Resection and repair of a Cas9 double-strand break at CTG trinucleotide repeats induces local and extensive chromosomal rearrangements

Microsatellites are short tandem repeats, ubiquitous in all eukaryotes and represent ∼2% of the human genome. Among them, trinucleotide repeats are responsible for more than two dozen neurological and developmental disorders. Targeting microsatellites with dedicated DNA endonucleases could become a viable option for patients affected with dramatic neurodegenerative disorders. Here, we used the Streptococcus pyogenes Cas9 to induce a double-strand break within the expanded CTG repeat involved in myotonic dystrophy type 1, integrated in a yeast chromosome. Repair of this double-strand break generated unexpected large chromosomal rearrangements around the repeat tract. These rearrangements depended on RAD52, DNL4 and SAE2, and both non-homologous end-joining and single-strand annealing pathways were involved. Resection and repair of the double-strand break (DSB) were totally abolished in a rad50Δ strain, whereas they were impaired in a sae2Δ mutant, only on the DSB end containing most of the repeat tract. This proved that Sae2 plays significant different roles in resecting a DSB end containing a repeated and structured sequence as compared to a non-repeated DSB end. In addition, we also discovered that gene conversion was less efficient when the DSB could be repaired using a homologous template, suggesting that the trinucleotide repeat may interfer with gene conversion too. Altogether, these data show that SpCas9 is probably not a good choice when inducing a double-strand break at or near a microsatellite, especially in mammalian genomes that contain many more dispersed repeated elements than the yeast genome.


Introduction
Microsatellites are short tandem repeats ubiquitously found in all eukaryotic genomes sequenced so far (Richard et al., 2008). Altogether, they cover ~2% of the human genome, a figure similar to the whole protein-coding sequence (International Human Genome Sequencing Consortium, 2004). Naturally prone to frequent repeat length polymorphism, some microsatellites are also prone to large expansions that lead to human neurological or developmental disorders, such as trinucleotide repeats involved in Huntington disease, myotonic dystrophy type 1 (Steinert disease), fragile X syndrome or Friedreich ataxia (Orr and Zoghbi, 2007). These expansion-prone microsatellites share the common property to form secondary DNA structures in vitro (Gacy et al., 1995) and genetic evidences suggest that similar structures may also form in vivo, transiently stalling replication fork progression (Anand et al., 2012;Nguyen et al.;Pelletier et al., 2003;Samadashwily et al., 1997;Viterbo et al., 2016). Among those, CCG/CGG trinucleotide repeats are fragile sites in human cells, forming frequent double-strand breaks when the replication machinery is slowed down or impaired (Sutherland et al., 1998). Similarly, CAG/CTG and CCG/CGG microsatellites are also fragile sites in Saccharomyces cerevisiae cells (Balakumaran et al., 2000;Freudenreich et al., 1998). Therefore, microsatellite abundance and the natural fragility of some of them make these repeated sequences perfect targets to generate chromosomal rearrangements potentially leading to cancer.
Double-strand break (DSB) repair mechanisms have been studied for decades in model organisms as well as in human cells and led to the identification of the main genes involved in this process (reviewed in Krogh and Symington, 2004). Many of these advances were made possible by the use of highly specific DNA endonucleases, such as the meganucleases I-Sce I or HO (Fairhead and Dujon, 1993;Haber, 1995;Plessis et al., 1992). Other frequently used methods involved ionizing radiations making genome-wide DSBs (Nelms et al., 1998).
However, the fate of a single double-strand break within a repeated and structured DNA sequence has never been addressed, until recently. In a former work, we used a TALE Nuclease (TALEN) to induce a unique DSB into a long CTG trinucleotide repeat integrated into a S. cerevisiae chromosome. We showed that 100% of yeast cells in which the TALEN was expressed exhibited a large contraction of the repeat tract, going from an initial length of ~80 CTG triplets to less than 35. POL32, DNL4 and RAD51 were shown to play no detectable role in repairing this DSB. On the contrary, RAD50, RAD52 and SAE2 were required for proper repair of the DSB, and a functional Sae2 protein was found to be essential for efficient DSB resection, suggesting that repeat contraction occurred by a single-strand annealing (SSA) process, involving preliminary resection of the break, followed by annealing of the two DSB ends carrying the repeat tract Richard et al., 2014).
In the present work, we used the Streptococcus pyogenes Cas9 endonuclease (SpCas9) to induce a DSB within the same long CTG trinucleotide repeat integrated in the yeast genome.
The break was made at the 3' end of the repeat tract ( Figure 1A), using a guide RNA that targets the repeat tract. Frequent rearrangements were found in surviving cells, with local deletions as well as more extensive ones involving recombination between retrotransposon LTRs. Survival and repair depended on RAD50, RAD52, SAE2 and DNL4 and double-strand break resection was abolished in rad50Δ and sae2Δ mutants. A more specific version of the nuclease, Enhanced SpCas9, generated the same rerrangements. In addition, we also discovered that gene conversion was less efficient when SpCas9 was used to induce a DSB within a CTG repeat tract that could be repaired with a homologous template, suggesting that the trinucleotide repeat may interfer with gene conversion too.

A Cas9-induced double-strand break within CTG repeats induces cell death and chromosomal rearrangements around the repeat tract
In previous works, we showed that a TALEN targeted at the CTG trinucleotide repeat from the human DMPK gene 3' UTR integrated in a yeast chromosome, was extremely efficient at contracting the repeat tract below the pathological length Richard et al., 2014). In order to determine whether the CRISPR-Cas9 system could be used in the same manner, a plasmid-borne Streptococcus pyogenes Cas9 nuclease (SpCas9) was expressed in Saccharomyces cerevisiae from a GAL1-inducible promoter (DiCarlo et al., 2013). The same plasmid also carried a CTG guide RNA (hereafter named gRNA#1) under the control of the constitutive SNR52 promoter. The PAM used in this experiment was the TGG sequence located right at the border between the CTG tract and non-repeated DNA ( Figure 1A). As controls, we used the same SpCas9-containing plasmid without the gRNA or a frameshift mutant of the SpCas9 gene resulting in a premature stop codon (SpCas9ΔNdeI) and the gRNA#1. The same genetic assay as previously was used Richard et al., 2014). It is based on a modified suppressor tRNA gene (SUP4) in which a CTG trinucleotide repeat was integrated. The length of the CTG repeat at the start of the experiment was determined to be approximately 80 triplets. Four hours after transition from glucose to galactose medium, two faint bands were visible on a Southern blot, corresponding to the 5' and 3' ends of the SpCas9 DSB. No signal was detected in both control strains ( Figure 1B). The DSB was quantified to be present in ca. 10% of the cells at any given time point and remained the same for the duration of the time course (4 hours). No evidence for repeat tract contraction was visible by Southern blot. Survival to the SpCas9 break was low (17.9%±4%), as calculated from CFU on galactose plates over CFU on glucose plates ( Figure   2, see Materials & Methods). Surviving colonies were picked, total genomic DNA was extracted and the SUP4::CTG-locus was analyzed by Southern blot. Patterns observed were remarkably different among clones, most of them showing bands of aberrant molecular weight, either much larger or much shorter than the repeat tract. In some clones, a total absence of signal suggested that the probe target was deleted and in other cases weakness of the signal was compatible with partial deletion ( Figure 3A). To understand these abnormal patterns, the genome of nine surviving clones were totally sequenced by paired-end Illumina.
As a control, one clone in which SpCas9 had not been induced was also sequenced. In all nine cases, a deletion around the repeat tract was found, extending from a few nucleotides to several kilobases ( Figure 3B). Some of the rearrangements involved flanking Ty1 retrotransposon LTRs, and in one case (clone #2), a complex event between a distant LTR (δ16) and the δ20 LTR close to the repeat tract was detected. Following this discovery, total genomic DNA was extracted from more surviving colonies, analyzed by Southern blot and rearrangement junctions were amplified by PCR and Sanger sequencing. Two different sets of primers were used to amplify the junction, su47/su48 allowing to amplify local rearrangements around the repeat tract, whereas su23/su42 were used to amplify larger rearrangements between Ty1 LTRs ( Figure 4, Table 2). Surviving colonies were classified into 12 different types, according to the SUP4::CTG locus after SpCas9 induction: type I corresponded to a colony in which the repeat tract was unchanged, types II-V corresponded to local rearrangements around the repeat tract and types VI-XII corresponded to more extensive rearrangements, the last category involving a complex event between δ16 and δ20.

Chromosomal rearrangements are under the control of RAD52, DNL4 and SAE2
We next decided to investigate the role of several genes known to be involved in DSB repair on chromosomal rearrangements generated by the SpCas9 nuclease. In a rad52Δ strain, in which all homologous recombination was abolished, survival significantly decreased below wild type (7.6%±0.7%, Figure 2). Molecular analysis of the survivors by Southern blot showed that rearrangements seemed to be less extensive than in the wild-type strain, fewer lanes showing a partial or total absence of signal ( Figure 3C). Sequencing of the junctions confirmed that rearrangements between Ty LTRs were lost (Type VI events), except for two cases in which the deletion occurred through annealing of eight or nine nucleotides and was therefore RAD52 independent ( Figure 4 and Supplemental Figure 1, clones #C4 and #C7).
This result shows that about 50% of colonies growing on galactose plates survived to the DSB by RAD52-dependent homologous recombination between two LTR elements flanking the trinucleotide repeat tract.
The possible role of non-homologous end-joining (NHEJ) in the observed rearrangements was also addressed by deleting the gene encoding yeast Ligase IV (DNL4). In the dnl4Δ strain, the level of detected DSBs was significantly lower than in wild type (Figures 1B and 1C), suggesting that DSB ends may be partially protected by the presence of Dnl4p, and were processed more rapidly when Ligase IV was absent. Survival was slightly decreased, but not significantly different from wild type (10.5%±6.3%, Figure 2). Molecular analysis of the survivors showed that local rearrangements were totally lost, whereas extensive rearrangements involving Ty LTR represented 84% of all events ( Figure 4 and Supplemental Figure 1). Hence, we concluded that all local rearrangements were NHEJ dependent.
In a recent work, we showed that SAE2 was essential to repair a DSB induced by a TALEN within a long trinucleotide repeat. In its absence, unrepaired breaks accumulated and DSB resection was lost on the trinucleotide repeat-containing end . We therefore tested the effect of a sae2Δ mutation on a SpCas9 DSB in the same experimental system. Southern blot analysis of repair intermediates showed that DSB ends accumulated twice as much in the sae2Δ mutant as compared to wild type ( Figures 1B and 1C). In addition, a smear was detected below the 5' DSB end ( Figure 1B, orange bracket), hallmark of a resection defect (Chen et al., 2013). Survival was similar to wild type (21.5%±2.9%, Figure   2). Southern blot analysis of surviving colonies displayed very little size changes as compared to uninduced controls ( Figure 3D). However, sequencing showed that the most frequent event was an insertion (or sometimes a small deletion) of one to eight nucleotides between the PAM and the repeat tract (Type III events, Figure 4). These local insertions represented 78% of all survivors, whereas only one Ty LTR recombination (Type VI) was detected (Supplemental Figure 1). This showed us that in the absence of SAE2, long range rearrangements were lost, probably due to the inability to resect the DSB into single-stranded DNA prone for homologous recombination.
The double mutant sae2Δ dnl4Δ was also built and showed an additive effect on survival, with a 30-fold reduction in CFU on galactose plates (0.6%±0.9%, Figure 2). This proved that in the absence of one of the two genes repair could occur by the other pathway, but absence of both genes was almost lethal to yeast cells receiving a SpCas9 DSB. Southern analysis showed that DSB levels were similar to sae2Δ levels (ca. 24% after 8 hrs versus 28% for sae2Δ), showing that SAE2 was epistatic to DNL4 for this phenotype. The smear corresponding to resection defects was also visible ( Figure 1B, orange bracket). Interestingly, 21% of survivors exhibited zero to two triplets lost, which could be attributed to natural microsatellite instability. These were classified as Type I events and were specific of the sae2Δ dnl4Δ double mutant (Figure 4 and Supplemental Figure 1). It is possible that given the low survival rate, cells in which SpCas9 and/or the gRNA was mutated were positively selected during the time course in liquid culture and were therefore subsequently recovered on galactose plates. It is however surprising that such events were not recovered in rad50Δ cells.
Remarkably, to the exception of the Type I events hereabove mentioned, all but one event corresponded to extensive rearrangements around the repeat tract, similarly to the single dnl4Δ mutant. This shows that although effects of both mutations were additive on cell survival, DNL4 was epistatic to SAE2, when chromosomal rearrangements induced by SpCas9 were considered.
Finally, in a rad50Δ strain, the DSB accumulated over the duration of the time course at levels similar to sae2Δ mutants ( Figures 1B and 1C). No smear was detected in this strain background, suggesting that the sae2Δ resection defect was specific of this gene and did not involve the integrity of the MRX-Sae2 complex. However, no survivor could be found on galactose plates, showing that the DSB was lethal in this mutant background ( Figure 2).
In conclusion, when a SpCas9 DSB was induced into a long CTG trinucleotide repeat, cell survival was low and depended on RAD50, RAD52, SAE2 and DNL4. Two classes of repair events were found: local rearrangements under the control of DNL4 and therefore the NHEJ pathway, and extensive rearrangements under the control of SAE2 and RAD52. In addition, the deletion of RAD50 almost completely recapitulated the sae2Δ dnl4Δ double mutation, except that the smear was not visible in rad50Δ and no surviving cell could be found, suggesting that in the absence of this gene DSB repair cannot occur at all on CTG trinucleotide repeats, following a SpCas9-induced DSB.

Enhanced SpCas9 generates the same chromosomal rearrangements as SpCas9
Over the last three years, several mutants of the widely used SpCas9 have been engineered or selected by genetic screens. SpCas9-HF1 and eSpCas9 were built to exhibit less off-target DSBs (Kleinstiver et al., 2016;Slaymaker et al., 2016), HypaCas9 was made to be even more accurate (Chen et al., 2017), Sniper-Cas9 also showed reduced off-target effects (Lee et al., 2018), while evoCas9 was selected in yeast for improved specificity (Casini et al., 2018). We decided to explore the possibility that chromosomal rearrangements observed in our experimental system were partly due to the fact that SpCas9 exhibited a high off-target activity on long CTG trinucleotide repeat tract, perhaps by generating more than one DSB within the repeat tract, or within the surrounding loci. In order to test this hypothesis, Enhanced SpCas9 (eSpCas9) was expressed in yeast, along with the same guide RNA as previously (gRNA #1, Figure 1A). Survival was slightly higher than with SpCas9 (26.3%±3.0%), but not significantly different (t test p-value= 0.06). DSB end accumulation was lower than SpCas9 ( Figure 1B, 1C). Molecular analysis of surviving yeast cells did not show any statistical difference between types of rearrangements observed with eSpCas9 as compared to SpCas9 (Chi2 p-value= 0.14) ( Figure 5 and Supplemental Figure 1). A second guide RNA (gRNA#2) was designed, so that the DSB would be made two nucleotides closer to the repeat tract end ( Figure 1A). Interestingly, the number of rearrangements involving a LTR was lower than with gRNA#1 (19% with gRNA#2 vs 92% with gRNA#1) but the proportion of very large deletions (Type XI) significantly increased from 4% to 31% (Chi2 p-value= 1.6 10 -3 ). We concluded that moving the DSB cut site two nucleotides toward nonrepeated DNA increased the outcome of very large deletions. Altogether, these results show that using a more specific version of SpCas9 did not decrease chromosomal rearrangements, but actually increased extensive deletions, suggesting that the effects seen were probably not due to extra off-target DSBs within the CTG repeat tract or the surrounding loci.

Gene conversion efficacy is decreased when a Cas9 DSB is made within a long CTG trinucleotide repeat
Gene conversion is a very efficient DSB-repair mechanism in S. cerevisiae. We previously showed that a single DSB induced by the I-Sce I meganuclease in a yeast chromosome was efficiently repaired using a CTG repeat-containing homologous template as a donor (Richard et al., 1999a(Richard et al., , 2000. In order to determine whether a Cas9-induced DSB within a CTG repeat was properly repaired by the recombination machinery, we reused a similar experimental system in which two copies of the SUP4 allele were present on yeast chromosome X, one containing a (CTG) 60 repeat tract and the other copy containing an I-Sce I recognition site ( Figure 6A). In this ectopic gene conversion assay, 80.2%±2.3% of yeast cells survived to an I-Sce I DSB and 100% of survivors were repaired by gene conversion using the ectopic SUP4::(CTG)n copy as a donor . When SpCas9 was induced in the same yeast strain along with gRNA#1, only 32.6%±3.8% of CFU formed on galactose plates ( Figure 2). Molecular analysis of surviving cells showed that 89% (34 out of 38) repaired by ectopic gene conversion, as expected, and now contain two I-Sce I recognition sites, one in each SUP4 copy ( Figure 6B, GC events). However, one expansion event was also detected, as well as one local rearrangement (Type IV) and two rearrangements involving a deletion and a DNA insertion (Type V). Intriguingly, the DNA insertion is a 211 bp piece of DNA from the YAK1 gene, located 158 kilobases upstream the ARG2 locus, on chromosome X left arm. This gene contains a long and imperfect CAG/CTG repeat within its reading frame, like many yeast genes (Field and Wills, 1998;Malpertuy et al., 2003;Richard and Dujon, 1996;Richard et al., 1999b). An unusual non-homologous recombination event occurred between the YAK1 CAG/CTG repeat and the ARG2 repeat, leading to a chimeric insertion ( Figure 6C). This may be the result of an off-target DSB generated by SpCas9 within the YAK1 repeat, or be due to an abnormal recombination event between the two CTG repeats following SpCas9 induction. Using the CRISPOR in silico tools, three off-target sites were found for SpCas9 gRNA#1 if no mismatch were allowed and 80 sites when up to three mismatches were permitted (Haeussler et al., 2016). The second best off-target score was YAK1. We concluded that gene conversion efficacy was reduced when the DSB was made by SpCas9 within a CTG repeat, as compared to an I-Sce I DSB made in a non-repeated sequence, partly due to occasional off-target DSBs in other CTG repeats of the yeast genome.

Resection of a Cas9-induced double-strand break
Quantitative PCR experiments were performed in order to determine the resection level in strains in which Cas9 was induced. The nuclease generates a DSB in the very last CTG triplets of the repeat tract ( Figure 1A). Therefore, the 5' end of the break contains most of the 80 triplets whereas the 3' end contains only two triplets. This allows to compare resection of a repeated and structured DNA end versus non-repeated DNA, concomitantly and in the same experimental setting. We took advantage of the convenient position of four EcoRV restriction sites, two on each side of the DSB, at different distances from the break ( Figure 7A). Primers were designed in such a way that EcoRV digested DNA could not be PCR amplified.
However, if DNA resection reached an EcoRV site, the resulting single-stranded DNA became resistant to digestion and therefore susceptible to amplification. In wild-type cells after eight hours, resection of the Cas9 DSB was always 100% at all EcoRV sites, except at the 3' distal site in which it was a little lower, around 70% ( Figure 7A). In dnl4Δ cells, resection was not statistically different from wild type. In the rad50Δ mutant, DSB resection was totally abolished on the 5' end of the break that contains most of the repeat tract and severely impaired on the other end, showing that the MRX-Sae2 complex was essential in this process, on both DSB ends. Interestingly, the sae2Δ mutant exhibited a resection defect on the 5' end of the break but not on the other side. This was also true for the double mutant sae2Δ dnl4Δ. All these data prove that: i) Ligase IV plays no role in DSB resection; ii) Sae2 is essential to resect a long CTG trinucleotide repeat but is dispensable to resect a non-repeated DSB end.

Genome-wide mutation spectrum in cells expressing SpCas9
When carefully looking at deletion borders in haploid strains in which SpCas9 was induced, they were found to be more extensive on the 5' side of the break than on the 3' side ( Figure 4). This suggests that larger 3' deletions encompassing the essential gene CDC8 or its promoter may not have been recovered because they would be lethal. This could also account for the lethality observed with SpCas9 in haploids. In order to check this hypothesis, we analyzed diploids containing SUP4::(CTG)n repeat tracts on both homologues, in which SpCas9 was induced. In these cells, both chromosomes could be cut by the nuclease. We quantified by qPCR CDC8 copy number in six independent diploid survivors. In all cases, it was reduced by half as compared to a control qPCR on another chromosome (Supplemental Figure 2A). This showed that in these cells, only one of the two CDC8 alleles was present, suggesting that the other was often deleted during DSB repair. In order to check if the whole chromosome could have been lost, we also amplified a region near the JEM1 gene, on the other chromosomal arm, near the ARG2 gene. Surviving clones showed a significantly higher signal, compatible with the presence of two chromosomes (Mann-Whitney-Wilcoxon rank test, p-value= 10 -3 ). Therefore, it was concluded that the mortality observed in haploid cells expressing SpCas9 was at least partly due to the frequent deletion of the essential CDC8 gene, but did not induce significant chromosome loss.
In order to detect possible off-target mutations, independent haploid and diploid colonies in which SpCas9 had been induced were deep-sequenced, using Illumina paired-end technology.
In diploid cells, two nucleotide substitutions were detected out of five independent clones when SpCas9 was repressed (Supplemental Figure 2B). When the nuclease was induced, six mutations were detected out of 18 sequenced clones, a similar proportion. One 36-bp deletion was found in the FLO11 minisatellite and two deletions of one repeat unit were found in AT dinucleotide repeats, but no mutation was found in any other CAG/CTG repeat tract.
Altogether, we concluded that SpCas9 expression in diploid cells did not significantly increase genome-wide mutation frequency. The genome of 10 independent haploid cells in which SpCas9 was induced was also completely sequenced. Eight mutations were detected among six of these survivors, all of them being nucleotide substitutions in non-repeated DNA (Supplemental Figure 2B). This is statistically not different from what was observed in diploids (Fisher exact test p-value= 0.003). We concluded that besides chrosomosomal rearrangements around the SUP4 locus observed in these haploids, SpCas9 did not induce other mutations in yeast cells in which it was expressed.

SpCas9-induced DSB repair within CTG repeats generates chromosomal rearrangements
In previous works, in which we induced a DSB within a CTG repeat using a dedicated TALEN, 100% of surviving yeast colonies repaired the break by contracting the repeat tract (Richard et al., 2014). These contractions occurred by an iterative single-strand annealing mechanism, that depended on RAD52, RAD50 and SAE2, but was independent of LIG4, POL32 and RAD51 . It was therefore striking and completely unexpected that a DSB made by the SpCas9 nuclease (or by its more specific mutant version, eSpCas9) at exactly the same location within the very same CTG trinucleotide repeat induced frequent chromosomal rearrangements around the repeat tract and almost no repeat contraction. With the TALEN, repeat contraction was shown to be an iterative phenomenon, involving several rounds of cutting and contraction until the repeat tract was too short for the two TALEN arms to dimerize and induce a DSB Richard et al., 2014).
A similar outcome was expected with SpCas9, iterative rounds of cutting and contraction could occur until the remaining CTG repeat tract would be too short for the gRNA to bind and induce a DSB. However, this was not observed here, surprisingly proving that a SpCas9induced DSB was differently repaired from a TALEN-induced DSB targeting the same exact repeated sequence.
Spontaneous homologous recombination events between delta elements surrounding SUP4 were already described by the past (Rothstein et al., 1987). However, in the present case, Cas9-induced deletions also involved microhomology sequences or no homology at all, in addition to delta LTR elements, suggesting that the initiating damage was different in both experimental systems (replication-induced single-strand nicks vs. nuclease-induced doublestrand breaks, for example). Our data are more reminiscent of a previous work in which spontaneous deletions around the URA2 gene were classified in seven different classes, six of them harboring microhomologies at their junctions and one showing no obvious homology (Welcker et al., 2000). In a recent work, using a GFP reporter system in human cells, it was shown that SpCas9 induced contractions as well as expansions of long CTG trinucleotide repeats, whereas the nickase mutant SpCas9-D10A only induced contractions. However, it was not possible to determine whether local rearrangements could be present in some of the surviving cells (Cinesi et al., 2016). In a recent work looking at the effect of a Cas9-induced DSB at the LYS2 locus in S. cerevisiae, the authors found frequent POL4-dependent small insertions (1-3 bp) in 42-68% of the survivors (depending on the PAM used) and local deletions (1-17 bp) in the remaining cases. In the present experiments, local rearrangements account for 19.6% with 20% of those being small insertions (Figure 4). The remaining events (80.4%) corresponded to extensive rearrangements involving retrotransposon LTRs.
However, given that there is no transposon or transposon remnant in the close proximity of the LYS2 locus, the authors could not retrieve LTR rearrangements (Lemos et al., 2018). This strongly suggests that rearrangements observed heavily depend on the surrounding chromosomal location where the DSB is made.
When an I-Sce I DSB was induced within a SUP4 allele, the break could be repaired by gene conversion with a CTG repeat-containing homologous donor at the ARG2 locus. All yeast cells repaired by gene conversion with the donor, generating repeat contractions and expansions in the process . Here, the exact reverse reaction was induced, the break was made within CTG repeats and repaired with a non-repeated sequence. DSB repair was much less efficient, since only 32.6% of the cells survived ( Figure 2) and less specific since 10% of the repair events were unfaithful recombination ( Figure 6B). This shows that when a Cas9 DSB was made into a CTG repeat, gene conversion was impaired, either by the repeat tract or by the Cas9 protein, or by both.
Ligase IV and Sae2 are respectively driving local and extensive chromosomal rearrangements Yeast Ligase IV is encoded by the DNL4 gene and is the enzyme used to ligate DSB ends during non homologous end-joining (Wilson et al., 1997). It was previously shown that RAD50 and SAE2 were essential to resect and process a TALEN-induced DSB but a DNL4 deletion had no effect on break processing, cell survival or repair efficacy . On the contrary, repair of a Cas9, an HO or an I-Sce I DSB at the MAT locus, in the absence of any homologous donor casette, was shown to be dependent on the product of the DNL4 gene (Frank-Vaillant and Marcand, 2001;Lemos et al., 2018). SpCas9 DSB repair has also been studied in human cells in the presence of a drug (NU7441), acting as a chemical inhibitor of non-homologous end-joining. In these conditions, the frequency of single-base insertions and small deletions decreased whereas larger deletions increased, suggesting that these repair events occurred by an alternative end-joining mechanism (alt-EJ/MMEJ) involving microhomologies flanking the DSB van Overbeek et al., 2016). Here, we showed that when DNL4 was inactivated, local deletions were totally lost.
However, survival was not significantly decreased because yeast cells could repair the DSB using LTR recombination, generating extensive deletions around the repeat tract (Figure 4).
Supporting this model, the absence of any resection defect in the dnl4Δ mutant proved that in the absence of end-joining, resection may take place very efficiently to repair the DSB by homologous recombination, using flanking homologies.
SAE2 is associated to the MRE11-RAD50-XRS2 complex, whose roles are multiple during DSB repair (Haber, 1998) and it was proposed to encode an endonuclease activity essential to process DNA hairpins (Lengsfeld et al., 2007), as well as to resect I-Sce I double-strand breaks (Mimitou and Symington, 2008). We previously showed that it was essential to resect a TALEN-induced DSB end containing a long CTG trinucleotide repeat, but less important to resect non-repeated DNA . In the present experiments, extensive rearrangements involving LTR elements were lost in a sae2Δ mutant, and 97% of yeast cells repaired the DSB by local rearrangements, most of them resulting in insertions or deletions between the PAM and the gRNA sequence (Figure 4), inactivating SpCas9 capacity to induce another DSB. Small insertions of a few nucleotides were also frequently detected following SpCas9 DSB induction at the VDJ locus in human B cells (So and Martin, 2019) or at the MAT locus in S. cerevisiae (Lemos et al., 2018). However, in our experiments, all nucleotides inserted were C, T or G, all three encoded by the gRNA. No insertion of an adenosine residue was found out of 28 insertions sequenced (Supplemental Figure 1). This intriguing observation suggests the possibility that the gRNA could be used as a template to repair the DSB, as it was demonstrated that a single-stranded RNA could be used to repair an HOinduced DSB into the LEU2 gene (Storici et al., 2007).
Although DNL4 and SAE2 trigger different types of chromosomal rearrangements and none of the single mutants significantly decreased survival, the dnl4Δ sae2Δ double mutant abolished repair, almost to the point of the rad50Δ mutant, since only 0.6% of the cells survived ( Figure   2), showing the synthetic effect of both mutations. However, repair events in the double mutant were similar to those observed in dnl4Δ, proving that DNL4 was epistatic on SAE2 ( Figure 4). All these results are compatible with a model in which a Cas9 DSB was tentatively repaired by NHEJ first ( Figure 7B). Then, if repair was unsuccessful or if DNL4 was inactivated, resection proceeded with MRX-Sae2 in charge of removing secondary structures present at DSB ends. When resection reached flanking LTRs, repair occured by RAD52mediated SSA. In the absence of this gene, the break was repaired by RAD52-independent local rearrangements. Finally, when SAE2 was inactivated, resection was impeded on the 5' DSB end containing most of the CTG repeat tract ( Figure 7A), and mutagenic NHEJ was favored, leading to local insertions and deletions. It is unknown whether Sae2 would play the same essential role on other secondary structure-forming trinucleotide repeats, like GAA or CGG triplets, or if its activity is specific of CTG triplets, hence of a structure rather than a repeat. It is also unclear whether other nucleases, like EXO1 or DNA2, would be important to perform long range resection on a long CTG repeat tract (Mimitou and Symington, 2008;Zhu et al., 2008), but the present experimental system allows us to address this question in a unique and unbiased way.

Yeast strains and plasmids
All mutant strains were built from strain GY6162-3D by classical gene replacement method (Orr-Weaver et al., 1981), using KANMX4 or HIS3 as marker (Supplemental Table 1).
VMY350 and VMY352 strains were respectively used to construct VMY650 and VMY352 by mating-type switching, as follows: the pJH132 vector (Holmes and Haber, 1999) carrying the HO endonuclease under the control of an inducible GAL1-10 promoter was transformed in the haploid MATα strains. After 5h of growth in lactate medium, HO expression was induced by addition of 2% galactose (final concentration) and grown for 1.5 hour. Cells were then plated on YPD and mating type was checked three days later by crosses with both MATa and MATα tester strains.
For SpCas9 inductions, addgene plasmid #43804 containing the nuclease under the control of the GalL promoter and the LEU2 selection marker was digested with HpaI and cloned into yeast by homology-driven recombination (Muller et al., 2012) with a single PCR amplified fragment containing the SNR52 promoter, the gRNA#1 and the SUP4 terminator, using primers SNR52Left and SNR52Right (Supplemental Table 2) to give plasmid pTRi203. A frameshift was then introduced in this plasmid by NdeI digestion followed by T4 DNA polymerase treatment and religation of the plasmid on itself, to give plasmid pTRi206. In this plasmid, the SpCas9 gene is interrupted by a stop codon after amino acid Ile 161 . The haploid GFY6162-3D strain (or its mutant derivatives), was subsequently transformed with pTRi203 or pTRi206 and transformants were selected on SC-Leu. The plasmid containing Enhanced  Table 2) and cloned into yeast cells in the Addgene#43804 plasmid digested with BamHI, by homology-driven recombination, with 34-bp homology on one side and 40-bp homology on the other side (Muller et al., 2012), to give plasmid pLPX11. For the gRNA#1, plasmid pLPX11 was digested with HpaI and cloned into yeast by homology-driven recombination (Muller et al., 2012) with a single PCR amplified fragment containing the SNR52 promoter, the gRNA#1 and the SUP4 terminator, as above to give plasmid pTRi207. For the gRNA#2, a guide RNA cassette was ordered from ThermoFisher (GeneArt), flanked by EcoRI sites and was cloned in pRS416 (Sikorski and Hieter, 1989) using standard procedures to give plasmid pLPX210.

In silico simulations of off-target sites
To assess the number of off-target sites for SpCas9 in Saccharomyces cerevisiae, online tools were used. CRISPOR is a software that evaluates the specificity of a guide RNA through an alignment algorithm that maps sequences to a reference genome to identify putative on-and off-target sites ). To predict off-target sites, the online tool sequentially introduces changes in the sequence of the gRNA and checks for homologies in the specified genome (Haeussler et al., 2016).

Cas9 inductions
Before nuclease induction, Southern blot analyses were conducted on several independent subclones to select one containing ca. 80 CTG triplets. For Cas9 inductions, yeast cells were grown overnight at 30°C in liquid SC-Leu medium, then washed with sterile water to remove any trace of glucose. Cells were split in two cultures, half of the cells were grown in synthetic -Leu medium supplemented with 2% galactose (final concentration) and the other half were grown in synthetic -Leu medium supplemented with 2% glucose (final concentration).
Around 4x10 8 cells were collected at different time points (T=0, 4, 5, 6, 7 and 8 hours) and killed by addition of sodium azide (0.01% final). Cells were washed with water, and frozen in dry ice before DNA extraction. To determine survival to Cas9 induction, 24 hours after the T0 time point, cells were diluted to an appropriate concentration, then plated on SC-Leu plates containing either 20 g/l glucose or galactose. After 3-5 days of growth at 30°C, ratio of CFU on galactose plates over CFU on glucose plates was considered to be the survival rate.

Double-strand break analysis and quantification
Total genomic DNA (4 μ g) of cells collected at each time point was digested for 6h with EcoRV (40 U) (NEB) loaded on a 1% agarose gel (15x20 cm) and run overnight at 1 V/cm.
The gel was vaccum transfered in alkaline conditions to a Hybond-XL nylon membrane (GE Healthcare) and hybridized with two randomly-labeled probes specific of each side of the repeat tract, upstream and downstream the SUP4 gene . After washing, the membrane was overnight exposed to a phosphor screen and signals were read and quantified on a FujiFilm FLA-9000.

SUP4 locus analysis after Cas9 induction
Several colonies from each induced or repressed plates were picked, total genomic DNA (4 μ g) was extracted with Zymolyase, digested for 6h by SspI (20 U) (NEB), loaded on a 1% agarose gel (15x20 cm) and run overnight at 1V/cm. The gel was vaccum transfered in alkaline conditions to a Hybond-XL nylon membrane (GE Healthcare) and hybridized with a randomly-labeled PCR fragment specific of a region downstream the SUP4 gene, amplified from the su8-su9 primer couple (Supplemental Table 2). After washing, the membrane was overnight exposed on a phosphor screen and signals were revealed on a FujiFilm FLA-9000.
Genomic DNA of each clone for which a signal was detected by Southern blot was subsequently amplified with su47-su48 primers and sequenced using su47 (Supplemental Table 2). Genomic DNA of clones for which no signal was detected by Southern blot of no PCR product was obtained with su47-su48 were subsequently amplified with su23-su42 primers and sequenced using su42 (Supplemental Table 2). Sanger sequencing was performed by GATC biotech.

Analysis of Cas9-induced DSB end resection by qPCR
A real-time PCR assay, using primer pairs flanking EcoRV sites 0.81 kb and 2.94 kb away from the 3' end of the CTG repeat tract (VMS20/VMAS20 and VMS21/VMAS21 respectively) and 0.88 kb and 1.88 kb away from the 5'end of the CTG repeat tract (VMS22/VMAS22 and VMS23/VMAS23 respectively), was used to quantify end resection.
Another pair of primers was used to amplify a region of chromosome X near the ARG2 gene (Viterbo et al., 2016), to serve as an internal control of DNA amount (JEM1f-JEM1r).
Genomic DNA of cells collected at T=0h, T=6h and T=8h was split in two fractions, incubated at 80°C for 10 minutes in order to inactivate any remaining active DNA nuclease, The same protocol was used to determine the relative amount of CDC8 and chromosome X in surviving clones after Cas9 induction, except that total genomic DNA was not digested prior to real-time PCR. Primer couples VMS23-VMAS23 were used to amplify CDC8 and JEM1f-JEM1r for chromosome X left arm. Primers Chromo4_f and Chromo4_r were were used to amplify a region of chromosome IV as an internal control for total DNA amount. See Supplemental Table 2 for all primer sequences.

Library preparation for deep-sequencing
Approximately 10 μg of total genomic DNA was extracted and sonicated to an average size of 500 bp, on a Covaris S220 (LGC Genomics) in microtubes AFA (6x16 mm) using the were respectively set between 30% and 70% in diploids. For haploids, the threshold for minimum variant allele frequency was set at 70%. Mutations less than 10 bp away from each other were discarded to avoid mapping problems due to paralogous genes or repeated sequences. To assess microsatellite mutations, we only retained reads uniquely anchored at least 20 bp on each side of the microsatellite (Fungtammasan et al., 2015). All detected mutations were manually examined using the IGV software (version 2.3.77), and compared between all sequenced libraries for interpretation. All the scripts used in order to process data are available on github (https://github.com/sdeclere/nuclease). All Illumina sequences were uploaded in the European Nucleotide Archive (ENA), accession number PRJEB16068.

Ackowledgments
We

Declaration of interest
The authors declare no competing interests.   Under the reference locus are cartooned the different chromosomal structures observed in some of the survivors. A yeast colony that was grown in glucose was also sequenced as a control. For each clone, vertical dotted lines represent junctions of rearrangements observed, with deletion sizes indicated in base pairs. Asterisk: clone #2 showed a complex rearrangement with a local inverted duplication involving the δ16 LTR and the 3' end of the KCH1 gene 5 kb upstream SUP4. Two clones (#8 and #9) exhibit exactly the same chromosomal rearrangement at precisely the same nucleotides. Note that CDC8 is an essential gene. C: Southern blot of genomic DNA at the SUP4 locus in the rad52Δ strain. Legend as for Figure 3A. Note that for this Southern blot genomic DNA was digested with EcoRV (instead of Ssp I, see Methods), therefore the expected CTG repeat length was around 1.8 kb,

Figure Legends
instead of 1 kb. D: Southern blot of genomic DNA at the SUP4 locus in the sae2Δ strain.
Legend as for Figure 3A. used on Southern blots is shown, as well as both primer couples used to amplify the locus. In order to assess a given clone to a rearrangement type, the following rules were followed: i) when a band was detected by Southern blot, primers su47 and su48 were used to amplify the locus an sequence it. These events corresponded to types I-V. The absence of a PCR product indicated that primer su48 genomic sequence was probably deleted and therefore primers su23 and su42 were used to amplify and sequence the locus. These were classified as types IV-V events; ii) when no band was detected by Southern blot, primers su 23 and su42 were directly used to amplify and sequence the locus. These events were classified as types VI-X and XII. When no PCR product was obtained, it meant that at least one of the two primers genomic sequence was probably deleted and these events were classified as type XI. Note that this last category may also contain rare -but possible-chromosomal translocations that ended up in puting each primer in a separate chromosome, making unobtainable the PCR product.