Stress combined with loss of the Candida albicans SUMO protease Ulp2 triggers selection of aneuploidy via a two-step process

A delicate balance between genome stability and instability ensures genome integrity while generating genetic diversity, a critical step for evolution. Indeed, while excessive genome instability is harmful, moderated genome instability can drive adaptation to novel environments by maximising genetic variation. Candida albicans, a human fungal pathogen that colonises different parts of the human body, adapts rapidly and frequently to different hostile host microenvironments. In this organism, the ability to generate large-scale genomic variation is a key adaptative mechanism triggering dangerous infections even in the presence of antifungal drugs. Understanding how fitter novel karyotypes are selected is key to determining how C. albicans and other microbial pathogens establish infections. Here, we identified the SUMO protease Ulp2 as a regulator of C. albicans genome integrity through genetic screening. Deletion of ULP2 leads to increased genome instability, enhanced genome variation and reduced fitness in the absence of additional stress. The combined stress caused by the lack of ULP2 and antifungal drug treatment leads to the selection of adaptive segmental aneuploidies that partially rescue the fitness defects of ulp2Δ/Δ cells. Short and long-read genomic sequencing demonstrates that these novel genotypes are selected via a two-step process leading to the formation of novel chromosomal fragments with breakpoints at microhomology regions and DNA repeats.


Introduction
Understanding how organisms survive and thrive in changing environments is a fundamental question in biology. Genetic variation is central to environmental adaptation because it facilitates the selection of fitter genotypes better adapted to a new environment. Different types of genetic changes contribute to genetic variability, including (i) whole-chromosome or segmental-chromosome aneuploidy, (ii) translocations and (iii) mutations [1]. Furthermore, diploid cells can undergo loss of heterozygosity (LOH) driven by mitotic events, such as cross-over, gene conversion or meiotic reversion [1,2]. Whole-chromosome or segmental chromosome aneuploidies have the greatest effect on adaptation as they generate copy number variations (CNVs) of multiple genes. These result in divergent phenotypes which may be selectively advantageous [3].
Genome plasticity-the ability to generate large-scale genomic variation-is emerging as a critical adaptive mechanism in human microbial pathogens that need to adapt quickly to extreme environmental shifts because it provides genetic diversity upon which selection can act [4][5][6][7][8]. One such organism is Candida albicans, a common human fungal pathogen and a prevalent cause of death due to systemic fungal infections [9]. C. albicans is part of the normal microbiota of most healthy individuals but, in immunocompromised individuals, it is a dangerous pathogen causing a wide range of infections, including life-threatening disseminated diseases [10]. Azole antifungal agents, such as fluconazole (FLC), are the most commonly prescribed drugs for treating C. albicans infections [9,11,12].
Several lines of evidence suggest that C. albicans genome plasticity provides a competitive advantage under host-relevant stress environments. C. albicans is a diploid organism with a heterozygous genome organised into 2 × 8 (2n = 16) chromosomes (Chr) [13,14]. Seven chromosomes are designated Chr1 to Chr7 according to size, while one is termed ChrR because it contains the rDNA locus [15]. Genomic analysis of clinical isolates reveals that many C. albicans strains have large-scale genomic changes including segmental and whole chromosome aneuploidies [16][17][18][19]. Furthermore, specific chromosomal variants are selected during hostniche colonisation [17,[20][21][22][23][24][25]. Accordingly, many drug-resistant isolates exhibit karyotypic diversity that can confer resistance due to increased copies of specific genes. For example, CNV for the gene ERG11 encoding for the target of FLC, lanosterol 14-alpha-demethylase is often observed in drug-resistant isolates [4,16,[26][27][28]. Several studies suggest that C. albicans genome instability is not random as it occurs more frequently at specific hotspots which are often repetitive [17,19,23,29,30]. Subtelomeric regions and the rDNA locus are among the most unstable genomic sites [17,31]. C. albicans subtelomeric regions are enriched in repetitive sequences derived from transposons and protein-coding genes [29,32]. Most notable are the telomere-associated TLO genes, a family of 14 closely related paralogues encoding proteins similar to the Mediator 2 subunit of the Mediator transcriptional regulator [33][34][35]. Most TLO genes are located at subtelomeric regions except TLO34, located at an internal locus on the left arm of Chr1 [33]. The rDNA locus consists of a tandem array of a~12 kb unit repeated 50 to 200 times; rDNA length polymorphisms frequently occur [14,17].
Despite the clear correlation between genomic variation and environmental adaptation, pinpointing the environmental pressure(s) selecting specific genotypes and understanding how complex karyotypes are formed is often difficult.
In this study, we performed a genetic screening to identify modulators of C. albicans genome stability. The screen led to the identification of the ULP2 gene, encoding for a SUMO protease. ULP2 deletion causes increased genome instability and enhanced genome variation leading to fitness defects and hypersensitivity to genotoxic agents. We show that loss of ULP2 combined with exposure to an additional stress (FLC) leads to the selection of multichromosome segmental aneuploidies with adaptive power. Long-read genomic sequencing demonstrates that these novel segmental aneuploidies are selected by a two-step process producing chromosomal fragments with breakpoints at microhomology regions and DNA repeats. Thus, exposure to stress can increase tolerance to unrelated stress by selecting novel complex genotypes.

A systematic genetic screen identifies Ulp2 as a regulator of C. albicans genotoxic stress response
To identify factors regulating C. albicans genome integrity, we utilised a deletion library comprising a subset (674/3000) of C. albicans genes that are not conserved in other organisms or have a functional motif potentially related to virulence [36]. As defects in genome integrity lead to hypersensitivity to genotoxic agents [37], the deletion library was screened for hypersensitivity to two DNA damaging agents: Ultraviolet (UV) irradiation which induces formation of pyrimidine dimers [38], and methyl methanesulfonate (MMS), which leads to replication blocks and base mispairing [39]. Genotoxic stress hypersensitivity was semi-quantitatively scored by comparing the growth of treated versus untreated on a scale of 0 to 4, where 0 indicates no sensitivity, and 4 specifies strong hypersensitivity (Fig 1A). The screen identified 28 gene deletions linked to DNA damage hypersensitivity (UV or MMS score �2). Of those deletion mutants, 9/28 hits show sensitivity to both UV and MMS, 6/28 hits are sensitive only to UV, and 13/28 hits are sensitive only to MMS (S1 Table). Functional prediction analysis demonstrated that~43% of the hits are genes predicted to encode components of the DNA damage response pathway (5/28) or for proteins necessary for cell division (7/28) (S1 Table). For example, the top 4 hits of the screen were GRR1, KIP3, MEC3 and RAD18 genes (S1 Table) One of the highest-ranked genes on our screen is ULP2 (CR_03820C/ orf19.4353: MMS score:3, UV score:3) encoding for a SUMO protease (Fig 1A and 1B and S1 Table).
Colony-forming unit (CFU) assays of UV-treated cells confirmed the importance of C. albicans ULP2 in DNA damage resistance as UV treatment reduced the number of CFU in a ulp2Δ/Δ strain (~14.5% survival) compared to a wild-type (WT) strain (~33.7% survival) ( Fig  1C). Furthermore, the ulp2Δ/Δ strain also displayed a reduced growth rate in liquid media containing MMS or Hydroxyurea (HU), a chemotherapeutic agent that challenges genome ULP2 is a regulator of the C. albicans genotoxic stress response. (A) Schematic representation of the screening strategy. 674 C. albicans deletion strains were screened for hypersensitivity to UV and MMS. Hypersensitivity was scored by comparing the growth of treated vs untreated on a scale of 0 (white) to 4 (magenta). Black � : genes encoding for DNA damage and sensing repair pathway components, Blue � : genes encoding for cell division and chromosome segregation machinery, Green arrow: ulp2Δ/Δ (B) Data for a plate containing ulp2Δ/Δ strain (magenta circle). Growth on Non-selective (N/S) media or following UV and MMS treatment is shown (C) Colony-forming unit assay (% survival) of UV treated WT and ulp2Δ/Δ strain. Statistical analysis was performed using the Kruskal-Wallis test with Mann-Whitney U test for post hoc analysis (D) Growth curve of WT and ulp2Δ/Δ strains grown in non-selective (N/S) and MMS-containing liquid media. Error bars: standard deviation (SD) of three biological replicates (E) Growth curve of WT and ulp2Δ/Δ strains grown in non-selective (N/S) and HU-containing liquid media. Error bars: standard deviation (SD) of three biological replicates. https://doi.org/10.1371/journal.pgen.1010576.g001

PLOS GENETICS
integrity by stalling replication forks [57] (Fig 1D and 1E). Thus, ULP2 has a role in responding to a wide range of genotoxic agents.

ULP2 but not ULP1 is required for survival under stress
C. albicans contains three putative SUMO-deconjugating enzymes: Ulp1, Ulp2 and Ulp3 (Fig  2A). Sequence comparison between the three C. albicans Ulp proteins and the two well-characterised S. cerevisiae Ulps (Ulp1 and Ulp2) reveals that although the C. albicans proteins are poorly conserved, the amino acid residues essential for catalytic activity are conserved (Fig 2A  and 2B). Accordingly, recombinantly expressed C. albicans Ulp1, Ulp2 and Ulp3 have SUMOprocessing activity in vitro [58]. Similarly to S. cerevisiae ULP1, C. albicans ULP3 is an essential gene and was not investigated further in this study [59,60]. Previous studies failed to detect a poly-histidine tagged Ulp2 protein by Western blot analyses of C. albicans protein extracts [58]. These results suggested that Ulp2 is unstable or expressed at undetectable low levels. We reassessed Ulp2 protein levels by generating strains expressing, at the endogenous locus, an epitope-tagged Ulp2 protein (Ulp2-HA). Western blot analyses show that Ulp2-HA expression is readily detected in extracts from four independent integrant strains (Fig 2C). Thus, a stable Ulp2 protein is expressed in cells grown under standard laboratory growth conditions (YPD, 30˚C). To assess whether ULP1, similarly to ULP2, is involved in genotoxic stress response, we engineered homozygous deletion strains for ULP1 (ulp1Δ/Δ) and ULP2 (ulp2Δ/Δ). Growth analysis demonstrated that deletion of ULP2 reduces fitness as the newly generated ulp2Δ/Δ strain is viable, but cells are slow-growing (Fig 2D and 2E). In contrast, the ulp1Δ/Δ strain grows similarly to the WT control in solid and liquid media (Fig 2D and 2E). Spot dilution assay confirmed that ULP2 is an important regulator of C. albicans stress response as, similarly to the deletion library mutant, the newly generated ulp2Δ/Δ strain was sensitive to different stress conditions including treatment with DNA damaging agents (UV and MMS), DNA replication inhibitor (HU), oxidative stress (H 2 O 2 ) and high temperature (39˚C) (Fig 2E). In contrast, deleting ULP1 did not cause any sensitivity to the tested stress conditions (Fig 2E).
Although ULP-1, ULP-2 and ULP-3 may have some partially redundant functions, our results suggest that ULP-1 does not play a major role in genotoxic stress response. In summary, loss of ULP2 leads to poor growth in standard laboratory growth conditions and hypersensitivity to multiple stresses.

Genome instability is exacerbated in the absence of ULP2
To assess whether the hypersensitivity to DNA damage agents observed in the ulp2Δ/Δ strain was indeed due to enhanced genome instability, we deleted ULP2 from a set of strains containing a heterozygous URA3 + marker gene inserted in three different chromosomes (Chr 1, 3 and 7) [61]. We quantified the frequency of URA3 + marker loss by plating on media containing the counter-selective drug 5-Fluoroorotic Acid (FOA) and scoring the number of colonies able to grow on FOA-containing media compared to non-selective (N/S) media. Deletion of ULP2 caused a dramatic increase in LOH rate at all three chromosomes (Chr1: 378X, Chr3: 18X, Chr7: 96X), indicating that ULP2 is required for maintaining genome stability across the C. albicans genome (Fig 3A). In C. albicans, hypersensitivity to genotoxic stress leads to filamentous growth [37, [62][63][64][65]. Accordingly, the ulp2Δ/Δ strain formed wrinkled colonies on solid medium and displayed a higher frequency of abnormal morphologies than the WT strain, including filamentous cells (Fig 3B and 3C). To assess whether the exacerbated ulp2Δ/Δ genome instability is linked to defective chromosome segregation, we deleted the ULP2 gene in a reporter strain in which TetO sequences are integrated adjacent to the centromere on Chr7 (CEN7) and a TetR-GFP fusion protein is expressed from the gene-free NEUT5L locus PLOS GENETICS [66,67]. The binding of TetR-GFP to TetO sequences allowed the visualisation of Chr7 duplication and segregation during the cell cycle. We found that deletion of ULP2 leads to abnormal Chr7 segregation. This included cells with no TetR-GFP signals or multiple TetR-GFP-foci, that were~5 fold higher in the ulp2Δ/Δ strain compared to the WT control strain (Fig 3D). Thus, deletion of C. albicans ULP2 leads to increased genome instability.

ULP2 loss coupled with stress triggers selection of segmental aneuploidies
Previous studies performed in the model system S. cerevisiae demonstrated that loss of ULP2 leads to the accumulation of a specific multichromosome aneuploidy (amplification of both ChrI and ChrXII). This aneuploidy rescues the lethal defects of ulp2 deletion by amplification of specific genes on both chromosomes [68,69]. To assess whether loss of C. albicans ULP2 triggers the selection of gross karyotypic abnormalities, we analysed the genome of WT and ulp2Δ/Δ strains at the beginning (Day 0) and the end (Day 30) of an in vitro evolution experiment where strains were passaged daily for 30 days in rich media (YPD 30˚C) (Fig 4A). Clamped homogeneous electrical field (CHEF) electrophoresis analysis did not detect any major chromosome rearrangements in both sets of evolved strains (Fig 4B). To further investigate the impact of ULP2 loss on genome organisation, we sequenced the genome of 3 randomly selected ulp2Δ/Δ colonies by whole genome Illumina sequencing (WGS) and compared their genome to the C. albicans reference genome. This analysis revealed that loss of ULP2 leads to very few (<10 across the 3 isolates) de novo mutations (S2 Table). Although we did not detect CNVs, we identified novel LOH tracts on different chromosomes in two of the three sequenced colonies (Fig 4C). For example, chromosome mis-segregation followed by reduplication of the remaining homologue is detected in isolate U1 (U1: ChrR) and the Karyotype organisation of C. albicans WT and ulp2Δ/Δ strains at the start (Day 0) and the end (Day 30) of the evolution experiment (C) Whole genome sequencing analysis of the progenitor (WT:SN152) and three single ulp2Δ/Δ colonies (U1, U2, and U3). Data were plotted as the log2 ratio and converted to chromosome copy number (y-axis, 1-4 copies) as a function of chromosome position (x-axis, Chr1-ChrR) using the Yeast Mapping Analysis Pipeline (YMAP) [109]. Heterozygous (AB) regions are indicated with grey shading, and homozygous regions (loss of heterozygosity) are indicated by shading of the remaining haplotype, either AA (cyan) or BB (magenta).
https://doi.org/10.1371/journal.pgen.1010576.g004 genome of U2 contains a long-track LOH (U2:Chr3L) that occurred within 4.6 kb of a repeat locus on Chr3L (PGA18, [19]) (Fig 4C). Therefore, loss of ULP2 can trigger selection of large chromosomal variations. We hypothesised that exposure of ulp2Δ/Δ cells to stress could facilitate the selection of novel adaptive karyotypes. To test this hypothesis, we challenged the ulp2Δ/Δ strain with high concentrations of FLC (128 μg/ml;~1000 fold above susceptibility breakpoint [70]) and isolated a FLC-adapted isolate (FLC-1) that was still able to grow at high drug concentration following two passages (T1 and T2) in non-selective (N/S) media (Fig 5A  and 5B). The phenotypes associated with the loss of ULP2 were partially rescued in FLC-1. as this isolate was less sensitive than the ulp2Δ/Δ progenitor to UV treatment and high temperature (39˚C) (Fig 5B). Furthermore, fewer wrinkled colonies are present in FLC-1 than ulp2Δ/Δ and the number of elongated cells was reduced in FLC-1 compared to ulp2Δ/Δ (Fig 5C and 5D).
To identify mutations that underlie the above phenotypes, we sequenced the genome of three FLC-1 single colonies (FLC-1a, b and c) by Illumina technology (Figs 5A and S1 and S3-S5 Tables). We also sequenced the genome of 3 additional ulp2Δ/Δ FLC-adapted isolates (FLC-2, FLC-3 and FLC-4) randomly selected from FLC plates and unable to grow at high FLC doses following passaging in N/S medium. The whole-genome sequencing revealed that the FLC-adapted colonies have a genotype distinct from the ulp2Δ/Δ progenitor (Fig 5E and S3  Table). FLC-1, but not FLC-2, FLC-3, or FLC-4 isolates, is marked by a segmental aneuploidy: a partial Chr1 amplification (~1.3 Mbp) containing 535 protein-coding genes (Fig 5E and S4 Table). Furthermore, all sequenced isolates have a partial deletion (~388 Kb) of the right arm of ChrR (ChrRR-Deletion). ChrRR-deletion occurs at the ribosomal DNA and extends to the right telomere of ChrR (ChrR:1,897,750 bp-2,286,380 bp), reducing the copy number of 204 genes from two to one (Fig 5E and S5 Table). In contrast, we detected very few (<10) de novo point mutations, and none of these are common among all the sequenced FLC isolates (S3 Table). Thus, exposure to an antifungal drug triggers the selection of adaptive chromosomal variations in the absence of ULP2.

Segmental aneuploidy is selected via a two-step process
Illumina sequencing is an inadequate technology for resolving complex chromosomal abnormalities because of the generated short reads. For example, we could not establish whether the increased Chr1 CNV was due to the formation of an extrachromosomal fragment or to a chromosomal fusion. Therefore, to understand further the genomic structure of FLC-1 we sequenced the genome of this isolate using long-read Oxford Nanopore Technologies (ONT) sequencing. To establish the temporal trajectory of FLC-1 aneuploidies, timepoints T1 (1X passage in non-selective media following FLC treatment) and T2 (2X passages in non-selective medium following FLC treatment) were sequenced (Fig 5A). Using this method, we could resolve the structure of FLC-1 aneuploidies completely (Fig 6A). We discovered that FLC-1 contains, in addition to the two endogenous Chr1 homologous chromosomes, an extra linear Chr1 (linChr1) copy that is selected in a two-step process. At time T1, we detected a linChr1 (~1.9 Mbp) containing an intact right arm and a truncated left arm (Fig 6B). At the 5' breakpoint, Chr7 subtelomeric and telomeric regions are fused to Chr1. This chromosomal fusion occurs within the internal TLOα34 gene deleting~1.3 Mbp. Sequence homology between TLOα34 and the subtelomeric TLOγ16 gene is likely to have guided the fusion between Chr1 and Chr7. At T2, linChr1 is further processed at its right arm by deletion of~0. 6 Mbp and the addition of telomeric repeats (Fig 6B). The 3' breakpoint contains a microhomology tract (6 bp 5'-TTCTTG-3') between internal sequences of Chr1 and telomeric repeats. The resulting linChr1 spans the centromere and is flanked by terminal telomeric repeats. To assess whether linChr1 was necessary for the ulp2Δ/Δ phenotypic rescue, we passaged FLC-1 in N/S media and selected three independent phenotypic revertants (R-1, R-2 and R-3) that, similarly to the ulp2Δ/Δ strain, are less able to withstand FLC and form wrinkled colonies on solid media (Fig  6C and 6D). Diagnostic PCR analysis with primers specific for linChr1 indicates that linChr1

PLOS GENETICS
was lost in the R-1, R-2 and R-3 revertants, while an amplification product was detected in DNA isolated from FLC-1 (Fig 6E). This result suggests that linChr1 has an adaptive value.
Finally, we discovered that the ChrRR-deletion is selected early following FLC treatment as this deletion is present at timepoint T1 and T2 (Fig 6A). A microhomology tract between the RDN18 (encoding for the 18S rRNA) genes and telomeric repeats facilitated the addition of telomeric repeats to one copy of the RDN18 gene stabilising the truncated chromosome ( Fig  6E). Thus, rearrangements at microhomology regions and DNA repeats guide the formation of adaptive segmental aneuploidies via two temporally separated events.

Discussion
There is a significant gap in our understanding of how, in microbial organisms, genome instability leads to increased fitness in stress and non-stress environments. Here, we identified the SUMO protease Ulp2 as a key protein ensuring genome stability in C. albicans. We demonstrated that ULP2 loss leads to enhanced genome variation and that dysregulation of the SUMO system combined with drug treatment facilitates the selection of adaptive segmental aneuploidies via a two-step process.

Ulp2 is a critical regulator of C. albicans genome stability
SUMOylation is a post-translational protein modification signalling in a large number of cellular processes by targeting nuclear eukaryotic proteins [71][72][73][74]. SUMO peptides are covalently attached to target proteins by the concerted action of E1, E2 and E3 enzymes while SUMOylation is reversed by SUMO-specific proteases [75][76][77][78][79]. We discovered that the SUMO protease Ulp2 promotes genome stability in C. albicans. These findings are consistent with the emerging role of SUMO proteases as a guardian of genome integrity across eukaryotes. Indeed, SUMO proteases ensure genome stability throughout eukaryotes [68,80]. We hypothesise that C. albicans ULP2 promotes genome stability by modulating SUMO levels of several substrates including: (i) kinetochore and centromere-associated proteins, (ii) the DNA replication machinery and (iii) factors involved in DNA repair. Indeed, it is well established that SUMO homeostasis modulates kinetochore function, DNA replication and DNA repair and that defects in these pathways lead to exacerbated genome instability [77,78,81,82]. For example, several S. cerevisiae kinetochore and centromere-associated proteins, including the centromeric-specific histone H3 variant Cse4 CENP-A , are SUMOylated and the knockdown of SENP6, the human ortholog of Ulp2, leads to mis-localisation of the inner kinetochore complex CENP-H/I/K causing chromosome segregation defects [74,[83][84][85]. Furthermore, SUMOylation of replication and repair proteins increases upon DNA damage [74,83,[86][87][88][89]. We still know very little about SUMOylation effect on C. albicans biology and its adaptation to hostile host environments. However, the observation that C. albicans protein SUMOylation patterns are different in normal and stress growth conditions agrees with our data and suggests that this post-translation modification has a critical role in adaption [90].

The adaptive power of segmental aneuploidy to overcome multiple stresses
It is well established that exposure to moderate stress can increase tolerance to unrelated stresses [91,92]. This increased tolerance is usually the result of coordinated gene expression changes known as the core stress response [91]. In contrast, we show that exposure to a stress (antifungal drug) can increase tolerance to unrelated stress (loss of ULP2) by selecting segmental aneuploidies. We demonstrate that two different segmental aneuploidies can co-exist when ulp2Δ/Δ cells are challenged with FLC: an amplification of a Chr1 fragment via the formation of an extra chromosome (linChr1) and a partial deletion of ChrR. We posit that these changes in karyotype provide a synergistic fitness advantage in response to the two independent stressors (the presence of FLC and lack of ULP2) by simultaneously changing the copy number of multiple genes. Indeed, linChr1 amplifies 535 protein-coding genes (S4 Table) and GO analyses demonstrated that 41 of those genes are associated with a "response to drug" phenotype (S6 Table). Among these, amplification of UPC2 encoding for the Upc2 transcription factor is likely to be critical. Indeed, it is well established that UPC2 overexpression leads to FLC resistance by ERG11 transcriptional upregulation [93,94]. Similarly, amplification of CCR4 and NOT5 (part of linChr1) might play a key role in rescuing the fitness defects of the ulp2Δ/Δ strain. Ccr4 and Not5 are subunits of the evolutionarily conserved Ccr4-Not complex that modulate gene expression at multiple levels, including transcription initiation, elongation, deadenylation and mRNA degradation [95]. It has been shown that S. cerevisiae CCR4 and NOT5 overexpression rescues the lethal defects associated with a ulp2 deletion strain [68]. Similarly, GO analysis revealed that ChrRR-Deletion leads to a reduced copy number of 34/204 genes associated with "response to stress" and 18/204 genes are linked to "response to drug" (S7 Table).

Selection of segmental aneuploidies via a two-step process involving microhomology tracts
One key question is to understand how complex novel genotypes are selected in C. albicans. We provide evidence that two temporally separated events led to the formation of linChr1 via regions with (micro)homology to telomeric and subtelomeric regions. Our findings support a model in which non-allelic homologous recombination (NAHR) between the TLOα34 (Chr1) gene and Chr7 subtelomeric sequences resulted in a first deletion on Chr1 that is stabilised by the addition of Chr7 telomeric regions. We propose that, in a second step, microhomologymediated break-induced replication (MMBIR) involving a microhomology region (6 bp 5'-TTCTTG-3') between internal sequences of Chr1 (position Chr1: 2594815-2594820 bp) and the 23 bp telomeric repeats caused the second Chr1 deletion and addition of telomeric repeats. We hypothesise that MMBIR is also responsible for the stabilisation of ChrRR-deletion. Indeed, a microhomology tract (4 bp AxGG) between the RDN18 regions and telomeric repeats is found at ChrRR-deletion breakpoint resulting in the addition of telomeric repeats at the rDNA locus and subsequent stabilisation of the broken chromosome. As MMBIR is caused by stalled replication forks [96], it is likely that defects in DNA replication trigger MMBIR and aneuploidy selection. We suspect that these mechanistic pathways are common in C. albicans and other fungal pathogens. Indeed, complex aneuploidies have been observed in other stress conditions [30, 97,98]. Complex novel karyotypes are most likely the result of independent events that accumulate over time. Furthermore, CNVs downstream of the rDNA locus have been described in C. albicans clinical isolates as well as in cells treated with the antifungal posaconazole [16,30,97]. Analysis of these genotypes by long-read sequencing will be instrumental in fully resolving these karyotypes and unveiling their origins.

Genetic screening
The genetic screening was performed using a C. albicans homozygous deletion library [36] arrayed in 96 colony format on YPD plates (145x20 mm) using a replica plater (Sigma #R2508). Control non-selective (N/S) plates were grown at 30˚C for 48 hours. UV treatment was performed using UVitec (Cambridge) with power density of 7.5 μW/cm 2 (0.030 J for 4 seconds). Following UV treatment, plates were incubated in the dark at 30˚C for 48 hours. For MMS treatment, the library was spotted onto YPD plates (145x20 mm) containing 0.005% MMS and incubated at 30˚C for 48 hours. UV and/or MMS sensitivity of selected strains was confirmed by serial dilution assays in control (YPD) and stress (UV: power density of 7.5 μW/ cm 2 , MMS: 0.005%) plates. Correct gene deletions were confirmed by PCR using gene-specific primers (S9 Table).

Yeast strain construction
Integration and deletion of genes were performed by transforming PCR products containing a marker gene and the appropriate target-gene sequence integration site [99]. Oligonucleotides and plasmids used for strain construction are listed in S9 and S10 Tables, respectively. For Lithium Acetate transformation, overnight liquid yeast cultures were diluted in fresh YPD and grown to an OD 600 of 1.3. Cells were harvested by centrifugation and washed once with dH 2 O and once with SORB solution (100 mM Lithium acetate, 10 mM Tris-HCl pH 7.5, 1 mM EDTA pH 7.5/8, 1M sorbitol; pH 8). The pellet was resuspended in SORB solution containing single-stranded carrier DNA (Sigma-Aldrich) and stored -80˚C in 50 μl aliquots. Frozen competent cells were defrosted on ice, mixed with 5 μL of PCR product and 300 μL PEG solution (100 mM Lithium acetate, 10 mM Tris-HCl pH 7.5, 1 mM EDTA pH 8, 40% PEG4000). Following incubation for 21-24 hours at 30˚C, cells were heat-shocked at 44˚C for 15 minutes and grown in 5 mL YPD liquid for 6 hours before plating on selective media at 30˚C.

UV survival quantification
Following dilution of overnight liquid cultures, 500 cells were plated in YPD control plates and 1500 cells were plated in YPD stress plates and UV irradiated with a power density of 7.5 μW/cm 2 (0.030 J for 4 seconds). Plates were incubated at 30˚C for 48 hours in the dark. Colonies were counted using a colony counter (Stuart Scientific). Experiments were performed in 5 biological replicates, and violin plots were generated using R and R Studio IDE (http://www.r-project.org/).

Growth curve
Overnight liquid cultures were diluted to 60 cells/μL in 100 μL YPD and incubated at 30˚C in a 96 well plate (Cellstar, #655180) with double orbital agitation of 400 rpm using a BMG Labtech SPECTROstar nanoplate reader for 48 hours. When indicated, YPD media was supplemented with MMS (0.005%) and HU (22 mM). Graphs show the mean of 3 biological replicates, error bars show the standard deviation.

Serial dilution assay
Overnight liquid cultures were diluted to an OD 600 of 4, serially diluted 1:5 and spotted into agar plates with and without indicated additives using a replica plater (Sigma Aldrich, #R2383). Images of the plates were taken using Syngene GBox Chemi XX6 Gel imaging system. Experiments were performed in 3 biological replicates.

URA3 + marker loss quantification
Strains were first streaked onto synthetic solid media lacking uracil and uridine (SC-Uri) to ensure the selection of cells carrying the URA3 + marker gene. Parallel liquid cultures were grown for 16 hours at 30˚C in YPD and plated on SC plates containing 1 mg/ml 5-FOA (5-fluorotic acid; Sigma) and on N/S SC plates. Colonies were counted after 2 days of growth at 30˚C. The frequency of the URA3 + marker loss was calculated using the formula F = m/M, where m represents the median number of colonies obtained on 5-FOA medium (corrected by the dilution factor used and the fraction of culture plated) and M the average number of colonies obtained on YPD (corrected by the dilution factor used and the fraction of culture plated) [63]. Statistical differences between results from samples were calculated using the Kruskal-Wallis test and the Mann-Whitney U test for post hoc analysis. Statistical analysis was performed and violin plots were generated using R Studio (http://www.r-project.org/). Microscopy 30 ml of yeast cultures (OD 600 = 1) grown in SC were centrifuged at 1550 x g for 5 minute and washed once with dH 2 O. Cells were fixed in 10 ml of 3.7% paraformaldehyde (Sigma #F8775) for 15 minutes, washed twice with 10 ml of KPO 4 /Sorbitol (100 mM KPO 4 , 1.2 M Sorbitol) and resuspended in 250 μl PBS containing 10 μg of DAPI. Cells were then sonicated and resuspended in a 1% low melting point agarose (Sigma Aldrich) before mounting under a 22 mm coverslip of 0,17 μm thickness. Samples were imaged on a Zeiss LSM 880 Airyscan with a 63x/1.4NA oil objective. Airyscan images were taken with a relative pinhole diameter of 0.2 AU (airy unit) for maximal resolution and reduced noise. GFP was imaged with a 488 nm Argon laser and 495-550 nm bandpass excitation filter. The DAPI channel was imaged on a PMT with standard pinhole of 1AU and brightfield images were captured on the trans-PMT with the same excitation laser of 405 nm. DAPI and brightfield images were taken with the same pixel size and bit depth (16bit) as the airyscan images. Images were of a 42.7x42.7μm field of view with a 33 nm pixel size resolution. z-stacks were taken containing cells of z interval of 500 nm. Airyscan Veena filtering was performed with the inbuilt algorithms of Zeiss Zen Black 2.3. Experiments were performed in 3 biological replicates and >100 cells/replicate were counted.

Drug selection
For fluconazole selection, strains were incubated overnight in Casitone liquid media at 30˚C with shaking. 10 4 cells were plated in a small plate (10 cm) containing Casitone medium plus 256μl DMSO or 128 μg/mL fluconazole. Plates were incubated at 30˚C for 7 days. Colonies able to grow on fluconazole-were streaked (2X) on non-selective (N/S) plates and tested by spotting assay in Casitone +DMSO, or Casitone+FLC. For selection of FLC-1 revertants, 100 cells were plated in YPD agar plates and single colonies were assessed for their ability to grow on casitone medium plus 256μl DMSO or 128 μg/mL fluconazole by serial dilution assays.

Contour-clamped homogeneous electric field (CHEF) electrophoresis
Intact yeast chromosomal DNA was prepared as previously described [101]. Briefly, cells were grown overnight, and a volume equivalent to an OD 600 of 6 was washed in 50 mM EDTA and resuspended in 20 μl of 10 mg/ml Zymolyase 100T (Amsbio #120493-1) and 300 μl of 1% Low Melt agarose (Biorad # 1613112) in 100 mM EDTA. Chromosomes were separated on a 1% Megabase agarose gel (Bio-Rad) in 0.5X TBE using a CHEF DRII apparatus. Run conditions were as follows: 60-120s switch at 6 V/cm for 12 hours followed by a 120-300s switch at 4.5 V/ cm for 26 hours at 14˚C. The gel was stained in 0.5x TBE with ethidium bromide (0.5 μg/ml) for 60 minutes and destained in water for 30 minutes. Chromosomes were visualised using a Syngene GBox Chemi XX6 gel imaging system.

Read depth and breakpoint analysis of short-reads sequencing
Whole-genome sequencing data were analysed for copy number and allele ratio changes as previously described [19,30]. Aneuploidies were visualised using the Yeast Mapping Analysis Pipeline (YMAP, v1.0) [109]. BAM files aligned to the SC5314 reference genome as described above were uploaded to YMAP and read depth was determined and plotted as a function of chromosome position, using the inbuilt SC5314 A21 reference genome and haplotype map. Read depth was corrected for both chromosome-end bias and GC-content. The GBrowse CNV track and allele ratio track identified regions of interest for CNV and LOH breakpoints, and more precise breakpoints were determined visually using IGV. LOH breakpoints are reported as the first and last informative homozygous position in a region that is heterozygous in the parental genome. CNV breakpoints were identified as described previously [19,30].

Long-read DNA sequencing
Oxford Nanopore Technologies (ONT) sequencing data have been deposited in the Sequence Read Archive under BioProject PRJNA879282. DNA was extracted from an overnight culture in YPD using the QIAGEN genomic tip 100/G kit (Qiagen, #10243) according to manufacturing protocol. Long read sequencing libraries were prepared using the SQK-LSK109 Ligation Sequencing Kit with the EXP-NBD104 Native Barcoding Kit (Oxford Nanopore Technologies) from approximately 1μg of high molecular weight genomic DNA, following the manufacturer's protocol. Long read libraries were sequenced on R9.4.1 Spot-On Flow cells (FLO-MIN106) using the GridION X5 platform set to super accurate base calling.

S1 Fig. Genomic variants are selected in ulp2Δ/Δ cells challenged with Fluconazole. (A)
Whole genome sequence data were plotted as the log2 ratio and converted to chromosome copy number (y-axis, 1-4 copies) as a function of chromosome position (x-axis, Chr1-ChrR) using YMAP. Heterozygous (AB) regions are indicated with gray shading and homozygous regions are indicated by haplotype AA (cyan) or BB (magenta). Allele ratio changes that occur within a CNV are indicated as dark blue (AAB) or purple (ABB). Colony B and C had allele ratio colouring that was corrected using IGV and allele frequency information.