Partner Choice in Spontaneous Mitotic Recombination in Wild Type and Homologous Recombination Mutants of Candida albicans

Candida albicans, the most common fungal pathogen, is a diploid with a genome that is rich in repeats and has high levels of heterozygosity. To study the role of different recombination pathways on direct-repeat recombination, we replaced either allele of the RAD52 gene (Chr6) with the URA-blaster cassette (hisG-URA3-hisG), measured rates of URA3 loss as resistance to 5-fluoroorotic acid (5FOAR) and used CHEF Southern hybridization and SNP-RFLP analysis to identify recombination mechanisms and their frequency in wildtype and recombination mutants. FOAR rates varied little across different strain backgrounds. In contrast, the type and frequency of mechanisms underlying direct repeat recombination varied greatly. For example, wildtype, rad59 and lig4 strains all displayed a bias for URA3 loss via pop-out/deletion vs. inter-homolog recombination and this bias was reduced in rad51 mutants. In addition, in rad51-derived 5FOAR strains direct repeat recombination was associated with ectopic translocation (5%), chromosome loss/truncation (14%) and inter-homolog recombination (6%). In the absence of RAD52, URA3 loss was mostly due to chromosome loss and truncation (80–90%), and the bias of retained allele frequency points to the presence of a recessive lethal allele on Chr6B. However, a few single-strand annealing (SSA)-like events were identified and these were independent of either Rad59 or Lig4. Finally, the specific sizes of Chr6 truncations suggest that the inserted URA-blaster could represent a fragile site.

During normal cell proliferation, spontaneous DNA lesions arise at measurable rates and their frequency is significantly increased by the presence of environmental compounds generally referred to as genotoxins. For instance, humans are estimated to generate up to10 5 mutations/cell/day (Hoeijmakers 2009). To repair DNA lesions, cells have evolved a variety of mechanisms that remove damage and accurately restore genetic information (Boiteux and Jinks-Robertson 2013;Wyrick and Roberts 2015). However, repair may also cause genomic rearrangements whose location and frequency are influenced by the genome structure, particularly by the presence of repetitive elements (Chan and Kolodner 2011). Repeated copies of DNA segments are potential targets for homologous recombination (HR) if resection of double strand breaks (DSB) exposes the complementary sequences (Aguilera et al. 2000;Prado et al. 2003;Heyer et al. 2010;Symington et al. 2014).
Single-strand annealing (SSA) plays a major role in direct-repeat recombination resulting in the loss of one repeat and the intervening sequence (Klein et al. 2019). Studies in haploid Saccharomyces cerevisiae (S. cerevisiae) strains on DSB-induced repeat recombination have shown that SSA was dependent on the annealing activity of Rad52 for repeat length of 1-2 kb (Rudin and Haber 1988;Fishman-Lobell et al. 1992;Sugawara and Haber 1992;Jablonovich et al. 1999) but not when repeats were much larger (e.g., CUP1 gene or rRNA gene arrays) (Ozenberger and Roeder 1991). Additional work revealed that this process was significantly impaired in the absence of RAD59, a RAD52 paralog, especially when the direct repeats were short (40-fold for 205 bp repeats) (Petukhova et al. 1999;Sugawara et al. 2000;Symington 2001, 2003;Wu et al. 2006;Pannunzio et al. 2008). In the presence of Rad52 and Rad59, DSB-induced SSA utilized repeats as short as 29 bp and showed linear dependency on the length of homologous repeats up to 415 bp (Ivanov et al. 1996).
Non-DSB direct-repeat recombination (spontaneous) via SSA-like mechanisms can also lead to loss of one repeat plus the intervening sequence. In S. cerevisiae, the rate of spontaneous direct-repeat recombination (not DSB-induced) was directly proportional to the substrate length and the minimal repeat length for efficient recombination was 285 bp; some recombination was detected for 80 bp repeats but not for 37 bp repeats (Jinks-Robertson et al. 1993). This suggests the existence of specific differences between DSB-induced and spontaneous directrepeat recombination via SSA. Importantly, SSA does not require strand invasion and is therefore independent of Rad51 (and its paralogs Rad55 and Rad57) (Ivanov et al. 1996;Jablonovich et al. 1999;Pannunzio et al. 2008).
The genome of C. albicans, the most common fungal pathogen, is particularly rich in direct repeats (Braun et al. 2005;Smichd et al. 2012;Todd et al. 2019). Not much is known in C. albicans about the recombination pathways involved in repeat number alteration and the potential consequences for overall genome structure and host-fungus interactions. It is believed that repeat number alterations are caused by replication slippage and recombination and may provide an evolutionary advantage in fluctuating environments thereby providing the population with a selection of proteins with different properties. Not only may these mechanisms alter repeat numbers and generating novel alleles of a specific ORF, recombination between repeats of two genes from the same family (i.e., agglutinin-like (ALS) sequence gene family in C. albicans) could lead to chimera formation, which may be endowed with novel properties advantageous for survival in the host (Zhang et al. 2003;Zhao et al. 2011). Several studies have shown that repeats within coding regions of genes may have functional roles. For example, the repeat copy number in ALS5 directly affects adhesion to fibronectin (Rauceo et al. 2006). Repeats of Hwp1, Pir1 and Eap1 are important in adhesion to buccal epithelial cells (Staab et al. 2004), protein localization (Sumita et al. 2005), and positioning of binding sites to several materials and cells, respectively (Li and Palecek 2008). Furthermore, repeat length variation in cell wall-associated proteins may contribute to the overall antigenic variation in C. albicans, which in turn aids in adaptation to and evasion from the host (Verstrepen and Fink 2009;Zhang et al. 2010;Zhao et al. 2011;Zhou et al. 2017).
Here, we took advantage of the URA-Blaster cassette which consists of the URA3 gene of C. albicans flanked by 1.1 kb hisG direct repeats (Alani et al. 1987;Fonzi and Irwin 1993). To study direct-repeat recombination and to test for allele-specific effects, we replaced each allele of RAD52 (located on the left arm of chromosome 6 (Chr6) with this cassette, measure rates of URA3 loss as resistance to 5-fluoroorotic acid (5FOA R ), and then analyzed 5FOA R derivatives by CHEF Southern and SNP-RFLP to determine the underlying genetic events and associated mechanisms (Forche et al. 2011). To assess the role of genes important for homologous recombination of direct repeats, we performed the same analyses in strains lacking RAD52, RAD51, RAD59, and LIG4. We found that URA3 loss in wild type, rad59, and lig4 backgrounds mostly resulted from URA3 pop-outs and to a lesser degree from interhomolog recombination. This bias was maintained in rad51 strains although with a significant reduction in the frequency of URA3 popouts compared to wild type. In rad51 5FOA R derivatives additional URA3 loss mechanisms were identified including chromosome loss and truncation as well as ectopic translocations. Interestingly, rad52 5FOA R derivatives underwent chromosome loss or truncation 85% of the time with interhomolog recombination being absent. The remaining URA3 loss events resulted from SSA-like mechanisms, which were independent of Rad59 and Lig4. As a collateral and unexpected finding, our results support the possibility that the insertion of the URA-blaster into the genome may have resulted in the generation of a slow replication zone and/or fragile site.

MATERIALS AND METHODS
C. albicans strains used in this study Single and double mutant strains used in this work were generated from strain CAI4, a Uraderivative of the reference strain SC5314 (Gillum et al. 1984), by disrupting the indicated allele with the hisG-URA3-hisG cassette flanked by promoter and terminator regions of the target gene (Table S1). Transformants were verified by PCR and/or Southern blot analyses as previously described (Ciudad et al. 2004;Bellido et al. 2015). To isolate 5FOA R derivatives, a single colony from the indicated genetic background was re-isolated on an YPD plate and then streaked on a new YPD plate supplemented with 0.1% (w/v) 5FOA and 25 mg ml -1 uridine, since C. albicans ura3 mutants are fed with uridine. To disrupt RAD52 with the SAT1-flipper cassette, the upstream and downstream regions of the RAD52 ORF were PCR-amplified from genomic DNA of strain CAF2-1, using oligonucleotides RAD52F-ApaI/RAD52R-XhoI and RAD52F-SacII/RAD52R-SacI respectively ( Fig. S1 and Table S2). Amplified fragments were cloned in pSFS2A plasmid flanking the SAT1-flipper cassette. The disruption cassette was released by digestion with ApaI and SacI and transformed into the indicated hemizygous strains RAD52/rad52Δ::hisG-URA3-hisG ( Figure 1) using a MicroPulser Electroporator system (Bio-Rad) (Ciudad et al. 2016). Nourseothricinresistant (Nou R ) colonies were selected on YPD plates supplemented with 200 mg/ml nourseothricin. Several transformants were initially selected based on their thorny colonies and filamentous cell morphology, two phenotypes of null rad52 strains (Andaluz et al. 2006) and then PCR verified for both integration of the SAT1-Flipper cassette in the RAD52 locus (oligonucleotides SAT1F-Flip/RAD52R) and absence of any residual RAD52 allele (oligonucleotides RAD52-IF/RAD52-IR). SAT1 loss was induced by overnight growth in liquid YPM (2% maltose, 1% yeast extract, 2% bactopeptone) (Reuss et al. 2004). The resulting nourseothricin-sensitive (Nou S ) derivatives were selected as small colonies on YPD plates supplemented with 20 mg/ml nourseothricin. They were verified by PCR for SAT1 loss (oligonucleotides RAD52-F and RAD52-R). These strains carry the rad52::FRT allele (FRT strains) (Fig. S1).

DNA extraction and analysis
Extraction of genomic DNA, preparation of chromosomes, and CHEF Southern hybridization have been described (Andaluz et al. 2011). Two different PFGE protocols were used. In the first protocol (short run), all chromosomes were separated. The second protocol separates both homologs of Chr7 and, in some strains, of Chr6 (Andaluz et al. 2011). To test for the presence of one or both homologs of Chr6 we used the SNP status (genotype) of multiple markers along chromosomes as proxy. Routine SNP-RFLPs analyses were carried out as described (Forche et al. 2009) using the indicated primers (Table S2).
Generation, verification and characterization of Chr6A and Chr6B tester strains In the strain background used in this study, Chr6 homologs exhibit size differences sufficient for separation on CHEF gels. Chr6 homolog length polymorphisms can be due to differences in the number of repeats either within the major repeat sequence (MRS) (Chibana and Magee 2009) or within members of the ALS family (ALS6, ALS1, ALS10, ALS5, and ALS2) located on this chromosome (Zhang et al. 2003;Zhao et al. 2011).
We used strains heterozygous for RAD52 and rad52 null strains to generate tester strains with the hisG-URA3-hisG construct either replacing RAD52 on the A or the B homolog (Table S1). CAGL4A and CAGL4.1A are two independent rad52 Ura + derivatives of the heterozygous parental, CAGL1B (Ciudad et al. 2004). We have previously shown that CAGL4A and CAGL4A.1 conserved both homologs of Chr6 (Andaluz et al. 2011). To identify the test homolog in CAGL4A and CAGL4A.1, we first performed a physical analysis of Chr6 homologs present in its parental heterozygous strain CAGL1B.1 (Ura -). CHEF Southern hybridization with a COX12 probe confirmed the presence of both Chr6 homologs in CAF2-1, CAI4, and CAGL1B.1, and hybridization with RAD52 and hisG probes localized RAD52 to the smaller Chr6 homolog (Chr6A) and rad52::hisG to the larger homolog (Chr6B) (Fig. S3A). In agreement with this, a spontaneous Hisderivative (GLH1-7) of a rad52 strain (TCR2.1.1) disomic for Chr6 only carried the small homolog and was homozygous, haplotype A, for multiple Chr6 SNPs markers (Forche et al. 2009;Andaluz et al. 2011).
We also took advantage of heterozygosity within the RAD52 ORF to identify the RAD52 allele present in each heterozygote using SNP/ RFLP. A 793 bp region of the RAD52 ORF was amplified with primers RAD52_501F and RAD52_1290R (Table S2) and subjected to a restriction digest with TaqI. This enzyme cuts twice in allele A (RAD52A) yielding 3 restriction fragments (251 bp, 237 bp, 305 bp) and once in allele B (RAD52B) resulting in 2 restriction fragments (251 bp and 542 bp) (Fig. S3B). As expected, both alleles were detected in strain CAI4 (as well as in parental strains SC5314 and CAF2-1, not shown) whereas CAGL1B was homozygous for RAD52A (Fig. S3B). We concluded that during the generation of CAGL4A and CAGL4A.1, the RAD52B allele present in the larger Chr6 (Chr6B) of CAI4 strain was disrupted first resulting in the intermediate strain CAGL1B (test chromosome B) (Fig. S3, top). Because of previous findings that the Chr6B allele may harbor recessive lethal alleles (and therefore cannot be lost) (Andaluz et al. 2011;Hickman et al. 2013;Feri et al. 2016), new strains were generated with the URA-Blaster inserted carrying Chr6A as the test chromosome (CAGL1A). These strains were used to generate rad52::hisG strains with the opposite configuration, i.e., if derived from CAGL1A, test chromosome was Chr6B, or rad52::FRT strains that conserved the parental configuration, i.e., if derived from CAGL1A, Chr6A remained as test chromosome (Figure 1, Table S1). All heterozygous and null RAD52 strains were tested for the presence of both Chr6 homologs by SNP RFLP (SNP122, SNP123 and SNP132) and for the lack of obvious GCR by PFGE (Fig. S4). We concluded that all of them were appropriate for the generation and subsequent genetic analysis of the 5FOA R derivatives.

Fluctuation test
Strains were streaked to single colonies on YPD and incubated for 2 -4 days at 30°. At least 10 independent colonies from each strain were resuspended in 100 ml of sterile water. Tenfold dilutions were generated using 10 ml of the initial resuspension and 40 ml of the 10 24 dilution were spotted onto YPD plates to determine the total amount of CFUs. The remaining 90 ml of the initial resuspension was spread onto 5FOA plates. Alternatively, fluctuation analysis using twenty overnight (16 h) liquid cultures seeded with single colonies was done as described by Forche et al. (2011). Importantly, for wild type strain CAGL1B, URA3 loss rates (5FOA R ) were similar for both methods (1.5 · 10 25 /cell generation for colonies vs. 2.5 · 10 25 /cell generation for liquid cultures). Therefore, fluctuation analyses were carried out using the former protocol. YPD and 5FOA plates were incubated at 30°for 3 days and colonies were counted. URA3 loss rates were calculated as described (Forche et al. 2011).

Molecular characterization of URA3 loss in 5FOA R derivatives
For most strains, a minimum of 20 5FOA R derivatives per strain background were analyzed. A scheme with the several steps for characterization of the 5FOA R derivatives at the RAD52 locus is shown in Fig. S4. SNP results are summarized in Table S3.
We used S. cerevisiae chromosomal markers to determine the size of SNCs (Argueso et al. 2008). The calculated size correlated well with the genotypes of markers snp122 and snp132, which are 832 kb and 545 kb away from the right telomere, respectively. SNCs from strains heterozygous for snp122 should be larger than 832 kb, whereas SNCs from strains homozygous for both snp122 and snp123 should be smaller than 545 kb. Importantly, all strains carrying SNCs were heterozygous for snp123 marker, an indication that no SNC was smaller than 545 kb ( Figure 1). Fisher's exact test was used to determine whether the frequency of different loss mechanisms in the mutant strains vs. wild type were significant (p value of , 0.05).
The occurrence of SSA at the SHE9 locus was investigated by PCR using primers SHE1 and SHE2, which amplify bands of 845 bp and 1171 bp for SHE9 and hisG repeat respectively, whereas the presence of URA3 in 5FOA R segregants (URA3 mutational inactivation) locus was verified using primers SHE1 and URA3det-R that amplify band of 1.3 kb (Table S2).

Data availability
Strains and plasmids are available upon request. The authors state that all data necessary for confirming the conclusions presented in the article are represented fully within the article. Supplemental material available at FigShare: https://doi.org/10.25387/g3.8796686.

Experimental system
A diagram showing the approach used to generate tester strains with the hisG-URA3-hisG cassette is shown in Figure 1A (right side upper branches). We used the RAD52 locus (Chr6L, left arm of Chr6, coordinates 97,421 to 95,727; see Figure 1B) to determine URA3 loss rates in Rad52 + strains (RAD52/rad52::hisG-URA3-hisG; wild type in the context of this study) because it also allows the analysis of rad52 null mutants (rad52::hisG/rad52:hisG-URA3-hisG), which are refractory to targeted gene replacement (Ciudad et al. 2004). 5FOA R derivatives from Rad52 + strains can arise through events shown in Figure 2 (Hiraoka et al. 2000;Cauwood et al. 2013). The strategy used to identify those events is summarized in Fig. S4. SNP-RFLP analysis allowed us to delimit the genomic region where genetic events responsible for the RAD52 genotype had occurred. When all SNP markers (snp122, 123 and 132) were homozygous for the same haplotype, the strain was considered having undergone a chromosome loss event ( Figure 2D) (Legrand et al. 2008;Forche et al. 2011). Truncation of the test chromosome (i.e., the chromosome carrying the URA-Blaster) ( Figure 2E) results in chromosome fragments detectable by pulsed-field gel electrophoresis (PFGE) and in hemizygosity of RAD52 and genes between RAD52 and the left telomere.
URA3 loss rates in wild type strains and in strains defective for recombination We determined URA3 loss rates and associated recombination mechanisms for wild type (RAD52 het, Ura + ) and for strains deleted for RAD59, LIG4, RAD52 or RAD51. To further limit the possibility for single strand annealing to occur, double mutants rad59 rad52 and lig4 rad52 were also analyzed. Importantly, for each genetic background (except for lig4 derivatives), two tester strains carrying the URA-Blaster on either allele of Chr6 at the RAD52 locus (Chr6A, tester A and Chr6B, tester B) were analyzed ( Figure 1 and Table S1).
URA3 loss mechanisms in wild type, rad59 and lig4 strains Next, we examined the nature of genetic alterations associated with URA loss using CHEF Southerns and SNP-RFLP analysis (Andaluz et al. 2011;Forche et al. 2011). PCR of the RAD52 locus showed that 5FOA R in strains derived from wild type CAGL1A resulted from URA3 deletion (95.5%, 42/44) and interhomolog recombination (4.5%, 2/44), which was similar for CAGL1B 5FOA R derivatives (Table 1). Karyotype analysis did not identify any gross chromosomal rearrangement (GCR) (Fig. S6) and SNP-RFLP analysis showed homozygosis for snp122 and snp123 for 5FOA R derivatives that resulted from interhomolog recombination and snp132 remained heterozygous (Table S3). This suggests that crossover/BIR between cen6 and snp123 led to 5FOA R in these derivatives and that both homologs of Chr6 were retained. Taken together, our data for the RAD52 het strain background show that, irrespective of the Chr6 homolog used as the tester allele, the formation of 5FOA R derivatives is rarely accompanied by chromosome loss or truncation and a strong bias exists for URA3 deletion vs. interhomolog recombination/other events.
Deletion of RAD59 or LIG4 in wild type strain background did not alter URA3 loss mechanisms. For strains CAGL2A and CAGL2B (Rad52 het, rad59DD, Table 1), the URA3 pop-out bias (88%) was similar to the related wild type strains (Fisher's exact test, P = 0.665, tester allele A, and P = 0.314, tester allele B) (Tables  1 and S3). However, three 5FOA R derivatives from strain CAGL2B showed supernumerary chromosomes (SNC) unrelated to Chr6 (two strains are shown in Fig. S7), suggesting that the absence of Rad59 does not alter frequencies of URA3 pop-out but may cause genetic instability. Similarly, a lig4 strain (Andaluz et al. 2001a;Sinha et al. 2016) carrying the URA-Blaster on Chr6A (CAGL01) did not show GCRs, and all 5FOA R derivatives resulted from URA3 deletion (Tables 1 and S3) suggesting that microhomology mediated end-joining does not contribute either to the observed URA3 popout bias in wild type.
Chromosome loss and truncation are the prevalent mechanisms leading to 5FOA R in a rad52 strain background Because C. albicans strains lacking RAD52 are intrinsically unstable (Andaluz et al. 2011), we first compared two independent, isogenic rad52 mutants (CAGL4A and CAGL4.1A) (Material and Methods, Table S1) that carry the URA-Blaster on Chr6A. CHEF Southerns indicated that 90% of 5FOA R derivatives of both strains showed novel SNCs between 815 and 945 kb ( Figure 4, Table 1) that hybridized to a COX12 probe (located on Chr6L, Figure 1) suggesting that these were Chr6 truncations. One CAGL4A derivative (CAGL4A-5) acquired a URA3 loss of function mutation (Figure 4, Table S3) and, quite strikingly, one CAGL4.1A derivative (CAGL4.1A-1) showed a wild type genotype except for the absence of URA3 ( Figure 4A and C) (see also below).
In contrast to our observations for CAGL4A and CAGL4.1A, chromosome loss was abundant in 5FOA R derivatives from strain CAGL4B: 6/20 (30%) showed LOH of all three SNP markers (Tables 1  and S3). All other derivatives (14/20; 70%) remained heterozygous for at least one SNP marker. CHEF Southerns using a COX12 probe showed that 71% (10/14) of them have Chr6-derived SNCs larger than 813 kb (Tables 1 and S3; Figs. S8 and S9). Importantly, as shown for strain CAGL4.1A-1, the remaining four CAGL4B derivatives did not show Chr6 size changes and remained heterozygous for all three SNP markers (Fig. S8B, lanes 9, 10, 12 and 16). In addition, PCR exclusively amplified the hisG repeat suggesting URA3 deletion ( Figure 5). Furthermore, and consistent with the heterozygosity of SNP makers, Chr6 bands were brighter and broader on CHEF Southern blots, as expected when two homologs with slightly different sizes are present, compared to less bright, single Chr6 homologs that remain after chromosome loss or truncation (i.e., compare lanes 1 and 2 in Figure 4C). These observations support the possibility that in contrast to S. cerevisiae (Haber and Hearn 1985;Ozenberger and Roeder 1991;Jablonovich et al. 1999;Paques and Haber 1999;Sugawara et al. 2000), homology-dependent recombination (SSA or interhomolog recombination) using repeats , 2 kb may occur at measurable rates (10 26 -10 27 /cell generation) in rad52 C. albicans strains. However, the frequency of these events was significantly lower than in wild type (P = 0.00005).
Although the absence of either Rad59 or Lig4 in wild type cells did not alter the ratio of URA3 pop-out vs. interhomolog recombination (Table 1), one could argue that either protein could have facilitated the putative URA3 pop-outs among 5FOA R derivatives in rad52 strains (Figures 4 and S8). Two independent rad52 rad59 double mutants (CAGL5A and CAGL5B) (Table 1) exhibited similar frequencies of chromosome loss, chromosome truncation and URA3 pop-out compared to single rad52 mutants (CAGL4A/4.1A and CAGL4B) (P = 0.57 for URA3 pop-outs), with chromosome loss only in 5FOA R derivatives from CAGL5B (Table 1, Fig. S10). In addition, the sizes of Chr6 SNCs were similar (compare Figures 4 and S8   Overview of possible mechanisms leading to inactivation of URA3 in a RAD52/rad52::hisG-URA3-hisG strain. Homolog A is shown in cyan and homolog B is shown is magenta. The resulting Uraderivatives (bottom row) are selected on 5FOA. Viable progeny (only Uraderivatives can grow on 5-FOA) is indicated with an asterisk. Events can occur in G1 phase of the cell cycle (top row; two homologs) or in G2 (second and third row; both chromatids are maintained together by the centromere), but only G2 events are shown, as follow: pop-out of URA3 and one copy of the two hisG repeats, which can occur via a single-strand annealing-like mechanism involving spontaneous intrachromatid direct-repeat recombination, intrachromosome or intra-chromatid crossover, or microhomology-mediated end joining (A); unequal sister chromatid exchange (B); inter-homolog recombination including crossover, break-induced replication (BIR) (C), or gene conversion (schematic not shown); ectopic recombination (schematic not shown); chromosome loss (D); chromosome truncation (E); and mutational inactivation of URA3 (F). Green line: Rad52; black line, hisG; gray line, URA3. Note that gene conversion (GC) without crossover at the RAD52 locus in G1 or G2 is also possible, but the absence of heterozygosities between RAD52 and the left telomere prevents its detection.
Overall, and similar to 5FOA R rad52 derivatives, URA3 pop-out strains 1) did not show SNCs, 2) remained heterozygous for all three SNPs, 3) conserved both Chr6 homologs (snp132 heterozygous and broader Chr6 bands on PFGE/Southern blots for most of them), and 4) PCR confirmed the presence of an unaltered hisG module and the absence of URA3 (Table S3).
In RAD52 strains, most SSA events are Rad51-independent URA3 pop-outs/deletions generally result from intra-chromatid recombination either via SSA-like mechanisms, intra-chromatid crossover or unequal sister chromatid exchange. Importantly, whereas both mechanisms To test RAD51 dependency of the observed URA3 pop-outs, we deleted RAD51 in the wild type strain background (CAGL3A and CAGL3B) (Table S1) and found that most CAGL3A 5FOA R derivatives still arose via pop-out (35/50) albeit at a decreased frequency (70%) compared to wild type (.90%). The remaining 5FOA R derivatives (15/50, 30%) retained RAD52 only. Of these, four strains likely underwent interhomolog recombination (Tables 1 and S3). CHEF Southerns with a COX12 probe failed to identify Chr6 SNCs among the 5FOA R derivatives from URA3 pop-outs (not shown). In contrast, in derivatives that retained RAD52 only, truncations were abundant with SNCs ranging in size from 813 kb to . 850 kb (Figures 6 and S12 and Table S3).
For strain CAGL3B, 76% (38/50) of 5FOA R derivatives arose from URA3 pop-outs, whereas only two retained RAD52 and remained heterozygous at all 3 SNPs suggesting that they arose via interhomolog recombination (GC or XO/BIR) near the RAD52 locus (Table 1). In contrast to CAGL3A derivatives, Chr6 truncations were not detected and chromosome loss was significantly more frequent (10/50; 20%) (Tables 3 and S3, Figure 6). Importantly, CHEF Southerns with a COX12 probe revealed novel SNCs with sizes larger than Chr6 for derivatives CAGL3A-2, CAGL3A-18, and CAGL3B-7 ( Figure 6, top right). These SNCs also hybridized to a CEN6 probe (Figure 6, bottom right), and it is therefore likely that these resulted from ectopic translocation involving a centromeric fragment of Chr6 and a fragment of a different chromosome ( Figure 6; Table S3). Together, results for 5FOA R derivatives from CAGL3A and CAGL3B show that a similar pop-out bias exists for both tester alleles (Fisher's exact test, P = 0.8299), and that lack of Rad51 significantly n■ For each strain (except CAGL27 and CAGL28) both genotype and test allele (A or B) are indicated. 1 and 2 refers to independent experiments. The number of derivatives analyzed in each experiment is shown in parenthesis. Genetic events are indicated at the top. For strains CAGL1, CAGL2 and CAGL3, 5FOA R derivatives resulting from URA3 pop-out conserved the wild type RAD52 allele and the disrupted rad52::hisG-URA3-hisG allele had been processed to rad52::hisG. Derivatives resulting from IHR carried only the wild type RAD52 allele (2 copies). CL and CT were further confirmed by SNP RFLP analysis and CHEF Southerns. nd-not determined. 5FOA R derivatives from FRT strains were screened for SSA, SNPs GRCs. Strain CAGL5B ÃÃ -FRT was intended to be CAGL5A-FRT (tester A) since it was derived from CAGL2A but behaved as if carrying RAD52B as the test allele. It is likely that a reciprocal exchange between both RAD52 alleles (gene conversion or crossover) occurred at some step during its generation.
decreased the number of the URA3 pop-outs independent of the allele (P = 0.0012, tester allele A; P = 0.0337, tester allele B). An interesting consequence of this observation is that, while most pop-out derivatives in the wild type background were generated by SSA, a few may have formed via intra-chromosome crossover or unequal sister chromatid exchange. Furthermore, the absence of Rad51 did not abolish interhomolog recombination since it was still observed among the RAD52 derivatives.
The majority of SNCs in rad51 5FOA R derivatives are formed by Chr6 truncation followed by telomere addition To determine how SNCs larger than wild type Chr6 arose, we tested the possibility that the presence of hisG on other chromosomes could serve as translocation hotspot leading to ectopic translocation and the formation of larger chimeric chromosomes. For example, the size of one of the two reciprocal translocation products involving the hisG repeats of rad52::hisG (Chr6) and rad51::hisG (ChrR, at $485 kb) would be $1.4 Mb (900 kb from Chr6 plus 485 kb from ChrR), which is close to the size calculated for the ectopic translocation chromosomal bands of 5FOA R derivatives CAGL3A-2, CAGL3A-18, and CAGL3B-7. However, while primers flanking the RAD52 ORF or the RAD51 ORF amplified the expected fragments, PCRs with mixed primer pairs did not amplify any products suggesting that ectopic translocations did not involve hisG repeats. In addition, CHEF Southerns with a hisG probe only hybridized to ChrR in these three strains (Fig. S12, suggesting that the hisG fragment on Chr6 had been lost. Importantly, an HDA1 probe from ChrRL (at $450 kb) failed to co-hybridize with the COX12/CEN6 containing SNCs but hybridized to a novel band of $1.4Mb in CAGL3A-23 ( Figure 6, bottom left), suggesting that this SNC was generated either by an internal deletion on ChrR or by a translocation involving ChrRL and a centric fragment of one of the smaller chromosomes (Chr5, 6, or 7) (see Figure 6, bottom right). Consistent with either possibility, this SNC, although faintly, hybridized to the hisG probe (Fig. S12), suggesting that it could carry sequences of the rad51::hisG allele.

RAD52 dosage does not affect URA3 loss
The wild type strains used in the above experiments (RAD52/ rad52::hisG-URA3-hisG) carry a single copy of RAD52. To investigate whether RAD52 dosage influences the rate and/or distribution Figure 4 Karyotypes of strains CAGL4A and CAGL4.1A (both test Chr6A) and their 5FOA R derivatives. (A) PFGE gel. B and C) separation of smaller Chrs (5-7) followed by Southern blot hybridization using a COX12 probe for derivatives of CAGL4A (B) and CAGL4.1A (C). The event determined for each derivative is indicated at the bottom of the CHEF-Southern. SNCs are marked with white arrowheads. Note that most SNCs were larger than 666 kb and the majority had sizes of $815 kb (see also Fig. S8), which is in agreement with the genotypes of markers snp122 and snps123 in the 5FOA R derivatives (Fig. 1, Table S3).

DISCUSSION
In the present study, we aimed to study the mechanisms involved in direct repeat recombination in C. albicans wild type strains and recombination mutants. We took advantage of the URA-Blaster (inserted at the RAD52 locus on Chr6) to determine rates of direct-repeat recombination measured as URA3 loss (resistance to 5FOA), and we used SNP-RFLP and CHEF-Southern analyses to determine the underlying mechanisms. We found that for the wild type strain (CAI4), URA3 popouts were the major events responsible for 5FOA R whereas other events identified here as interhomolog recombination and URA3 mutational inactivation were much less frequent and independent of the Chr6 allele examined. Importantly, interhomolog recombination resulted in very long LOH tracts in about 5-10% of 5FOA R derivatives, most of them leading to homozygosis of all SNP markers, which is consistent with BIR or reciprocal crossovers (Forche et al. 2011;Cauwood et al. 2013;Symington et al. 2014). It is worth noting that the major repeat sequence on Chr6 is located between snp123 marker and cen6 (Figure 1), the region where most interhomolog recombination occurred, suggesting that it could act as a recombination hotspot in wild type cells (Lephart and Magee 2006;Marton et al. 2019). The possibility that these strains exhibit phenotypes attributable to off-target effects calls for an exhaustive characterization of C. albicans genetically engineered strains involving recombination as recently demonstrated for CAI4 (Ciudad et al. 2016).
In the absence of recombination proteins Rad51 or Rad52 chromosome loss and chromosome truncations were frequent, which supports the existence of selective pressure to maintain a complete Chr6A homolog; its loss likely may result in cell death. This conclusion is consistent with previous results indicating that only one homolog of several chromosomes can be lost or is preferentially lost (Andaluz et al. 2011;Hickman et al. 2013). As a diploid, C. albicans may allow the generation of high levels of heterozygosity including the appearance and persistence of recessive lethal alleles of one or more essential genes on one homolog, as was recently shown for Chr4 and Chr7 (Feri et al. 2016;Marton et al. 2019). Therefore, the nature and relative frequency of events responsible for the loss of URA3 may depend significantly on the homolog used as tester chromosome.
Rates of URA3 loss in mutants with defective homologous recombination We found little variation in URA3 loss rates for single mutants (including rad52) compared to wild type. This is in contrast with the significant decrease in 5-FOA R frequency exhibited by haploid S. cerevisiae rad52 and, to a lesser extent rad59, in a similar assay using URA3 flanked by 2.4 kb-long repeats (Halas et al. 2016). Under these conditions, loss of URA3 via SSA is drastically decreased (rad52) and other mechanisms such as chromosome loss and chromosome truncation may be detrimental. By contrast, a diploid rad52 strain has the potential to become Uraby chromosome loss and chromosome truncation. This is particularly true for C. albicans, whose genome plasticity is well documented (Rustchenko 2007;Selmecki et al. 2009;Forche 2014). Some variation between the URA3 loss rates of the several strains may also derive from the diploid state of C. albicans. For instance, in wild type S. cerevisiae variation in URA3 loss rates from the URA-Blaster inserted at five different loci was intrinsically greater in diploids compared to haploids (Cauwood et al. 2013). The validity of our assay is further Figure 5 PCR products of hisG fragments from rad52 URA3 popout derivatives. Numbers in parentheses identify 5FOA R derivatives for each initial strain.
supported by the observation that URA3 loss rates were locus-and dosage-independent. The significant increase in URA3 loss rate in rad52 rad59 and rad52 lig4 double mutants compared to rad52 single mutants (5.5 and 7.eightfold, respectively) may simply suggest that Rad59 and Lig4 are suppressing the formation of lesions that form in a rad52 background, most likely because they are repairing these lesions. However, regardless of the mechanism(s) involved, it may stem from the additive effect on genetic instability caused by the lack of more than one gene.
In Rad52 + cells, URA3 pop-outs are Rad59-independent and occur through an SSA-Like mechanism The absence of Rad59 did not alter the frequency of SSA in C. albicans. This result contrasts with previous work in S. cerevisiae showing that depletion of Rad59 significantly decreased the formation of the SSA product and the number of survivors regardless of repeat length (Sugawara and Haber 1992;Sugawara et al. 2000). An important difference between both systems is that the S. cerevisiae study (Sugawara et al. 2000) used a haploid strain and created a DSB between the repeats whereas we have determined spontaneous recombination in a diploid cell. However, other studies in diploid S. cerevisiae strains have found that Rad59 is also required for spontaneous SSA-like events of short repeats (Halas et al. 2016). It will be interesting to determine whether direct repeats shorter than hisG (1.1 kb) would affect URA3 loss in the absence of CaRad59.
It should be noted that, unlike wild type, the rad59::hisG alleles (Chr4) in rad59 strains could represent potential sites for ectopic translocation involving sequences of the rad52::hisG-URA3-hisG allele (Chr6). We think this is unlikely because translocations involving Chr6 were not observed among the 84 5FOA R derivatives in the rad59-DD strain background. However, it is possible that hisG-mediated ectopic translocations may occur at rates below the limit of detection. In S. cerevisiae, for example, spontaneous ectopic recombination between Ty1 interspersed direct repeats occurred at a much lower rate (10 28 /cell generation, respectively), which is below the threshold of detection in the system used here (Chan and Kolodner 2011). Figure 6 Karyotypes (top-left) of 5FOA R derivatives from strains CAGL3A (test Chr6A) and CAGL3B (test Chr6B). Only derivatives that did not undergo URA3 loss via pop-out from Expt 2 are shown (see Table S3). Chromosomes were transferred to nitrocellulose paper and hybridized to the COX12 probe (Chr6) (top-right). Nitrocellulose was stripped and re-probed with the HDA1 probe (ChrR) (bottom left). SNCs are indicated with white arrowheads. Events accompanying 5FOA R are indicated at the bottom of the top-right panel. Strains 23 (CAGL3A-derivative) and 6 (CAGL3B-derivative) were likely formed by interhomolog recombination involving Chr6; the former also shows an ectopic translocation event involving ChrR (see Fig. S12) (for a summary of the SNP-RFLP analysis see Table S3). Strains showing SNCs larger than Chr6 when probed with COX12 and HDA1 (2, 18, 23 from CAGL3A and 7 from CAGL3B) were further analyzed by CHEF Southern using a cen6 probe. Only derivatives 2, 18 and 7 carried centromeric fragments from Chr6 (bottom right).
Deletion of CaRAD51 does not abolish interhomolog recombination, reduces SSA frequency, and induces ectopic translocation, chromosome loss and chromosome truncation In S. cerevisiae, SSA is Rad51-independent whereas crossovers are dependent on Rad51. We found that depletion of Rad51 caused a statistically significant drop in the rates of URA3 pop-outs. This is in striking contrast to a reported increase of SSA between direct repeats in rad51 null mutants of haploid S. cerevisiae (McDonald and Rothstein 1994;Bai and Symington 1996;Ivanov et al. 1996;Jablonovich et al. 1999;Sugawara et al. 2000;Pannunzio et al. 2010;Halas et al. 2016), in rad51 loss-of-function mutants in mammalian cells (Stark et al. 2004), and for Rad51 inhibition of Rad52-mediated annealing of complementary ssDNA in vitro (Wu et al. 2008). The decreased frequency of SSA in the absence of Rad51 suggests that although most URA3 pop-outs in wild type C. albicans were due to SSA, a few likely resulted from intra-chromatid crossover. Differences in SSA requirements between both yeasts might arise from the assay conditions (spontaneous in C. albicans vs. mostly DSBinduced in S. cerevisiae), the ploidy of the strains analyzed, or a differential regulation of SSA/intra-chromatid crossovers that could have evolved in response to the higher number of repeats in C. albicans or other species-specific traits.
A second interesting finding was that depletion of Rad51 did not abolish interhomolog recombination, which could still be caused by Rad51-independent BIR (VanHulle et al. 2007) or, more likely, by true crossover/BIR catalyzed by the Rad51-paralog DLH1 (Figure 7). DLH1 is the ortholog of the meiotic recombinase DMC1 that mediates strand invasion in S. cerevisiae meiotic cells (Bishop et al. 1992;Diener and Fink 1996). It is clear, however, that an important fraction of the events requiring strand invasion (inter-homolog, inter-sister chromatid, or intra-chromatid crossovers) initiated in the absence of Rad51 are defective and channeled toward ectopic translocation, chromosome loss and chromosome truncation (Table 1, Figure 7). Importantly, whereas ectopic translocations were completely absent in the presence of Rad51, they occurred at rates of 1.5 · 10 27 events/cell generation in rad51 strains and likely involved endogenous homologous sequences of another chromosome, but not hisG repeats (Figure 6), further supporting the idea that the latter are not hotspots for translocation.
We have previously shown that centric fragments of truncated chromosomes observed in rad52 strains of C. albicans are maintained when sealed by de novo telomere addition using junction sequences Figure 7 Model for genetic events in rad51 mutants. Left. Spontaneous (likely DLH1-mediated) recombination (crossover/BIR) between Chr6 homologs in G1 followed by the occurrence of a DSB. Each line accounts for ssDNA. Right. A DSB within the cassette region (fragile site) is followed by resection. Telomere addition at the resected end results in SNC formation (see Discussion). In rad51 strains ectopic translocation can occur if a complementary ssDNA tailed duplex of another broken and resected chromosome (red lines) is detected using the annealing activity of Rad52. If resection continues and trespasses a threshold (which we have traced to the neighborhood of SNP123) the Chr6 fragment cannot be maintained in the absence of Rad52 (and perhaps Rad51) resulting in chromosome loss (loss of Chr6B) or cell dead (loss of Chr6A). common to both chromosome and telomere (Andaluz et al. 2011). Therefore, it is likely that some resectioning of DSBs occurs before bases complementary to telomere repeats are exposed. Similarly, resectioning of DSBs can also expose ssDNA tracts complementary to sequences present on a different chromosome (Figure 7). In this scenario the absence of Rad51 would increase the substrate pool for translocation mediated by RAD52/RAD9-dependent SSA, consistent with the absence of translocations in rad52 mutants. We conclude that Rad51 is a strong suppressor of spontaneous ectopic translocation in C. albicans. This is consistent with findings in rad51 haploid S. cerevisiae where spontaneous Ty1-mediated GCR rates were increased sevenfold (Chan and Kolodner 2011) and with the spontaneous ectopic translocation frequency between 300 bp of identical sequence in diploids (Pannunzio et al. 2008;Manthey and Bailis 2010). In the latter case, the rate of translocation for Scrad51 strains was 1.1 · 10 27 /cell generation, which is similar to what we found for Carad51 in our study.
On the generation of SNCs in the absence of CaRad52 Our assay did not select for rad52 5FOA R derivatives carrying a centromeric SNC that retains a functional URA3. Given the size of Chr6 (1032 kb) and the distance between rad52::hisG-URA3-hisG and the left telomere (% 95 kb), the maximum expected size for a SNC is 940 kb, which is consistent with the observation that SNCs were always smaller than 945 kb (the size of Chr7). However, there are no constraints for the minimal size of a Chr6 SNC other than retention of cen6, which is only 53 kb away from the right telomere. In fact, a 95 kb centromeric fragment was conserved following truncation of Chr6 in vivo (Baum et al. 2006). The large size of SNCs (usually . 813 kb and never smaller than 666 kb), could stem from the inability of rad52 cells to maintain SNCs smaller than 500 -600 kb (high chromosome loss frequency). In agreement with this, it was shown previously that the only Chr6derived SNC identified among spontaneous histidine auxotrophs generated by rad52 cells was % 600 kb (Andaluz et al. 2011), which matches the size of the smallest SNCs detected in this study (% 630 kb). One explanation for the abundance of large SNCs is the presence of a fragile site between snp122 and the left telomere whose tendency to break increases during DNA replication in the absence of Rad52 and, to a lesser extent, Rad51. Importantly, in rad52 or rad51 haploid S. cerevisiae, defective fork restart at damaged (methylated) sites results in chromosome breakage and cell death (González-Prieto et al. 2013), and a similar defect could generate SNCs in diploid C. albicans rad52 and rad51 derivatives. An attractive possibility is that the URA-Blaster acts as a fragile site due to unusual DNA or chromatin structure that converts it into a locus difficult to replicate and thereby mimicking DNA damaged sites. To account for the actual size of SNCs, the initial break at the URA-Blaster would require a minimum of 100 -120 kb resection which, according to data from S. cerevisiae, could be too long for wild type (Zhu et al. 2008) but not for rad52 strains (Sugawara and Haber 1992). We do not rule out that the C. albicans major repeat sequence also may act as a recombination hotspot (Lephart and Magee 2006;Chibana and Magee 2009;Marton et al. 2019) and become a breakage site in the absence of homologous recombination proteins. However, if that were the case, resulting SNCs would be too small to be maintained in rad52 strains.
In this study we show that although recombination pathways are basically conserved, C. albicans exhibits specific requirements for mitotic recombination that affect expansion and contractions of repeated sequences, including Rad52-independent SSA (for repeat lengths as short as 1.2 kb) and Rad51-independent interhomolog recombination. This opens up the exciting possibility that, in addition to affecting genome structure, specific features of the recombination machinery may have evolved to facilitate variation. Ongoing and future research will be studying the impact of specific stresses on repeat stability and identifying recombination requirements for repeats of reduced length.

ACKNOWLEDGMENTS
We are indebted to Andrés Aguilera for allowing us to do experiments in his lab and to Joachim Morschhäuser for providing the SAT1 cassette. A.F. is funded by NIH grant R15AI090633.