain, whereas the opposite would lead to growth inhibition.We confirmed this prediction (Fig 6B).For both backgrounds, conditioning the medium with the fcy1Δ mutant led to the fur1Δ mutant growing significantly better than the WT.On the other hand, conditioning the medium with the fur1Δ mutant led to growth inhibition of both the WT and the fcy1Δ mutant.
In this assay we show the ultimate outcome of applying selective pressure to both mutants, but what happens if the only source of 5-FC comes from the solid medium on which the population is plated?To evaluate such a scenario, we designed a competition experiment, in which both mutants are grown individually with no selective pressure.For each background, the fcy1Δ is pooled with the fur1Δ culture.Serial dilutions are then plated on large Petri dishes with 5-FC, and mutants are identified by multiplexed colony PCR.We show that the survival of fcy1Δ mutants significantly depends on the inoculum concentration (Fig 6C).Specifically, if a 1:1 mix of fcy1Δ and fur1Δ mutants is plated at 10 −5 OD 600 , the same ratio will be observed for the growing colonies.However, when the same mix is plated at a concentration 100 times higher, all tested colonies are fur1Δ, meaning fcy1Δ cells are killed by 5-FU cross-feeding.To further confirm the link between cell density and cross-feeding, we performed an inhibition assay, where a culture of the fur1Δ mutant is spotted on a fcy1Δ lawn spread on medium with 5-FC.For both backgrounds, we were able to see a clear inhibition zone around the fur1Δ spot (Fig 6D ).
In summary, the design of our experimental evolution, where the mutants are selected by spotting precultures on a plate, negatively selects against spontaneous FCY1 mutants.This would likely occur as well in a mixed community.FCY1-mediated resistance is context-dependent.A) Predicted cross-feeding interactions between cells with different alleles of FCY1 and FUR1.Uracil (top) and 5-FU (middle) are expected to diffuse in the medium and be converted by Fur1 of neighboring cells, which would allow or inhibit growth, respectively.Notably, the growth of any FUR1 + cells, regardless of their FCY1 allele, will be inhibited because of 5-FU cross-feeding.On the contrary, any fur1 -cells would grow despite 5-FU cross-feeding (bottom).Figure created with BioRender, using the structures of Fcy1 (1P6O) and Fur1 from C. albicans (7RH8).B) Medium conditioning assay.First, cultures of the deletion mutants were prepared in SD with or without 1.56 μg/mL 5-FC.After reaching an OD 600 of 0.6, the cultures were filtered and the medium was transferred to a 96-well plate.Cultures were inoculated with either the WT or the other deletion mutant.Relative growth corresponds to the area under the curve (AUC) normalized by the WT and was obtained from biological triplicates.A two-way ANOVA followed by Tukey's multiple comparison test was performed to compare the

Efflux-mediated resistance to 5-FC in rho -mutants
We generated and analyzed the fitness of hundreds of independent 5-FC resistant mutants to identify the most frequent molecular signatures of resistance to this antifungal.Challenging a common assumption of the usual paths to 5-FC resistance, we show that around a third of mutants gained resistance through a previously unknown mechanism likely involving drug efflux rather than inactivation of key metabolic enzymes.This path to resistance, often referred to as a pleiotropic drug response, is well characterized for azole resistance [42][43][44].Mutants which have lost mitochondrial function are no longer able to respirate and display a specific petite morphotype on glucose-containing medium.Resistance in this case therefore comes with a tremendous tradeoff.In turn, they display enhanced efflux activity, which confers generalized resistance.To date, this resistance mechanism has not been described for 5-FC.
Here, we confirm that 5-FC treatment at least selects for rho -mutants, which are crossresistant to 5-FC and fluconazole (Fig 3A).However, we find no significant correlation between rhodamine accumulation, indicative of increased efflux capacity through ABC transporters, and fitness in 5-FC (Fig 3D).This corroborates previous observations that 5-FC treatment, while selecting for rho 0 mutants, does not affect the expression of transcription factors such as Pdr1 or the related expression of ABC transporters [22].On the other hand, we do observe a significant correlation between rhodamine accumulation and fitness in fluconazole for NC-02 mutants (Fig 3C).Since azole resistance has also been associated with MFS transporters [45], one possible explanation for the discrepancy would be that a mechanism overlapping both regulatory pathways is sufficient to confer resistance to both classes of antifungals.One study supporting this hypothesis reported that a substitution in the gene MRR1 was sufficient to confer cross-resistance to 5-FC and azoles in clinical isolates of C. lusitaniae [25].This gene encodes a transcriptional regulator, which when mutated upregulates the expression of the multidrug transporter encoding gene MFS7.Another study reported reduced susceptibility to azoles in 5-FC resistant isolates of several pathogenic fungi, including C. albicans [46].
Although future work is needed to uncover the exact mechanisms governing 5-FC resistance through efflux, it is plausible that pathogenic fungi could evolve along this path, in addition to the more commonly described resistance paths in the canonical pathway.Losing mitochondrial function would be unfavored in environments that require respiration, given the strong fitness cost in drug-free conditions.However prolonged treatments could render that evolutionary path more accessible, as suggested by reports of such selection in a clinical isolate of N. glabrata in response to fluconazole [47].

FUR1-inactivating mutations confer 5-FC resistance in rho + mutants
Using WGS, we found that most rho + mutants gained resistance through FUR1 inactivation.Interestingly, a large diversity of mutations was detected in this gene.In contrast, the MARDy database only lists two FUR1 mutations conferring resistance to 5-FC at the time of writing relative growth in SD with 5-FC to the one obtained in SD without 5-FC in each condition.Statistical significance is as follows: ****, adj.pvalue < 0.0001; ns, not significant.C) Competition assay.Individual cultures of the deletion mutants were prepared in SD without selective pressure, then pooled in equal volumes, serially diluted and plated on large Petri dishes containing SD with 1.56 μg/mL 5-FC.For each counted colony, the genotype (fcy1Δ or fur1Δ) was confirmed by PCR.For each background, the relationship between CFU count of either mutant and the inoculum concentration was assessed by a Chi-square test of independence, performed on the cumulative sum of counts obtained from two independent experiments.D) Inhibition assay.For each background, a culture of the fcy1Δ mutant was inoculated in SD 1.7% agar with or without 25 μg/mL 5-FC and poured on top of thin 2% agar plates containing the matching medium.A culture of the fur1Δ mutant was then spotted on top.Uncropped pictures are provided in S8 Fig. (version 1.1DB:1.3WS,http://mardy.dide.ic.ac.uk/).Despite the low number of reported mutations (including the ones reported in other studies but not listed in MARDy [11]), our experiment still recapitulates some of them.Most importantly, they correspond to mutations of clinical relevance, such as F211I and G190D found in clinical isolates of C. auris and N. glabrata, respectively [48,49].Our work shows that the available paths through mutations in FUR1 are far more diverse than the literature lets appear.Most mutations in FUR1 appear to be lossof-function mutations.We observed several indels (deletions and insertions of up to 49 and 39 bp, respectively) and five mutations affecting the translation start site of Fur1 (M1V, M1I, M1L, M1K, M1T) (S2 Data).Many more are predicted to impact the stability of the protein.This would make resistance through mutations in FUR1 very likely since many of the possible amino acid changes are predicted to impact the stability of Fur1 and/or conserved regions (S4 It is also possible that the drug concentration used to select for resistant mutants was high enough that a partially functional Fur1 would not confer resistance.It would be interesting to study the relationship between the extent of Fur1 inactivation and inhibitory concentrations to formally test this hypothesis.In any case, our results strongly suggest that the number of resistance-conferring mutations in fungal pathogens is profoundly underestimated.They do confirm however that resistance can emerge rapidly in response to 5-FC monotherapy [50].That being said, rapid emergence of resistance does not seem to be specific to 5-FC, whether it is observed through experimental evolution or in the clinic.For example, cases have been reported where resistance through nonsense mutations in FUR1 was rapidly acquired upon combination therapy with caspofungin [51].Cross-resistance to fluconazole (and potentially other azoles) has also been reported to evolve rapidly in clinical settings [46].The same can be said of nystatin resistance in experimental evolution [52].Additionally, the fact that resistance to different antifungals is connected through shared resistance mutations despite their apparent lack of similarity in their mode of action is worrisome and stresses the need for more studies on cross-resistance.

Alternative paths to resistance in rho + mutants
Our experiment led to the identification of three other genes potentially conferring resistance to 5-FC when mutated: URA6, GFA1 and ARG2 (Fig 4C).
URA6 encodes the enzyme which catalyzes the step right after the reaction performed by Fur1 (Fig 1), therefore it is highly likely that the detected mutations inactivate URA6 and are sufficient to confer 5-FC and 5-FU resistance.A temperature-sensitive mutant of this enzyme in a laboratory strain is known to be resistant to 5-FU, confirming that loss of function is responsible for resistance [31].GFA1 encodes an important enzyme involved in the biosynthesis of chitin.Some studies have shown fungal cells can display increased chitin levels in response to fluconazole or echinocandins [53][54][55], however it appears to be one of many consequences of a generalized stress response.Both URA6 and GFA1 are essential in auxotrophic laboratory strains [31,56].This shows that epistasis can play an important role in the evolution of resistance, and that the use of prototrophic strains in experimental evolution experiments will likely lead to different outcomes.
Finally, it has previously been shown that deletion of ARG2 can confer resistance to 5-FU, through crosstalks between the ornithine biosynthesis pathway and the pyrimidine biosynthesis pathway [57].The hypothesized mechanism is that blocking ornithine formation hinders the consumption of carbamoyl-phosphate, both being required for the synthesis of arginine.
As a result, carbamoyl-phosphate is used as an early precursor of UTP, which ends up being produced in larger quantities than 5-FUTP [57].This would corroborate a previous observation that five other genes involved in arginine metabolism are linked with 5-FC resistance [23].
Our results highlight the need for studies using prototrophic strains, instead of the popular laboratory strains, as they allow the identification of rare alleles conferring resistance.

FUR1 is the main target to confer 5-FC resistance
Out of all genes involved in the metabolism of 5-FC, we unexpectedly detected mutations only in FUR1 and URA6 (Fig 4C).Resistance to 5-FC in clinically relevant fungi typically results from mutations in FCY2, FCY1 and/or FUR1 [17,58,59].Since resistance often requires loss of function of those genes, redundant entry routes in S. cerevisiae can explain the absence of resistance mutations in genes involved in the import of 5-FC, namely FCY2, FCY21, FCY22 and FUR4 [13] as their loss of function would be masked by other, intact genes.The case of FCY1 was more difficult to interpret as there is no known redundancy for its function.We hypothesized that depending on the context, FUR1 could be a more likely target than FCY1 due to cross-feeding.In a mix of FCY1 + and fcy1 -cells, 5-FU would be produced and secreted, leading to the death of FCY1 + cells, as well as loss-of-function mutants.The benefits of losing the cytosine deaminase function would therefore be highly specific to the ecological context.
Accordingly, we performed a growth-based experiment in which the medium was previously conditioned by the growth of either a fcy1Δ mutant or a fur1Δ mutant (Fig 6B ).A medium conditioned by the latter contains 5-FC, as well as 5-FU, as a result of deamination of 5-FC by Fcy1.5-FU being toxic for any cell that encodes a functional Fur1, it imparts selective pressure upon FUR1 mutants.We further confirm this hypothesis by showing that plating a mix of fcy1Δ and fur1Δ mutants at high cell density prevents the growth of fcy1Δ cells (Fig 6C).Finally, spotting the fur1Δ mutant on top of a fcy1Δ lawn leads to the apparition of an inhibition zone, delimiting the range of diffusion of 5-FU (Fig 6D and S8 Fig).
Compound sharing during the selection step dilutes the populations of cells in which mutations impact any step upstream of the conversion of 5-FU.This is of the utmost ecological importance, since the ecological context will define which mutations are more likely to lead to resistance [60].

No strong background-dependent effect for path to resistance but many background-dependent effects on fitness
Even though we used two distinct environmental isolates as parental strains, we observe strong parallel evolution of resistance.Notably, the main outcome of our evolution experiment did not seem to be affected by the presence of active transposable elements in only one of the two parental strains (LL13-040).Indeed, for both backgrounds, most evolved strains gained resistance through loss-of-function mutations in FUR1.However, differences in fitness reflect the different behaviors of each background before and after evolution.For example, even though both parental strains grow at comparable rates in all tested conditions (S1 Fig) , NC-02 displays more rhodamine accumulation than LL13-040 (Fig 3B).After evolution however, rho -strains of both backgrounds display strikingly similar distributions in rhodamine accumulation (Fig 3B).One background, NC-02, therefore evolved to gain much more efflux capacity than the other.This observation is corroborated by the significant correlation with fitness in fluconazole, which is again markedly higher for NC-02 rho -strains compared to LL13-040 counterparts (Fig 3C).Similarly, the same mutants show a lower fitness tradeoff in drug-free media (Figs 2 and 3A).Together, our results could indicate that NC-02 would potentially evolve along a broader range of paths to resistance.Interestingly, this hypothesis is also supported by the mutations detected in the sequenced genomes.For example, more indels were detected in FUR1 for NC-02 strains (Fig 5A).Two genes potentially sufficient to confer resistance, GFA1 and ARG2, were found to be mutated only in NC-02 strains (Fig 4C).On the other hand, a single mutation in Fur1 (R110G) was found to be present in 39 different LL13-040 strains (S3 Fig) .And despite selection seemingly acting more strongly in one background than the other, we found an identical number of unique mutations in FUR1 for each background (52 background-specific mutations for each background and 14 shared mutations, S2 Data and S5 Fig).

Experimental design and evolutionary relevance
Our experimental design does not represent a real-world clinical situation in which 5-FC would be used for treatment.However, we revealed a very interesting community that can affect the route of evolution.Upon treatment of an infection, rapid establishment of resistance would arguably be impacted by two main factors: the local concentration of antifungal and the local population size.The first factor determines for instance the efficiency of selection and the second the size of the mutational supply and the efficiency of selection.Our results indicate that population size or density could also affect the route of evolution.
Drug concentration is rarely homogeneous, but instead exists as a spatiotemporal gradient linked to the inherent pharmacodynamic properties of the drug, such as imperfect drug penetration [61].This can have a crucial effect on the establishment of resistance.For example, it can allow (or not) the expansion of a resistance-conferring mutation from a single cell to a whole subpopulation [62].Drug concentration has also been shown to impact the nature of beneficial mutations and affect whether a population will develop resistance or tolerance [63].Local population size is also heterogeneous.For example, a localized higher density of cells could be the result of a biofilm having developed on a catheter [8,64].Our findings suggest that the benefits of losing either of two consecutive steps in a path, both of which confer resistance on their own, will be dictated by the local conditions and density.In conditions that allow cross-feeding and diffusion of small molecules, the loss of the initial step (Fcy1) would not be advantageous.However, for isolated cells or in conditions where the diffusion of nutrients is highly limited, it might be, hence the discrepancy in the resistance phenotype of the fcy1Δ mutant on a shared plate versus in an isolated well (S7 Fig) .In comparison, the loss of the second step (Fur1) is advantageous, no matter if cells are isolated or in close contact (Fig 6C ).
Ultimately, evolution experiments such as our study highlight relatively narrow but still very relevant features of evolution of resistance.And even with the complex challenge that is the accurate reproduction of clinically relevant environments, these studies still manage to capture resistance mutations found in clinical isolates [44].

Strains, plasmids, primers and culture media
Strains used in this study are listed in S1 Data.We used two environmental strains of S. cerevisiae: LL13-040 [65] and NC-02 [66], both wild isolates collected from trees in North America.Unlike LL13-040, previous whole-genome analysis of NC-02 revealed no active transposable element [66], which could potentially change the types of resistance mutations the strains would have access to.For both strains, only haploids were used to be able to detect recessive mutations.First, the HO locus was replaced by a nourseothricin resistance marker in LL13-040 and a hygromycin resistance marker in NC-02.Strains underwent sporulation and dissection.Haploid selection and mating type were confirmed by multiplexed PCR.All subsequent experiments were performed using the corresponding haploid strains of mating type MATa.They are referred to thereafter as the "parental" strains in our experimental evolution.The two strains have 0.39% nucleotide divergence (averaged across chromosomes).

Experimental evolution
The evolution experiment was adapted from [67] to isolate a large number of independent 5-FC resistant mutants.Parental strains were streaked from glycerol stocks onto YPD agar medium, then incubated three days at 30˚C. 10 mL YPD was inoculated with a single colony, then incubated for 24 hours at 30˚C with shaking.The corresponding number of mitotic generations was calculated from measurements of cell concentrations before and after incubation using cell counts estimated by flow cytometry for four biological replicates, with a Guava easy-Cyte HT cytometer (Cytek, blue laser): LL13-040, 5.3 ± 0.5 generations; NC-02, 5.4 ± 0.5 generations.Each 24 h preculture was used to inoculate a 96-deep-well plate with 1 mL synthetic minimal medium (SD) supplemented with 1.56 μg/mL 5-FC, at a final concentration of 0.1 OD 600 .Border wells were filled with 1 mL sterile water to prevent evaporation.Plates were sealed with porous adhesive film and incubated for 72 hours at 30˚C with shaking.Finally, cultures were spotted on SD agar plates with or without 6.25 μg/mL 5-FC to select 5-FC resistant mutants.All mutants were streaked on YPD agar medium to isolate single colonies, only one of which per mutant was selected for further experiments.
rho -(petite) mutants were identified by their morphotype.Once enough mutants were generated, a 96-well plate with 1 mL YPD was inoculated and incubated overnight at 30˚C.The plate was then used to prepare a glycerol stock, as well as spotting on YPG agar medium to confirm rho status.Overall, the evolution experiment was iterated eight times, for a total of 49 precultures (26 for LL13-040 and 23 for NC-02), generating a total of 682 5-FC resistant mutants (296 LL13-040 mutants and 386 NC-02 mutants).Two more iterations without 5-FC selection were performed to evaluate whether acclimation to minimal medium could be enough to acquire resistance.It was not, since neither generated any 5-FC resistant mutant.
Media conditions for the evolution experiment were initially determined based on preliminary growth assays using the constructed haploid LL13-040 strain.Specifically, we noted that the use of minimal medium greatly increases 5-FC sensitivity, the absence of uracil being partly responsible for this effect (S9A Fig) .The final concentration of 5-FC used in the evolution experiment was chosen based on dose-response curves in SD medium obtained from biological triplicates of both strains.In the conditions of this experiment (see section below), a concentration of 1.56 μg/mL 5-FC corresponds to inhibition coefficients of 93.5 ± 0.8% and 91.8 ± 0.3% for LL13-040 and NC-02, respectively (S9B Fig) .In an initial experimental evolution trial, this concentration yielded 15 LL13-040 spontaneous mutants, compared to only 3 when using 0.78 μg/mL 5-FC.A 10% cutoff (at least six mutants out of 60 wells inoculated on each plate) ultimately prompted us to use 1.56 μg/mL 5-FC for all iterations of the evolution experiment.

Automated growth measurements on solid medium
The growth of resistant strains, as well as parental strains and deletion mutants for FCY1 and FCY2 was assayed with a BM3-SC robot (S&P Robotics Inc.).First, plates of mutants were printed on YPD OmniTrays.Then, mutants were rearrayed depending on their rho status.Finally, the plates were replicated onto solid medium, either rich medium, or minimal medium with or without antifungal.The following sections describe in more detail the protocol followed depending on the rho status.rho + mutants.A liquid culture of strain s_009 was used to print a border on a YPD 384-array.300 rho + mutants (including controls) were then rearrayed onto the same 384-array from five 96-array sources.The resulting 384-array was expanded into a 1536-array to have four replicates for each mutant.The 1536-array was replicated six times to obtain the source plates for the fitness measurements.Source 1 was replicated onto five destination plates: YPD, SD, SD + 25 μg/mL 5-FC, SD + 1.56 μg/mL 5-FU, SD + 6.25 μg/mL 5-FU, which were incubated for 22 h at 30˚C.Sources 2-6 were each replicated onto three destination plates for a total of 15 conditions: YPD, SD, SD + 25 μg/mL 5-FC, SD + 16/32/64 μg/mL fluconazole, SD + 4/8/16 μg/mL nystatin, SD + 0.25/1/2 μg/mL caspofungin and 0.0625/0.25/0.5 μg/mL micafungin.Plates were incubated for 90 h at 37˚C.During both incubations, pictures of each plate were taken every 2 h in a spImager custom robotic platform (S&P Robotics Inc.).rho -mutants.Similarly, two 96-arrays were prepared, one with only LL13-040 mutants, the other with only NC-02 mutants.For both, a liquid culture of the corresponding fcy1Δ mutant was used to print the border.118 rho -mutants as well as controls were rearrayed from either three (NC-02) or four (LL13-040) 96-array sources.The two resulting 96-arrays were expanded into two 384-arrays to have four replicates for each mutant.For each background, the 384-array was replicated five times to obtain the source plates for the fitness measurements.Similar growth media conditions were tested.Additionally, both plates were replicated onto YPG arrays to confirm rho status.Following this quality control step, four mutants were detected as being initially misannotated as rho -(three LL13-040 mutants and one NC-02 mutant) and were therefore excluded from all analyses (S10 Fig) .Colony size analyses.Pictures were cropped, then converted into inverted gray levels using scikit-image [68].Colony sizes were quantified using pyphe-quantify in batch mode [69].The area parameter was used to generate growth curves and calculate the corresponding area under the curve (AUC) using the composite trapezoidal rule.For each mutant, the relative fitness was calculated as the ratio of the mean AUC (absolute fitness) divided by the corresponding mean AUC obtained for the WT parental strain.

Cytometry
Rhodamine accumulation assay.In order to evaluate the efflux capacity of rho -mutants, we used rhodamine 6G, a fluorescent dye known to be a substrate of ABC transporters.Measurement of intracellular accumulation of rhodamine was adapted from [70].The two 96-arrays of rho -mutants for LL13-040 and NC-02 (obtained as described above) were inoculated in 96-deep-well plates containing 1 mL YPD and incubated overnight with shaking at 30˚C.For both plates, two controls were added: the parental strain and a strain with deletions for all ABC transporters (PDR5 SNQ2 YBT1 YCF1 YOR1, s_012 in S1 Data) [71], therefore incapable of rhodamine efflux.In the morning, both plates were subcultured at 0.15 OD 600 , then incubated at 30˚C with shaking until they reached 0.6 OD 600 .For this step, one of the mutants was subcultured twice to later have a control without rhodamine.200 uL was transferred to a sterile V-shaped plate.Rhodamine 6G (Millipore Sigma) was added to a final concentration of 10 μg/mL except for 1 well.Plates were sealed with porous adhesive film and incubated 30 min at 30˚C with shaking.Cultures were centrifuged 5 min at 230 g and pellets were washed with sterile PBS.Cultures were centrifuged again and pellets were resuspended in 200 μL sterile PBS with 0.2% glucose.Cultures were incubated an extra 30 min at 30˚C to activate energy-dependent efflux, then diluted 1:10 prior to fluorescence measurements by flow cytometry.Fluorescence in the green and orange channels using blue and green lasers, respectively, was acquired for approximately 2,000 events per sample with a Guava easyCyte HT cytometer (Cytek).Fluorescence values in both channels were normalized with the cell size using the FSC value.Thresholds were set for both normalized fluorescence values to maximize events with fluorescence below the thresholds for the negative control (sample not treated with rhodamine), as well as maximize events with fluorescence above the thresholds for the positive control (s_012 in S1 Data).The percentage of cells with fluorescence above both thresholds is considered to be inversely correlated with the efflux capacity.
Fluorescence reporter assay.A GFP reporter was used to evaluate the level of expression of FCY1 in one of the parental strains to assess FCY1 functionality.Precultures of LL13-040 with or without FCY1-mEGFP (s_002 and s_003 in S1 Data) were prepared from single colonies in 5 mL YPD medium and incubated overnight at 30˚C with shaking.Precultures were diluted at 1.5 OD 600 in sterile water.24-well plates containing either SC, SC -ura, SD (MSG) or SD (NH 4 ) with 0, 0.78 or 1.56 μg/mL 5-FC were inoculated at 0.15 OD 600 , sealed with porous adhesive film and incubated for 4 h at 30˚C with shaking.Cultures were diluted at 0.05 OD 600 .Fluorescence in the green channel using a blue laser was acquired for 5,000 events.Events with a SSC value (proxy for cell granularity) below 300 or a FSC value (proxy for cell size) below 2,000 were filtered out.The fluorescence values were normalized by the FSC value.

DNA extraction and sequencing
In order to identify which mutation(s) were conferring 5-FC resistance in rho + mutants, we performed whole-genome sequencing.The mutants were cultured overnight at 30˚C in 24-well plates with 2 mL YPD, sealed with porous adhesive film.Cells were pelleted from a total of approximately 20 OD units per sample.DNA was extracted using the MasterPure Yeast DNA Purification kit (Epicentre) following the kit's protocol, except for the following adjustments.Isopropanol precipitation was performed for 1 h at room temperature.Pellets were dried at 55˚C for 25 min, after which they were resuspended in 50 μL 0.2 ng/μL RNase.Samples were incubated an additional 5 min at 55˚C, then immediately purified using SPRI beads (1:1 ratio, Axygen).Attempts to extract genomic DNA from rho -mutants proved unsuccessful with this method.
Genomic DNA was stored at -20˚C, then quantified in technical triplicates using the Accu-Clear Ultra High Sensitivity dsDNA Quantitation kit.A total of 282 samples (276 out of 294 rearrayed rho + mutants, as well as the parental strain and the fcy1Δ and fcy2Δ mutants for each background, s_004, s_005, s_009, s_010) were diluted to 10 ng/μL for a total of 40 ng and arranged on three 96-well plates.For eight samples, whose initial concentration was too low, a volume of 15 μL was pipetted and dried in a SpeedVac, then resuspended in the appropriate volume to the desired final concentration of 10 ng/uL.Sequencing libraries were prepared using the Riptide High Throughput Rapid DNA Library Prep kit (iGenomx), according to the kit's protocol.The three 96-well plates were each pooled separately.For the amplification step, we modified the protocol to use barcoded primers for each pool instead of a single universal PCR primer.The three pools were SPRI-purified in two steps with protocol option 3 (recommended for PE150 sequencing).The pools were checked on agarose gel, quantified with the AccuClear Ultra High Sensitivity dsDNA Quantitation kit and analyzed on BioAnalyzer (Agilent).Finally, they were sequenced in paired-end 150 bp on an Illumina NovaSeq 6000 S4 system at the Centre d'expertise et de services Ge ´nome Que ´bec.
FCY1 and FUR1 were amplified from the same genomic DNA using primers listed in S1 Data, then sequenced by Sanger amplicon sequencing.For FUR1 amplicons, all samples for which enough material could be recovered (after preparing WGS libraries) were processed.The sequencing reactions were performed at the Centre Hospitalier de l'Universite ´Laval sequencing platform (Universite ´Laval, Que ´bec).

Validations and complementations
Once we identified FUR1 as being the hotspot for mutations in rho + mutants, we confirmed the detected mutations were causal by using two types of growth assays: validation refers to a growth assay in isolated liquid medium where we compared the growth of mutants to that of the fur1Δ mutant, whereas complementation refers to measuring the growth of mutants upon expression of the WT allele of FUR1 carried by a plasmid.
Complementation.32 out of the 35 validation mutants (16 from each background) and the fur1Δ mutants were transformed with pMoBY or pMoBY-FUR1 [72] using a modified yeast transformation method [73].Transformants were selected with combinations of nourseothricin or hygromycin B and G418.Precultures were prepared in a single 96-deep-well plate with 1 mL YPD + G418, inoculated from a single colony for each strain and incubated overnight at 30˚C with shaking.Precultures were diluted at 1 OD 600 in sterile water and inoculated at a final concentration of 0.1 OD 600 in a sterile Greiner 96-well plate with SD + 100 μg/ mL 5-FC.OD 600 measurements were performed at 30˚C every 15 min until a plateau was reached in a BioTek Epoch 2 Microplate Spectrophotometer (Agilent).

Growth assays in isolated liquid cultures
Three types of growth assays in isolated liquid cultures were carried out.1-'Small-scale' assays (precultures in 5 mL tubes and growth measurements in 96-well plates) were used a) to evaluate 5-FC sensitivity in different media (minimal and rich) and b) to measure the growth of parental strains in minimal medium with different sources of nitrogen.2-A single 'largerscale' growth assay (precultures in 96-well plates and growth measurements in 384-well plate) was used to measure the growth of a subset of rho + mutants, as well as the fcy1Δ mutant, in liquid medium, in order to compare with the growth measured on solid medium.3-A doseresponse curve assay was used to measure the inhibition coefficient corresponding to the 5-FC concentration used in the evolution experiment.
'Small-scale' growth assay.For the 'small-scale' growth assays, strains were streaked on YPD agar medium supplemented with the appropriate antibiotics.Precultures were prepared from three isolated colonies in 5 mL YPD medium with appropriate selection and incubated overnight at 30˚C with shaking.Precultures were diluted at 1 OD 600 in sterile water.Sterile Greiner 96-well plates were prepared with the indicated culture medium and compounds and cultures were inoculated at a final concentration of 0.1 OD 600 .OD 600 measurements were performed at 30˚C every 15 min until a plateau was reached in a Tecan Infinite M Nano (Tecan Life Sciences).
'Larger-scale' growth assay.For the 'larger-scale' growth assay, precultures were prepared in a 96-deep-well plate with 1 mL YPD and incubated overnight at 30˚C with shaking.Precultures were diluted at 1 OD 600 in sterile water.A sterile 384-well plate containing four different conditions was used: SD, SD + 25 μg/mL 5-FC, SD + 100 μg/mL 5-FC and YNB + 2% glucose + 250 μg/mL cytosine.The lid was conditioned by incubating 3 min with 5 mL 0.05% Triton X-100 / 20% ethanol and dried under sterile conditions.Cultures were inoculated in a single replicate at a final concentration of 0.1 OD 600 in a final volume of 80 μL per well.OD 600 measurements were performed at 30˚C every 15 min in a Tecan Spark (Tecan Life Sciences) with an active Tecool temperature control module, until curves showed signs of evaporation (13 h).
Dose-response curve assay.A single 96-well plate was prepared as described for the 'small-scale' growth assays, with cultures of LL13-040 and NC-02 in biological triplicates.5-FC was serially diluted 1:2 seven times starting from a final concentration of 3.125 μg/mL.The maximum growth rate was transformed into the inhibition coefficient, with an inhibition coefficient of 0 corresponding to the maximum growth rate measured in the absence of 5-FC.

Assessment of cross-feeding-induced toxicity
Three types of growth-based assays were carried out to investigate the selection of fur1 - mutants over fcy1 -mutants.A liquid medium conditioning assay was used to evaluate if a toxic compound was secreted by the fur1Δ mutant.A competition assay was used to assess the importance of cell density in liquid medium.Finally, an inhibition assay was performed to confirm the phenotype on solid medium.
Medium conditioning assay.Precultures of LL13-040 and NC-02 with or without deletion of FCY1 or FUR1 (s_002, s_004, s_006, s_008, s_009, s_011) were prepared from three isolated colonies in 5 mL YPD medium and incubated overnight at 30˚C with shaking.Precultures were diluted at 1 OD 600 in sterile water.Subcultures of 5 mL SD medium with or without 1.56 μg/mL 5-FC in 50 mL Falcon tubes were inoculated at a final concentration of 0.1 OD 600 .After a 5h incubation at 30˚C with shaking, cultures had reached 0.5-0.6OD 600 .Cultures were centrifuged 5 min at 500 g and the supernatants were filtered, then transferred to a sterile Greiner 96-well plate.The media "conditioned" by the growth of the fcy1Δ mutant were inoculated with either the FUR1 null mutant or the WT.The media conditioned by the growth of the FUR1 null mutant were inoculated with either the fcy1Δ mutant or the WT.Cultures were inoculated at a final concentration of 0.1 OD 600 .Growth curves were acquired as described above.
Competition assay.Precultures of LL13-040 and NC-02 with or without deletion of FCY1 or FUR1 (s_002, s_004, s_006, s_008, s_009, s_011) were prepared as described above.Precultures were diluted at 1 OD 600 in sterile water.Subcultures of 5 mL SD medium without 5-FC were inoculated at a final concentration of 0.1 OD 600 .After a 5h incubation at 30˚C with shaking, cultures had reached 0.6 OD 600 .For each background, cultures of both deletion mutants were pooled in equal volumes and serially diluted in sterile water.0.5 mL of each dilution (10 −3 , 10 −4 and 10 −5 OD) was plated on large Petri dishes containing SD with 1.56 μg/mL 5-FC.From each plate, 16 colonies were streaked on the same medium to make sure they were 5-FC resistant.Additionally, the genotype (fcy1Δ or fur1Δ) was identified by multiplexed colony-PCR using one common primer annealing in either marker (NAT or HPH) and two genespecific primers annealing upstream the deletion site.The reported CFU counts correspond to "W".End chromosome bias (increasing read count towards chromosome ends) was corrected by fitting a curve for the windows closest to one chromosome end (each half of the chromosome), using the LOWESS smoothing method, in python v3 statsmodels package, with a default of 2/3 of data points to estimate each y value.Windows were then divided by the fitted line to obtain the normalized read count.
Predictions of impact of mutations.Mutations were submitted to the online tool mutfunc to predict their impact, notably on stability and/or conservation in a homology model [79].

Fig 1 .
Fig 1. Pyrimidine and 5-FC import and metabolism around some of the known and potential mechanisms of resistance.5-Fluorocytosine (5-FC), cytosine, uracil and 5-fluorouracil (5-FU) are represented as blue, dark purple, light purple and light blue beads, respectively.For clarity, only two permeases known to import pyrimidines are shown: Fcy2 and Fur4.In prototrophic yeast strains, pyrimidines can be synthesized de novo from amino acids such as glutamine and aspartate[29].The steps leading to the obtention of the nucleotide precursors uridine monophosphate (UMP) and uridine diphosphate (UDP) are catalyzed by Ura3 and Ura6, respectively, in the de novo pathway[30,31].Alternatively, pyrimidines can be imported from the medium[32].Uracil can also be obtained by deamination of cytosine by Fcy1[33].Finally, UMP can be obtained by the addition of a phosphate group by Fur1[34].5-FC is a prodrug, which is metabolized the same way cytosine is, except its deamination leads to the obtention of the cytotoxic compound 5-FU[35].Based on work done in laboratory strains or other fungal species, resistance to 5-FC can arise when either of the following three steps are compromised: import, conversion and activation, typically through mutations inactivating Fcy2 (for example), Fcy1 and Fur1, respectively[36].We also hypothesize an alternative path to resistance in the pleiotropic drug response, wherein loss of mitochondrial function leads to derepression of the transcription factor Pdr3, which in turn overexpresses efflux pumps such as Pdr5. Figure created with BioRender, using the structures of Fcy1 (1P6O), Fur1 from C. albicans (7RH8), Ura3 (3GDL), Ura6 (1UKZ) and Pdr5 (7P04) and AlphaFold predictions of Fcy2 (P17064) and Fur4 (P05316)[37,38].

Fig 2 .
Fig 2. Growth of individual mutants relative to the parental strains in various media.Relative growth corresponds to the mean area under the curve (AUC, calculated on 22 h) from four replicate colonies, normalized by the WT for individual strains arrayed on solid media: YPD, SD, SD + 25 μg/mL 5-FC and SD + 6.25 μg/mL 5-FU.A two-way ANOVA followed by Tukey's multiple comparison test was performed to compare the relative fitness of rho + mutants to that of rho -mutants for each background in each condition.Statistical significance is as follows: ****, adj.p-value < 0.0001; ***, adj.pvalue < 0.001; *, adj.p-value < 0.05; ns, not significant.Insets show the correlation between relative fitness in SD + 5-FC and relative fitness in SD + 5-FU, with the corresponding Spearman's rank correlation coefficient (S. rho) and p-value.https://doi.org/10.1371/journal.pgen.1011002.g002

Fig 3 .
Fig 3. Cross-resistance of rho -mutants to 5-FC and fluconazole may be explained by efflux capacity.A) Relative growth corresponds to the mean area under the curve (AUC, calculated for the first 22 h of incubation) from four replicate colonies, normalized by the WT for individual strains arrayed on solid media: YPD, SD, SD + 25 μg/mL 5-FC, SD + 0.5 μg/mL micafungin, SD + 2 μg/mL caspofungin, SD + 16 μg/mL nystatin and SD + 64 μg/mL fluconazole.A two-way ANOVA followed by Tukey's multiple comparison test was performed to compare the relative fitness of rho + mutants to that of rho -mutants for each background in each condition.Statistical significance is as follows: ****, adj.p-value < 0.0001; ***, adj.pvalue < 0.001; *, adj.p-value < 0.05; ns, not significant.B) Rhodamine retention assay.Blue bars indicate the resulting detection thresholds (one negative control for each background i.e. cells not treated with rhodamine).abcΔ (s_012 in S1 Data) was used as the positive control.C, D) Rhodamine retention (data from panel B normalized with the WT) and relative growth in C) fluconazole or D) 5-FC (data from panel A) are compared, with the corresponding Spearman's rank correlation coefficient (S. rho) and p-value.https://doi.org/10.1371/journal.pgen.1011002.g003

Fig 4 .
Fig 4. Genome sequences analysis and validation.A) Upset plots showing the number of unique mutated genes (# genes) detected by gatk, samtools or both for each background.Categorical plots on top show the number of distinct genomes for which at least one variant was detected in a given gene (one dot per gene, colored according to the number of distinct mutations which have been detected).B) Distribution of the number of genes with variants per genome (as detected either by gatk or samtools) for each background.C) Data from Fig 2, where dots are colored by the mutated/deleted gene most likely to confer resistance.D) Growth assay in liquid medium for 35 resistant mutants for which a variant in FUR1 was detected.Cultures were inoculated in SD with or without 100 μg/mL 5-FC.Relative growth corresponds to the mean area under the curve from two replicates (AUC, calculated on 46h) normalized by the WT.https://doi.org/10.1371/journal.pgen.1011002.g004

Fig 5 .
Fig 5. Mutations observed in Fur1 lead to loss of function.(A) Upset plots showing the number of unique mutations identified in Fur1 for each background.Categorical plots on top show the number of distinct genomes for which the same mutation was detected.Only one amino acid change resulted from two adjacent mutations (back-to-back in the same codon), all others corresponding to a single SNP or indel.B) Growth assay for 32 out of 35 resistant mutants tested in Fig 4D, with pMoBY expressing or not FUR1 from its native promoter.Relative growth corresponds to the area under the curve calculated on 25 h for the mutant with pMoBY-FUR1 divided by the same parameter for the mutant with pMoBY.Mutants were grown from single colonies in SD + 100 μg/mL 5-FC (+ G418 to maintain pMoBY).Colored dots indicate the same mutation was detected in the mutants from both backgrounds.Two mutations for which expression of the WT allele of FUR1 did not restore 5-FC sensitivity are highlighted.A closer look at the residues in the next panel suggests a putative role in assembly of the tetramer.C) Location of substitutions from our dataset on the predicted structure of Fur1, visualized using UCSF ChimeraX[40].The tetramer (C.albicans Fur1 assembly 7RH8) is represented in light gray, with 1 chain in dark gray.Residues found to be mutated at least once in our dataset are colored in pink.UTP molecules are represented in sticks colored by atom.Two insets zoom in on one interface each (the slightly opaque one depicts the interface at the back of the represented structure).The highlighted residues (each one from a different monomer) appear to interact with one another to ensure proper assembly of the tetramer.

Fig 6 .
Fig 6.FCY1-mediated resistance is context-dependent.A) Predicted cross-feeding interactions between cells with different alleles of FCY1 and FUR1.Uracil (top) and 5-FU (middle) are expected to diffuse in the medium and be converted by Fur1 of neighboring cells, which would allow or inhibit growth, respectively.Notably, the growth of any FUR1 + cells, regardless of their FCY1 allele, will be inhibited because of 5-FU cross-feeding.On the contrary, any fur1 -cells would grow despite 5-FU cross-feeding (bottom).Figure created with BioRender, using the structures of Fcy1 (1P6O) and Fur1 from C. albicans (7RH8).B) Medium conditioning assay.First, cultures of the deletion mutants were prepared in SD with or without 1.56 μg/mL 5-FC.After reaching an OD 600 of 0.6, the cultures were filtered and the medium was transferred to a 96-well plate.Cultures were inoculated with either the WT or the other deletion mutant.Relative growth corresponds to the area under the curve (AUC) normalized by the WT and was obtained from biological triplicates.A two-way ANOVA followed by Tukey's multiple comparison test was performed to compare the https://doi.org/10.1371/journal.pgen.1011002.g006 Fig), in addition to all possible stop codons and other possible frameshifting indels.Not only the mutations picked up in our experiments capture both of these phenomena (S4 Fig), some also appear to likely affect proper assembly of the Fur1 tetramer (Fig 5C).