Correction: mlh3 mutations in baker’s yeast alter meiotic recombination outcomes by increasing noncrossover events genome-wide

[This corrects the article DOI: 10.1371/journal.pgen.1006974.].


Introduction
During mismatch repair (MMR), insertion/deletion and base-base mismatches that form as the result of DNA replication errors are recognized by MutS homolog (MSH) proteins, which in turn recruit MutL homolog (MLH) proteins to form ternary complexes containing mismatched DNA, MSH factors, and MLH factors. These interactions result in the recruitment of downstream excision and resynthesis proteins to remove the error [1]. In S. cerevisiae repair of insertion deletion loops greater than one nucleotide in size primarily involves the MSH heterodimer Msh2-Msh3 and the MLH heterodimer Mlh1-Pms1 [1]. The MLH heterodimer Mlh1-Mlh3 has been shown to play a minor role in this process and can partially substitute for Mlh1-Pms1 in Msh2-Msh3-dependent MMR [2][3][4]. However, Mlh1-Mlh3 has been shown to play a major role in meiotic crossing over [5][6][7][8]. Accurate chromosome segregation in Meiosis I in most eukaryotes requires reciprocal exchange of genetic information (crossing over) between homologs [9][10][11][12]. Failure to achieve at least one crossover (CO) per homolog pair results in homolog nondisjunction and the formation of aneuploid gametes. Errors in meiotic chromosome segregation are a leading cause of spontaneous miscarriages and birth defects [13].
Yeast Mlh1-Pms1 and its human ortholog MLH1-PMS2 both exhibit an endonuclease activity that is essential for MMR [14][15]. This activity is dependent on the integrity of a highly conserved (DQHA(X) 2 E(X) 4 E) metal binding motif also found in Mlh3. Previous work demonstrated that a point mutation within this motif (mlh3-D523N) conferred mlh3Δlike defects in MMR and crossing over. These included a mutator phenotype, a decrease in spore viability to 70% (from 97% in wild-type), and a two-fold reduction in genetic map distances [5]. Consistent with these observations, Mlh1-Mlh3 is an endonuclease that nicks circular duplex DNA in vitro, and Mlh1-mlh3-D523N is defective in endonuclease activity [16][17].
Approximately 200 double strand breaks (DSBs) are induced throughout the genome in a S. cerevisiae cell in meiotic prophase, of which~90 are repaired to form COs between homologous chromosomes, with the rest repaired to form noncrossovers (NCOs; [18][19][20][21][22][23]). In this pathway a DSB, which forms on one chromatid of a homologous pair, is resected by 5' to 3' exonucleases, resulting in the formation of 3' single-strand tails on both sides of the DSB (Fig  1). One of these tails invades the other unbroken homolog and is extended and stabilized to events (class II COs). In this pathway, which does not involve the ZMM proteins, the two junctions are resolved independently by the Mus81-Mms4 endonuclease, leading to a mixture of CO and NCO products [7,8,32,33].
Genetic and physical studies summarized below support a major role for Mlh1-Mlh3 in promoting meiotic CO formation in the interference-dependent CO pathway. 1. Genetic studies performed in yeast showed that mlh1 and mlh3 mutants display approximately two-fold reductions in crossing over [7,35,36]. 2. There is significant redundancy of factors required to resolve dHJs into COs. This redundancy involves the endonucleases Mlh1-Mlh3, Mus81-Mms4, Yen1, and Slx1-Slx4 [5,7,8], with Yen1 and Slx1-Slx4 acting in cryptic or backup roles. When all four factors were removed, crossing over was reduced to very low levels; however, in an mms4 slx4 yen1 triple mutant, in which Mlh1-Mlh3 is maintained, relatively high CO levels (~70% of wild-type levels) were observed, suggesting that Mlh1-Mlh3 is the primary factor required for CO resolution in the interference-dependent CO pathway [8]. 3. MLH1 and MLH3 play critical roles in mammalian meiosis [37,38]. For example, mlh3 -/mice are sterile with an 85-94% reduction in the number of COs; germ cells in these mice fail to maintain homologous pairing at metaphase and undergo apoptosis [37,39].
Much remains to be understood on how biased resolution of dHJs in the interferencedependent pathway is achieved. A working model, supported by genetic and molecular studies outlined below, is that the STR complex and a subset of ZMM proteins process and interact with DSB repair and SEI intermediates to create a dHJ substrate that can be resolved by the Mlh1-Mlh3 endonuclease and Exo1 to form primarily COs [5-8, 16, 29, 30, 31, 36, 40-46]. In this model, the biased cleavage of a dHJ suggests coordination between the two junctions that would likely require asymmetric loading of meiotic protein complexes at each junction. However, little is known at the mechanistic level about how such coordination could be accomplished. A recent bioinformatics study by the Fung group, which involved the analysis of COassociated gene conversion patterns in yeast tetrads, suggested that Zip3, a SUMO E3 ligase, is required for biased cleavage [34]. Curiously, they found that biased resolution of dHJs was maintained in msh4 mutants. Based on these findings and other observations they propose that Msh4-Msh5 is required at the invading end of the DSB to stabilize recombination intermediates such as SEIs, while Zip3 acts to promote second-end capture steps at the ligating end of the DSB [34]. In support of this model, the ZMM heterodimer Msh4-Msh5 has been shown to promote COs in the same pathway as Mlh1-Mlh3, and human MSH4-MSH5 was shown to bind to SEI and Holliday junction substrates in vitro [7,43]. Furthermore, cytological observations in mouse have shown that MSH4-MSH5 foci appear prior to MLH1-MLH3 [37,44,47,48]. Consistent with these observations, MLH1 and MLH3 foci formation requires MSH4-MSH5 [48].
Additional support for the above model was obtained from analysis of the STR complex [8,30,31,46,49]. The STR complex has recently been labeled the master regulator of meiotic DSB repair, acting as both a positive and negative CO coordinator (Fig 1; [30,49]). Initially, the Sgs1 helicase was characterized as anti-CO because it facilitates unwinding of DSB repair intermediates to promote NCOs via synthesis-dependent strand annealing (SDSA). However, deleting either Sgs1 or Mlh3 in yeast strains that lack all other meiotic resolvases (mms4, slx4, yen1) results in a similar reduction of CO levels (~10% of wild-type levels) suggesting a pathway where Sgs1-dependent COs require Mlh1-Mlh3 [8]. Similar results were observed in mms4, slx4, yen1 strains deficient in Top3 or Rmi1 [30,31]. These data indicate that the STR complex promotes the majority of COs in conjunction with a resolvase that is not Mus81-Mms4, Slx1-Slx4 or Yen1.
A role for Exo1 in crossing over is supported by genetic studies that show Exo1 and Mlh3 acting in the same CO pathway [29]. Interestingly, Exo1's role in maintaining wild-type levels of crossing over is independent of its catalytic activity, suggesting a structural role for this pro-CO factor [29]. Consistent with the above observations, Msh4-Msh5, STR, Exo1 and Zip3 have all been shown to interact with one another and/or with Mlh1-Mlh3 [50].
In this study, we created a structure-function map of Mlh3 by analyzing 60 new mlh3 alleles in S. cerevisiae. Five alleles predicted to disrupt the Mlh1-Mlh3 endonuclease motif conferred defects in both MMR and crossing over, providing further support that endonuclease activity is required for both functions. Importantly, we identified five mlh3 mutations that specifically disrupted MMR, and one mutation that specifically disrupted crossing over. By performing biochemical and genetic analyses of the separation of function Mlh1-mlh3 complexes we suggest that the defects seen in our mutants can be explained by a weakening of protein-protein interactions, which can be tolerated in meiosis, but not MMR. Importantly, our high-resolution recombination mapping of these mutants revealed ways in which defective meiotic components can alter the fate of meiotic recombination intermediates.

Rationale for site-directed mutagenesis of MLH3
Mlh3 contains a highly conserved N-terminal ATP binding motif, a dynamic and unstructured motif known as the linker arm, and an endonuclease active site that overlaps with a Cterminal Mlh1 interaction domain [51]. We performed a clustered charged-to-alanine scanning mutagenesis of the S. cerevisiae MLH3 gene to create 60 mlh3 variants (Fig 2; S1-S3 Tables). Charged residues were considered "clustered" if there were at least two charged residues, consecutive or separated by at most one amino acid, within the primary sequence of Mlh3. Such a directed approach, in the absence of a complete crystal structure, is aimed at targeting the surface of a protein where clusters of charged residues likely reside, while minimizing changes within the interior. In this model, replacement of a charged patch from Mlh3's surface with alanine residues would disrupt protein-protein or protein-DNA interactions without affecting Mlh3 structure. This unbiased mutagenesis has been successfully applied to study the functional domains of several proteins [52,53], and has provided a comprehensive view of the functional organization of MLH1 [54]. As shown below, we identified a subset of mutations that caused strong defects in either MMR or crossing over, but not both, likely through disrupted interactions with Mlh1 and other MMR and meiotic CO factors.

Structure-function analysis of Mlh3
We analyzed the effect of mlh3 mutations on MMR in vegetatively grown cells and on meiotic COs in diploids induced to undergo sporulation. For MMR we employed the lys2-A 14 reversion assay to assess the mutation rate in mlh3 haploid strains (S1 Table; [55]). In this assay, the median reversion rate of mlh3Δ is six-fold higher than wild-type (Fig 3B; S1 Fig; Table 1; [5,6]). To measure meiotic crossing over we crossed mutant mlh3 strains to mlh3Δ strains (MLH3 is haplosufficient [6]) to form diploids that were then sporulated (S2 Table). The resulting tetrads were directly visualized for chromosome VIII CO events using a spore autonomous fluorescence assay ( [56]; Fig 3A). In mlh3Δ strains we observed a more than two-fold decrease in crossing over, as measured by percent tetratype, compared to wild-type (Fig 3B;  S1 Fig; Table 1). Similar effects of the mlh3Δ mutation on crossing over were seen at other genetic intervals [5][6][7][8]. It is important to note that nonparental ditype (NPD) events were not scored because they cannot be distinguished from Meiosis I nondisjunction events [56]. Similar to work performed on a smaller number of mlh3 alleles and a structure-function analysis of MLH1, we found that MLH3 MMR functions were more sensitive to mutagenesis than CO functions (S1 Fig; [6 , 54]). Phenotypes exhibited by mlh3 strains containing mutations in the ATP-binding motif suggested that this region plays a more critical role in MMR compared to crossing over. However, a region just beyond the ATP-binding domain appeared insensitive to mutagenesis. A null phenotype for both functions was observed in strains bearing mutations in endonuclease motifs, further confirming that endonuclease activity is essential for MMR and crossing over (S1 Fig; Table 1; [5,8,16]).
Comparison of the MMR and CO assay results for each individual allele led to the identification of six separation of function mutations, defined as showing strong defects in one function (e.g. MMR) relative to another (e.g. CO), in the Mlh3 ATP-binding motifs, N-terminal domain beyond the ATP-binding motifs, linker arm, and the interaction domain ( Fig 3B;  S1 Fig, indicated by stars). One of these alleles (mlh3-32) conferred a nearly wild-type phenotype for MMR and a null phenotype for crossing over on chromosome VIII (hereafter referred to as MMR + , CO -). The remaining five mutations (mlh3-6, mlh3-23, mlh3-42, mlh3-45, and A. Spore-autonomous fluorescent protein expression was used to quantify crossing over [56]. Shown is the starting parental configuration on chromosome VIII with a map distance of 20 cM separating the red fluorescent protein (RFP) marker and the blue fluorescent protein (CFP) marker. Percent tetratype at this interval in wild-type meiosis is 36.7%. B. MMR (top) and CO (bottom) phenotypes for MLH3 and mlh3 null (mlh3Δ), separation of function, endonuclease, and C-terminal tail (mlh3-60) mutants. Mismatch repair was measured using the lys2-A 14 reversion assay [55] and crossing over was measured using the assay depicted in panel A. Bars represent the median reversion rates (error bars based on 95% confidence intervals) and percent tetratype normalized to MLH3 (1X). For mismatch repair (top), bars represent reversion rates of at least 10 independently tested cultures from two independently constructed strains presented here normalized to MLH3 median rate of 1X = 1.43x10 -6 (n = 140). For crossing over (bottom), bars represent percent tetratype of at least 250 tetrads from two independently constructed strains presented here normalized to MLH3 percent tetratype 1X = 36.7% (n = 226; Table 1 and S2 File). Blue and red dotted lines represent MLH3 and mlh3Δ respectively. C. mlh3-42, -54 weaken Mlh1 interaction yet maintain crossover function. Yeast two-hybrid interactions between lexA-Mlh1 (target) and Gal4-Mlh3 (amino acids 481-715; prey) or Gal4-mlh3-39, -40, -41, -42, -45, -54, -60 derivative constructs, as measured in the ONPG assay for βgalactosidase activity. Error bars indicate standard error of mean from at least three independent assays (S2 File). mlh3 separation of function alleles indicated in green font. https://doi.org/10.1371/journal.pgen.1006974.g003 Structure-function analysis of Mlh1-Mlh3 Structure-function analysis of Mlh1-Mlh3 mlh3-54) conferred null MMR phenotypes and nearly wild-type levels of crossing over (hereafter referred to as MMR -, CO + ). As indicated below, mlh3-23 displayed different CO phenotypes in different strain backgrounds, and so its designation as a separation of function allele is less clear. The phenotypes observed in the separation of function mutants may result from a defect in DNA binding/substrate specificity, endonuclease activity, interactions with specific MMR and meiotic CO factors, or changes in protein conformation. It is important to note that a co-crystal structure of the N-terminal domain of E. coli MutL (LN40) and E. coli MutS was recently solved. This work showed that conformational changes license MutS-MutL interaction and are essential for MMR [57,58].
The endonuclease active sites in Mlh3 and Pms1 appear to be similar S. cerevisiae Mlh1-Pms1 and Mlh1-Mlh3 and human MLH1-PMS2 display latent endonuclease activities dependent on the integrity of a highly conserved metal binding motif DQHA(X) 2 E (X) 4 E [14][15][16][17]. This motif is critical for Mlh3's MMR and meiotic functions [5]. Two additional motifs were implicated in MLH family endonuclease function based on sequence alignment: ACR and C(P/N)HGRP [59]. In the Mlh1-Pms1 C-terminal domains crystal structure, five Pms1 residues, located in the three conserved motifs (H703, E707, C817, C848, H850), form a metal binding site through folding of the Pms1 C-terminal domain (Fig 2C; [51]). This Structure-function analysis of Mlh1-Mlh3 organization was also seen in the crystal structure of the C-terminal domain of B. subtilis endonuclease MutL (all but H703 are conserved; [60]). We performed a sequence alignment of Mlh3, Pms1, and B. subtilis MutL and found all three possess conserved metal binding motifs, with the following five residues predicted to form the endonuclease active site in Mlh3: H525, E529, C670, C701, and H703. In addition, we constructed a homology model of S. cerevisiae Mlh3, and found the C-terminal domain can potentially fold in a similar manner to Pms1 such that these five conserved residues form a single putative metal binding site (Fig 2C). These residues were targeted for site directed mutagenesis of MLH3 (Fig 2B, shown in red). As shown in S1 Fig (represented in the C-terminal domain by dotted white squares) and Fig 3B and Table 1, mutations in the putative conserved metal binding motifs of Mlh3 (mlh3-39(D523A, H525A), -40 (D528A, E529A, R530A), -57(C670A), -58(C701A), and -59(H703A)) conferred null phenotypes for MMR and crossing over, indicating they are essential for Mlh3 function. Thus, these genetic data, combined with the high sequence homology, suggest that H525, E529, C670, C701, and H703 in the C-terminal domain of Mlh3 form the catalytic active site ( Fig 2C).
The endonuclease motifs in Mlh3 overlap with the C-terminal Mlh1 interaction domain. To determine if mutations in the Mlh3 endonuclease motifs disrupted interaction with Mlh1, three alleles spanning the DQHA(X) 2 E(X) 4 E endonuclease motif (mlh3- 39, -40, and -41) were analyzed by yeast two-hybrid for interaction with Mlh1. We also tested these alleles because a previously characterized mutation in the DQHA(X) 2 E(X) 4 E endonuclease motif (mlh3-E529K) disrupts Mlh1-Mlh3 interactions [5]. As shown in Fig 3C, these mutations disrupted Mlh1-Mlh3 interactions, possibly by altering the endonuclease active site structure. This idea is supported by the Mlh1-Pms1 crystal structure. In this model, heterodimer stability is maintained through interactions between the C-terminal domain of Mlh1 and the endonuclease active site of Pms1 [51]. We cannot rule out the possibility that the null phenotypes observed for MMR and crossing over in mlh3-39, -40, and -41 were caused by specifically mutating residues that comprise the Mlh1-Mlh3 dimerization interface without causing a gross disruption in protein folding.
The Mlh1-Pms1 C-terminal domain structure reveals three patches constituting the heterodimerization interface of Mlh1-Pms1 [51]. Patch I is a pseudosymmetric hydrophobic core, Patch II is composed of the last 12 residues of Pms1 and contributes two salt bridges, and Patch III involves the C-terminus of Mlh1 and contributes to the Pms1 metal binding site [51]. Patches I and III are likely maintained in the Mlh1-Mlh3 heterodimerization interface, but Mlh3 lacks the last 11 residues that comprise the bulk of Patch II. This finding gives a likely explanation for partial disruption of the Mlh1-Mlh3 complex when we attempted to analyze it further by gel-filtration [16]. We hypothesized that restoring Patch II to the Mlh1-Mlh3 interaction interface will strengthen this interaction. We engineered a fusion construct of Mlh3 carrying the last 11 residues of Pms1 (mlh3-60, Fig 2B in red). As shown in Fig 3C, when we inserted the last 11 residues of Pms1 after the C-terminal residue of Mlh3, we observed a striking increase in the strength of the interaction between Mlh1 and Mlh3 as measured in the yeast two-hybrid assay (2.6 ± 0.5 Miller units of β-galactosidase activity for wild-type Mlh1-Mlh3 compared to 14.2 ± 2.1 for Mlh1-mlh3-60). We initially hypothesized that such an enhanced interaction would be detrimental to MMR because it would sequester Mlh1 from the major MMR endonuclease Pms1. Surprisingly we did not observe a significant effect of the mlh3-60 mutation on MMR or on crossing over ( Fig 3B; Table 1), suggesting that strengthening the interaction between Mlh1-Mlh3 does not affect formation of the Mlh1-Pms1 heterodimer.

mlh3-42 and mlh3-54 weaken interaction with Mlh1 yet maintain wildtype levels of crossing over
Three of the five MMR -, CO + alleles (mlh3-42, -45, and -54) contained mutations that mapped to the Mlh1 interaction interface. We performed a two-hybrid assay to test whether these mutations affected Mlh1-Mlh3 dimerization. The mlh3-45 mutation did not alter Mlh1-Mlh3 interactions; however, both the mlh3-42 and mlh3-54 mutations disrupted this interaction ( Fig  3C). While such a result could explain the null MMR phenotype conferred by mlh3-42 and mlh3-54, it does not explain why these strains are functional for meiotic crossing over. One explanation is that additional pro-CO factors act as structural scaffolds to stabilize the weakened Mlh1-mlh3 heterodimer, thus allowing it to perform its function at dHJs. Several observations support this idea: 1. The pro-CO factors Msh4-Msh5, Sgs1-Top3-Rmi1, Zip3, and Exo1 interact with one another and/or with Mlh1-Mlh3 [29,40,41,42,61,62]. 2. Studies in mice showed that MLH1 and MLH3 do not form a complex until mid to late pachytene; at early to mid pachytene, only MLH3 foci are seen [37,44]. 3. Exo1's role in crossing over is independent of its enzymatic activity; it is suggested to play a structural role, acting as a platform for pro-CO factors [29]. Together these observations support the presence of a resolvase complex at CO sites that regulates the endonuclease activity of Mlh1-Mlh3 (see Discussion). Alternatively, a weak Mlh1-Mlh3 interaction defect is sufficient to inhibit a yeast-two hybrid interaction, but not affect meiotic recombination if the strength of the Mlh1-Mlh3 interaction is not a limiting factor for CO resolution.
Cumulative genetic distance and spore viability measurements confirm fluorescent assay results for mlh3 separation of function alleles The meiotic CO phenotype for the separation of function mutants was determined at four consecutive intervals in chromosome (XV) using traditional genetic map distance analyses ( Fig  4A; [5][6][7]). The overall effect of mlh3 mutations in crossing over on chromosome XV was similar to that determined on chromosome VIII (Figs 3B and 4B; Table 1; S4 Table). mlh3-6, mlh3-42, mlh3-45, and mlh3-54 appear similar to wild-type; mlh3-23 displays an intermediate phenotype, and mlh3-32 appears similar to the null ( Fig 4B; S4 Table). Thus, these mutants, phenotypes of which are summarized in Table 2, were confirmed as separation of function alleles, and are candidates for in-depth characterization and high-resolution recombination mapping.
It is important to note that the spore viability and genetic distance measurements of the six separation of function alleles indicated that CO levels, represented by genetic distance measurements in Chr. XV, can be reduced from 115 cM to~70 cM without compromising spore viability ( Fig 4C; Table 2). A similar observation was made with msh4/5 hypomorph alleles where crossing over in the four same intervals in Chr. XV could be reduced to~50 cM without affecting spore viability [63][64]. Together, these observations and the high resolution mapping below indicate that the baker's yeast meiotic cell does not require the full amount of COs maintained by CO homeostasis (~90; see [65]) for accurate chromosome segregation and to form viable spores.

High-resolution recombination maps illustrate unexpected effects of mlh3 mutants on resolving meiotic recombination intermediates
We characterized three mlh3 alleles (mlh3-23-MMR -, CO + , mlh3-32-MMR + , CO -, mlh3-D523N) in a genome-wide meiotic recombination assay [66]. mlh3-23 and mlh3-32 were chosen as they confer stronger defects in one function relative to the other. The mlh3-D523N endonuclease mutation contains an aspartic acid to asparagine substitution in the DQHA(X) 2 E(X) 4 Table). The solid circle indicates the centromere. The distances between markers are not drawn to scale. The actual physical and genetic distances in the wild-type diploid are given numerically for each interval and for the entire region between CENXV and HIS3 [7]. B. Cumulative genetic distances between URA3 and HIS3 markers from tetrads of MLH3 and indicated mlh3 variants. Each bar is further divided into sectors that correspond to the four genetic intervals that span URA3-HIS3 (S4 Table). C. Spore viabilities are plotted vs. genetic map distances from panel B for MLH3 (dark blue), mlh3Δ (red), and the separation of function mutants (light blue). Yellow diamonds represent data from Sonntag Brown et al. [6] (S2 File).
Seven four-viable spore tetrads each of mlh3-23 and mlh3-32, and ten four-viable spore tetrads each of mlh3-D523N, mlh3Δ, and MLH3 were sequenced (S5 Table). The sequence data for all the above mlh3 point and null mutants along with the control strain are available from the National Centre for Biotechnology Information Sequence Read Archive under the accession numbers SRP096621 and SRP110341. The segregation of the SNPs in all 44 tetrads is shown in S1 File. The CO and NCO counts for all tetrads and the average CO and NCO counts per chromosome are shown in S6 and S7 Tables, respectively.
As described below, high-resolution recombination mapping analysis using groupEvents [34] revealed overall patterns of average CO and NCO events that grouped mlh3-23 close to wild-type, and mlh3-D523N and mlh3-32 close to mlh3Δ (Fig 5). Importantly, this analysis showed that mlh3 point and null mutants displayed, compared to wild-type, genome-wide increases in NCO events. We did not obtain any evidence that the number of DSBs increased in mlh3 point mutants based on measuring the total number of interhomolog (IH) events ( Fig  6A); such an increase would have provided a simple explanation for why an increase in NCO events was observed, though it is important to note that small changes in total events are challenging to interpret, partly because NCO events can escape detection due to a lack of a polymorphic marker, and we do not have a good estimation of how many such invisible events exist, or how they could be altered by differences in conversion tract length. Thus, we recognize that analyzing total events does not provide a particularly good proxy for DSB (e.g. [69]) levels. Together, these data provide evidence for mlh3 mutants altering the resolution of meiotic recombination intermediates (see Discussion). mlh3-23, mlh3-32, and mlh3-D523N display distinct CO phenotypes. We used groupEvents ( [34]; 5 kb threshold for merging nearby events) to categorize all IH events, focusing primarily on the major CO (E2, E3) and NCO events (E1; Figs 1 and 5B), but also examining minority NCO (E4) events that can provide information on whether events resulted from defective mismatch repair or multiple invasion intermediates (Figs 5 and 6; S3 Fig; Table 3; S9  Table; Methods).
Wild-type and mlh3-23 displayed the same average number of CO events (E2+E3) (p = 0.99), and mlh3Δ, mlh3-D523N, and mlh3-32 displayed CO levels lower than wild-type (p <0.01). mlh3-D523N and mlh3Δ displayed similar average CO levels (p = 0.23), but mlh3-32 displayed levels higher than mlh3Δ (p = 0.05) (Fig 5C and Table 3). An analysis of average CO counts per chromosome was consistent with these observations; mlh3-23 had a CO distribution similar to wild-type, and mlh3-32 and mlh3-D523N had distributions similar to mlh3Δ (S2 Fig). Significant reduction in crossovers were observed primarily on medium (0.5 to 0.9 Mb-II, V, VIII, X, XI, XIV) and large (>0.9 Mb-IV, VII, XII, XIII, XV, XVI) chromosomes in tract, and not within 5 kb of another CO or NCO), simple CO with discontinuous gene conversion tracts (E3; same definition as for E2, except with one or more gene conversions within 5 kb and on one of the same chromatids as the CO chromatid), and discontinuous NCOs (E4; two or more NCOs consecutively on one chromatid, with 2:2 marker segregation separating them) as presented in Oke et al. [34]. C. Crossover (CO, E2+E3) and noncrossover (NCO, E1) counts per meiosis for wildtype, mlh3-23, mlh3-32, mlh3-D523N, and mlh3Δ. The minimum, first quantile, median, third quantile and maximum count are indicated in the box plot. The ratio of CO to NCO events is presented above the box plots. The proficiency of the mlh3 alleles in mismatch repair is shown as +, MMR proficient, or -, MMR deficient. D. Average number of simple NCO (E1) and NCO with discontinuous tract (E4) events per tetrad (+/-standard error). * p 0.05 compared to wild-type (Table 3 and Table) were significantly longer (p 0.03) for mlh3Δ (2.19 kb) compared to MLH3 (1.79 kb) and mlh3-32 (1.84 kb), but were not significantly longer (p ! 0.18) for mlh3-23 (1.96 kb), and mlh3-D523N (1.96 kb). We found this interesting because MLH3 and mlh3-32 mutants appeared similarly functional for MMR, whereas mlh3Δ, mlh3-23, and mlh3-D523N all displayed MMR defects (Table 1). These observations encouraged us to test if Mlh1-Mlh3 MMR functions affected the processing of gene conversion tracts (see below and Discussion). mlh3-23, mlh3-32, mlh3-D523N and mlh3Δ mutants. The average numbers of NCOs (E1 events) in mlh3Δ (49), mlh3-23 (51), mlh3-32 (56), and mlh3-D523N (50) were significantly higher than the wild-type value (41; t-test, p = 0.01, 0.0067, 0.0012, 0.0044, respectively; Table 3; Fig 5D; S2C Fig). This was unexpected because CO levels were similar to wild-type for mlh3-23 mutants and similar to mlh3Δ for mlh3-32 and mlh3-D523N mutants (see Discussion). Compared to mlh3Δ, the average number of NCO events in mlh3-23, mlh3-32, and mlh3-D523N was similar (p = 0.48, 0.080, 0.70, respectively). Significant increases in NCOs on chromosomes were observed for mlh3-32 (X and VII), mlh3-DN (VIII, X, XIII, VII, XV, IV), mlh3Δ (III, VIII, X, XIV, XIII, XVI, XII, VII, XV, IV). No increase in NCOs on specific chromosomes was observed for mlh3-23 (S2B Fig). It is possible that some of the increase in NCOs seen in mlh3 mutants was due to the presence of discontinuous gene tracts that arose from the same initiating DSB. Such discontinuous tracts could have resulted from defective MMR but are not included in the E1 NCO class above. The E4 class (Fig 5B) is a minority discontinuous NCO event that is hypothesized to arise from defective MMR or multiple invasion intermediates [34]. As shown in Fig 5D, none of the E4 events in the mlh3 mutants were significantly different from wild-type (p = 0.33 to 0.80). This suggests that discontinuous tracts are unlikely to significantly contribute to the increased level of NCO events seen in mlh3 mutants.  for mlh3-23, mlh3-32, mlh3-D523N, and mlh3Δ  Structure-function analysis of Mlh1-Mlh3

Genome-wide increase in noncrossovers in
The average NCO (E1) tract lengths were similar to wild-type (1.46 kb) in all of the mlh3 mutants analyzed (1.34 to 1.50 kb; p = 0.23 to 0.91; S3 Fig). The NCO tract length distribution suggested an enrichment of short NCO tract lengths (<1 kb) in the mlh3 mutants compared to wild-type (Fig 6B). We found this potentially interesting because Oke et al. [34] identified such an enrichment in sgs1 mutants; they hypothesized that the short NCO class arose from joint molecules or joint molecule-like intermediates that were resolved by the activity of structureselective nucleases (SSNs). However, both Kolmogorov-Smirnov and Wilcoxon Rank-Sum tests indicated that an apparent enrichment of short NCO tract lengths in mlh3 mutants, compared to wild-type, was not significant. A similarly weak pattern (an insignificant increase in mlh3 mutants compared to wild-type) was seen for the E5A class (S9 Table, [34]), which is thought to represent double COs and is predicted to arise if joint molecule resolution is not biased (Class II COs resolved by SSNs; Fig 1).  Table), and each event type (E1-E7) as a fraction of the total, in MLH3, mlh3-23, mlh3-32, mlh3-D523N, and mlh3Δ mutants. * p 0.05 compared to wild-type (S2 File). B. Average number of tracts ordered by size for simple CO (E2) and simple NCO (E1) events. Median CO and NCO tract sizes are also presented, and events were assigned as described by Oke et al. [34] (S2 File). The Y axis represents pooled data from all tetrads, normalized by dividing the number of tetrads. https://doi.org/10.1371/journal.pgen.1006974.g006 Structure-function analysis of Mlh1-Mlh3 One explanation for the increase in NCO events seen in mlh3Δ, mlh3-23, mlh3-32 and mlh3-D523N mutants is that these mutants experienced meiotic progression delays that resulted in the continued accumulation of NCOs, possibly through increased DSB formation. Increases in NCO events and a meiotic delay were observed in ndt80 and the ZMM zip1, zip3 and msh5 mutants as a result of impeding feedback circuits that inhibit DSB formation [27,30,49,71,72]. To investigate this possibility, we examined total interhomolog (IH) events, which consists of the average of all joint molecule (E2, E3, E5, E6, E7) and NCO (E1, E4) events, and serves as a rough estimate of double-strand break levels [34]. As shown in Fig 6A and S9 Table, the IH value for wild-type was 135.7 events per tetrad. None of the mlh3 mutants displayed a significantly higher value than wild-type, but mlh3-D523N (122.6 events, p = 0.011) and mlh3Δ (114.2 events, p = 0.002) displayed lower values.
We also examined meiotic progression in MLH3, mlh3Δ, mlh3-32, mlh3-23 and mlh3-D523N SK1 strains by measuring the completion of the first meiotic division. This would be difficult to do in S288c/YJM789 strains because they do not show the highly synchronous and efficient meiotic progression profile seen in SK1. As shown in S4 Fig, MLH3, mlh3Δ, mlh3-32, mlh3-23, and mlh3-D523N mutants showed similar kinetics for completion of at least the first meiotic division (MI+MII). These data suggest that the increase in NCO events in mlh3Δ, mlh3-32, mlh3-23, and mlh3-D523N cannot simply be explained by an increase in DSB levels, MMR defects, or a meiotic progression delay.
In summary, the increase in NCO events in mlh3 mutants is consistent with a possible role for SSNs and/or the STR complex (see Discussion), and is unlikely to be explainable by an increase in DSBs or MMR defects.
Mlh1-mlh3-32 and Mlh1-mlh3-45 display wild-type endonuclease activities but only Mlh1-mlh3-32 endonuclease is stimulated by Msh2-Msh3 We examined Mlh1-mlh3 mutant complexes for endonuclease activity [16,17], focusing on opposite separation of function mutants Mlh1-mlh3-32 (MMR + , CO -), Mlh1-mlh3-6, and Mlh1-mlh3-45 (MMR -, CO + ). Mlh1-mlh3-45, located in the C-terminal Mlh1 interaction domain, was chosen because it is the only separation of function mutant in that domain that displayed wild-type Mlh1-Mlh3 interactions as measured in the two-hybrid assay ( Fig 3C). As shown in Fig 7 and S5 Fig, all three mutant complexes purified as heterodimers and display endonuclease activities similar to wild-type. When this work was initiated we thought that separation of function mutant complexes might show endonuclease defects indicating that this activity is more critical for MMR or crossing over, or show no endonuclease defects because mutant complexes were defective in interacting with MMR or CO specific factors. Our finding that all three mutants have enzymatic activity comparable to wild-type is consistent with the interaction defect model (see below). The mlh3-6 mutation maps close to conserved sites in the ATP binding motif (S1 Fig). We then tested whether the mutant complex displayed a defect in ATPase activity. As shown in S5C Fig, Mlh1-Mlh3 and Mlh1-mlh3-6 displayed similar ATPase activities.
Because Mlh1-Mlh3's endonuclease activity is enhanced by Msh2-Msh3 [16], we tested whether the opposite separation of function phenotypes of Mlh1-mlh3-32 and Mlh1-mlh3-45 could be explained by defective interactions with MSH complexes. As shown in Fig 7D, Mlh1-mlh3-32 endonuclease activity but not Mlh1-mlh3-45 could be stimulated by Msh2-Msh3. These data are consistent with the MMR -, CO + phenotype exhibited by mlh3-45 mutants resulting from a defect in interacting with the MMR component Msh2-Msh3, and the mlh3-32 mutant likely being defective in interactions with meiosis-specific factors. In a first step to test if mlh3-32 mutants are defective in interactions with meiotic CO factors, we determined if the mlh3-32 mutation is dominant; such a phenotype could provide hints on the nature of the mlh3-32 meiotic defect. We mated EAY3552 (mlh3-32, CEN8Tomato::LEU2) to EAY3339 (relevant genotype MLH3, THR1::m-Cerulean-TRP1; S1 and S2 Tables), and the sporulated progeny displayed a tetratype frequency similar to wild-type (40.9%, n = 252 tetrads), indicating that mlh3-32 is recessive.
Sgs1 but not sgs1-hd overexpression differentially affects spore viability in mlh3Δ vs. mlh3-32 As presented in Fig 1, the STR complex can act as both a negative and positive regulator of CO formation in meiotic prophase [8,30,31,46,49]. In its role as a negative regulator, STR is thought to prevent the formation of aberrant recombination structures by disassembling branched recombination intermediates to form early NCOs via synthesis dependent strand Structure-function analysis of Mlh1-Mlh3 annealing (SDSA), or by re-forming the DSB intermediate. In its role as a pro-CO factor STR promotes stabilization of ZMM complexes on recombination intermediates, leading to the resolution of dHJs by an interference-dependent CO pathway (class I) that requires the Mlh1-Mlh3 endonuclease. In sgs1Δ mutants COs have been shown to be ZMM independent [73]. Strand invasion intermediates that escape STR disassembly are thought to be resolved as COs or NCOs using an alternative interference-independent CO pathway (class II) that involves the SSNs Mus81-Mms4, Yen1, and Slx1-Slx4.
The increase in NCO events seen in mlh3 mutants is suggestive of their formation through SSN or STR activities. To test for genetic interactions between SGS1 and MLH3, we expressed SGS1 via its native promoter on a 2μ multi-copy vector. Sgs1 overexpression enhanced the mlh3Δ spore viability defect (Fig 8: 76% in mlh3Δ+2μ vs. 69% in mlh3Δ+SGS1-2μ; p<0.05,χ 2 test) and conferred a more apparent MI nondisjunction pattern (an excess of 4, 2, 0 viable Distribution of viable spores in tetrads of MLH3, mlh3Δ, mlh3-23, mlh3-32, and mlh3-D523N strains containing no insert (2μ), pSGS1-2μ, or psgs1-hd-2μ (helicase defective mutant). In all plots, the horizontal axis corresponds to the classes of tetrads with 4, 3, 2, 1 and 0 viable spores, and the vertical axis corresponds to the frequency of each class given in percentage. The overall spore viability (SV) and the total number of spores counted (n) are shown (S2 File). *p<0.05 (χ 2 test), comparing spore viability for the indicated strain transformed with a 2μ SGS1 plasmid to that of the same strain transformed with a 2μ no insert plasmid.

Discussion
We performed a structure-function analysis of Mlh3, a factor that acts in both MMR and meiotic crossing over. This work was pursued because little is known about how Mlh1-Mlh3 acts as a meiotic endonuclease. This is due in part to Mlh1-Mlh3 sharing little in common with the well-characterized structure-selective endonucleases (SSNs; Mus81-Mms4, Slx1-Slx4, and Yen1) in terms of homology and intrinsic behavior in vitro (reviewed in [50]). Obtaining new mechanistic insights has been complicated by the fact that Mlh1-Mlh3 can bind to model HJ substrates, but cannot cleave them, and by genetic studies suggesting that Mlh1-Mlh3 acts in concert with other pro-CO factors [16,17,50,74]. Recent work has suggested that multiple Mlh1-Mlh3 heterodimers are required to activate the endonuclease and that the complex is at least partially inhibited by incorporation of a DNA secondary structure that is not part of a continuous homoduplex substrate [74]. These data suggest that other protein factors likely recruit and position Mlh1-Mlh3 complexes during meiotic recombination. The identities of these factors are for the most part known, though it is not understood how they directly contribute to Mlh1-Mlh3's ability to nick DNA in the directed manner required to generate COs.
Our analysis of mlh3-32 (MMR + , CO -) and mlh3-45 (MMR -, CO + ), support the above hypothesis that protein-protein interactions are critical for directing Mlh1-Mlh3 endonuclease activity ( Table 2; Fig 7). Mlh1-Mlh3 has been shown genetically to act downstream of Msh4-Msh5 [40,41,44,47]; this order of events is analogous to steps in DNA MMR where MLH acts following MSH recognition [14,75]. As outlined in the introduction, Msh4-Msh5, STR, Exo1 (independent of its enzymatic activity) and Zip3 have been classified as pro-CO factors, and have all been shown to interact with one another and/or with Mlh1-Mlh3 (reviewed in [50]). Our biochemical studies are consistent with Mlh1-mlh3-45 having interaction defects that prevent its endonuclease activity from being stimulated by Msh2-Msh3 in MMR. We hypothesize that MLH complexes interact with MSH complexes via a common mechanism, and that defective interactions with Msh2-Msh3 are also indicative of defective interactions with Msh4-Msh5, but during meiotic CO resolution, additional factors act in concert to strengthen a possibly weakened Msh4-Msh5-Mlh1-mlh3-45 interaction. This model also helps explain why we identified several MMR -CO + mlh3 mutants (mlh3- 42 and -54) in which the mutant mlh3 protein fails to interact with Mlh1. For the Mlh1-mlh3-32 complex, the MSH interaction and enhancement is retained, but interaction with other critical meiotic factors is likely lost, possibly resulting in an unstable complex that cannot resolve dHJs.

Mlh1-Mlh3 in DNA mismatch repair
In MMR the asymmetric loading of PCNA by the RFC complex is thought to direct the endonuclease activity of MutLα (Mlh1-Pms1 in S. cerevisiae, MLH1-PMS2 in humans) to act in strand-specific repair [76]. Additional studies suggest that specific protein-protein interactions influence and activate MLH endonuclease activity, and direct nicking to a specific location. For example, in vitro studies performed with yeast proteins showed that RFC-loaded PCNA can activate Mlh1-Pms1 but not Mlh1-Mlh3 endonuclease on circular plasmids (Mlh3 lacks a PCNA binding motif present in Pms1; [15-17; 34]. Our finding that endonuclease active site residues are highly conserved between Mlh1-Mlh3 and Mlh1-Pms1, which has no role in meiotic crossing over, suggests that the different functions of the two complexes are a result of the different protein-protein interactions. Data from the Crouse lab suggest that Mlh1-Mlh3 acts in conjunction with Mlh1-Pms1 in Msh2-Msh3 dependent MMR [4]. This observation helps address how Mlh1-Mlh3 is involved in MMR in the absence of a PCNA interaction. In one scenario consistent with the above, Mlh1-Mlh3 is recruited and activated by Msh2-Msh3, and also forms a complex with Mlh1-Pms1, which can be directed by PCNA to promote efficient repair. If dimerization between Mlh3 and Mlh1 is weakened, the ability to be recruited by Msh2-Msh3 and interact with Mlh1-Pms1 is likely inhibited, creating a defect in MMR that is minor because Mlh1-Pms1 is still active. For meiotic crossing over, a relatively slow process compared to DNA MMR at the replication fork, we suggest that a weakened dimer can be compensated for by interactions with other meiotic factors (e.g. Msh4-Msh5, Exo1 and STR). Thus, our work provides further motivation to examine Mlh1-Mlh3 activity on recombination substrates in the presence of pro-CO factors.

Mlh1-Mlh3 in meiotic recombination
Current meiotic DSB repair models postulate an enzymatic role for Mlh1-Mlh3 in the class I CO pathway after DSB intermediates have been captured and stabilized by the ZMM proteins [8,30,49]. In these models DSB intermediates that escape capture by ZMM proteins are resolved into class II COs or NCOs by SSNs, though NCOs mostly arise from the action of the STR complex through synthesis dependent strand annealing (SDSA; [30,49] ; Fig 1).
We observed a genome-wide increase in NCOs compared to wild-type in mlh3-23, mlh3-32, mlh3-D523N, and mlh3Δ mutants. This suggested to us that in mutants defective in Mlh1-Mlh3 functions, dHJs formed in the ZMM pathway become susceptible to the actions of SSNs to form COs and NCOs. They could also be acted upon by STR through unwinding and convergent migration of the two HJs until a single pair of crossing strands in a hemicatenane can be removed by topoisomerase activity, giving rise to NCOs [30,31]. In this model Mlh1-Mlh3, in concert with the ZMM proteins, acts to protect recombination intermediates from STR. In support of this idea, we identified genetic interactions between Mlh1-Mlh3 and Sgs1 that suggest an active interplay (Mlh1 and Mlh3 both interact with Sgs1 [42,77]) between these factors: Sgs1 overexpression decreased the spore viability of mlh3Δ strains but not MLH3, mlh3-23, or mlh3-D523N strains, where a structurally intact Mlh1-Mlh3 complex is thought to be present, and modestly increased the spore viability of mlh3-32 mutants (Fig 8).
We also observed an increase in tract length for gene conversion events associated with COs in mlh3Δ compared to wild-type (S3 Fig; S8 and S9 Tables). The latter observation is interesting because recent work from Duroc et al. [78] provided evidence that another MLH complex, Mlh1-Mlh2, which lacks endonuclease activity, acts to limit the extent of meiotic gene conversion. They found that gene conversion tract lengths associated or not associated with COs increased from~1 kb in wild-type to~2 kb in mlh2Δ. Thus, regulating gene conversion tracts may be a common property of the MLH complexes that act in meiotic recombination, and is consistent with the fact that Mlh1 has been shown to physically interact with helicases that can modulate heteroduplex extension/migration. An alternative explanation for the increase in NCO events in mlh3 mutants is that delays in meiotic progression result in the accumulation of NCO events as the result of increased DSB formation [72]. However, we did not observe such delays in any of the mlh3 mutant backgrounds or any dramatic change in DSB formation based on an analysis of total interhomolog events in mlh3 mutants, though a direct measure of DSB levels is critical to resolve this issue (Fig 6A; S4 Fig; S9 Table; [69]).

Mlh3's linker arm is critical for its meiotic function
MLH proteins act as dimers and contain long unstructured linkers that connect the N-and Cterminal domains of each subunit. These linkers vary in length and are resistant to amino acid substitutions [54]. Previous work showed that the Mlh1-Pms1 heterodimer undergoes large global conformational changes in an ATP binding and hydrolysis cycle [79]. In this cycle the linkers act as arms that can switch between extended and condensed states. These conformational changes are hypothesized to be important to expose different domains of the heterodimer for new protein-protein or protein-DNA interactions in addition to mediating the timing of these interactions [79], and have also been implicated in B. subtilis MutL for "licensing" its latent endonuclease activity [60]. In addition, a series of truncation mutants in Mlh1-Pms1 indicate that the Pms1 linker arm appears more important than the Mlh1 linker arm for DNA binding [80]. Extending these ideas to Mlh1-Mlh3, it is interesting to note that the MMR + ,CO -mlh3-32 allele maps to the unstructured linker, suggesting that this domain is particularly important in crossing over (S1 Fig; Table 1), possibly facilitating interactions with CO promoting factors that in turn direct and position Mlh3's endonuclease activity on recombination substrates. It is important to note that Claeys Bouuaert and Keeney [81] identified mutations in the Mlh3 linker domain based on a biochemical analysis of Mlh1-Mlh3 that overlap with residues mutated in the mlh3-32 allele. Interestingly, the mutations that they identified also conferred a greater defect in crossing over than in MMR, consistent with our analysis of mlh3-32. In addition, they found that mutations within and near the mlh3-32 allele compromised DNA binding activity of Mlh1-Mlh3, suggesting that DNA binding within the linker region may be important for meiotic functions, though we did not detect any apparent defect in the endonuclease activity of Mlh1-mlh3-32.
Alanine-scan mutagenesis of Mlh1 [54] and Mlh3 have provided us with additional information regarding the unstructured linkers in Mlh proteins. Previously we used protein structure prediction and molecular analyses to map the Mlh1 unstructured linker to amino acids 336 to 480 [80]; a similar analysis mapped the Mlh3 unstructured linker to amino acids 373 to 490 [16]. As in the analysis of the Mlh3 random coil, few mutations were identified in the Mlh1 unstructured linker that conferred defects in MMR and crossing over. For example, in Mlh1, no mutations were identified between amino acids 427 and 490 that conferred mutator phenotypes. However, similar to results seen for Mlh3 (S1 Fig; Table 1), mutations were identified just before the unstructured linker in Mlh1 (253-312) that conferred strong mutator phenotypes [54]. Curiously, the corresponding region in MutL contains residues that have been linked through crystallographic analysis to DNA binding [82], suggesting that the organization of the DNA binding and unstructured linker domains in the MLH proteins is conserved. Finally, in both Mlh1 and Mlh3, a localized set of mutations within the center of the unstructured linker (390-403 in Mlh1, 414-416 in Mlh3) affect function, suggesting that this specific region is likely to play an important function beyond serving as a random coil.

Closing thoughts
Mlh1-Mlh3 appears to be acting in CO resolution through a novel mechanism distinct from known structure-selective endonucleases (75). Mlh1-Mlh3 does not share conservation with the known endonuclease superfamilies (XPF, URI-YIG, Rad2/XPG), and does not appear capable of resolving model HJ substrates [50,74]. As mentioned previously, dHJ resolution by Mlh1-Mlh3 results in only CO products whereas the interference-independent CO pathway, which is dependent on Mus81-Mms4, resolves dHJs into a mixture of CO and NCO products [8]. Thus, Mlh1-Mlh3's distinct activity suggests that its nicking is positioned by pro-CO factors such as Msh4-Msh5, Zip3, the STR complex, and Exo1. Such factors are likely to orient Mlh1-Mlh3 to promote asymmetric cleavage of dHJs in a highly regulated and coordinated manner. Thus our work provides further motivation to examine Mlh1-Mlh3 activity on recombination substrates in the presence of pro-CO factors.
Polymorphisms in human MLH3 genes have been associated with male and female infertility [83][84][85], and errors in meiotic chromosome segregation are considered a leading cause of spontaneous miscarriages and birth defects [13]. It is interesting to note that the mlh3-23 mutation, which only weakly affected crossing over, conferred an alteration in meiotic recombination outcomes that was similar to that seen in mlh3 mutants that conferred more severe defects (Fig 5). This observation suggests that some polymorphisms in meiotic recombination genes could have more severe defects in human fertility than expected.
Site-directed mutagenesis of MLH3 60 mlh3 alleles were constructed, resulting in the mutagenesis of 139 amino acids in the 715 amino acid Mlh3 polypeptide (S1 Table). The single-step integration vector (pEAI254), containing the SK1 MLH3 gene with a KANMX4 selectable marker inserted 40 bp downstream of the stop codon [5], was used as a template to create plasmids (S3 Table) bearing the mlh3 mutant alleles via QuickChange site directed mutagenesis (Stratagene, La Jolla, CA). mlh3-60, in which the last 11 residues of Pms1 (DWSSFSKDYEI) were inserted before the MLH3 stop codon, was also made by QuickChange. Mutations were confirmed by sequencing the entire open reading frame (Sanger method), as well as 70 bp upstream and 150 bp downstream. Primer sequences used to make and sequence these variants are available upon request.

Mlh3 homology model
The amino acid sequence of S. cerevisiae Mlh3 (YJM789) was used to construct a homology model from HHpred (http://toolkit.tuebingen.mpg.de/hhpred) and Modeller software. PyMOL was used for imaging.

Construction of strains to measure meiotic crossing over and MMR
The SK1 strain EAY3255 (S1 Table) was constructed to allow for the simultaneous analysis of mlh3 MMR and meiotic crossing over phenotypes. It carries a spore autonomous fluorescent protein marker (RFP) on chromosome VIII to monitor chromosome behavior (crossing over and nondisjunction; [56]) as well as the lys2::InsE-A 14 cassette to measure reversion to Lys + [55]. pEAI254 and mutant derivatives described above and in S3 Table were digested with BamHI and SalI and introduced into EAY3255 by gene replacement using the lithium acetate transformation method as described in Gietz et al. [89]. At least two independent transformants for each genotype (verified by sequencing) were made resulting in a total of 120 haploid strains bearing the mlh3 variants described in this study (S1 Table). These haploid strains were used to measure the effect of mlh3 mutations on reversion rate and were mated to EAY3486, an mlh3Δ strain containing the CFP marker, resulting in diploid strains suitable for analysis of crossing over (S2 Table). Diploids were selected on media lacking the appropriate nutrients and maintained as stable strains. Meiosis was induced upon growing the diploid strains on sporulation media as described in Argueso et al. [7]. Wild-type strains carrying the fluorescent protein markers used to make the above test strains were a gift from the Keeney lab.

Lys + reversion assays
The haploid strains described above were analyzed for reversion to Lys + as described in Tran et al. [55]. At least 10 independent cultures were analyzed for each mutant allele alongside wild-type or mlh3Δ controls. Analyses were performed for two independent transformants per allele. Reversion rates were measured as described [90,91], and each median rate was normalized to the wild-type median rate (1X) to calculate fold increase. Alleles were classified into a wild-type, intermediate, or null phenotype based on the 95% confidence intervals which were determined as described [92].
Spore autonomous fluorescent protein expression to measure percent tetratype Diploids in the EAY3255/EAY3486 background described above (S2 Table) were sporulated on media described in Argueso et al. [7]. Spores were treated with 0.5% NP40 and briefly sonicated before analysis using the Zeiss AxioImager.M2 [56]. At least 250 tetrads for each mlh3 allele were counted to determine the % tetratype. Two independent transformants were measured per allele. A statistically significant difference (p<0.01) from wild-type and mlh3Δ controls based on χ 2 analysis was used to classify each allele as exhibiting a wild-type, intermediate, or null phenotype.

Meiotic time courses
Meiotic time course were performed as described in Sonntag Brown et al. [93] for the diploid strains EAY3252/EAY3486 (MLH3), EAY3255/EAY3486 (mlh3Δ), EAY3534-35/EAY3486 (mlh3-23), EAY3552-53/EAY3486 (mlh3-32), and EAY3819-20/EAY3486 (mlh3-D523N; S2 Table). Strains in single time courses were grown in the same batch of media under identical conditions. Aliquots of cells at specific time points were stained with DAPI to determine the percentage of cell that completed the first meiotic division (cells in which 2, 3, or 4 nuclei were observed by DAPI staining, presented as MI+MII). Cells were visualized using a Zeiss Axio Imager M2 microscope equipped with a DAPI filter. At least 150 cells were counted for each time point. Two independent transformants were analyzed per allele.
Endonuclease assay on supercoiled plasmid DNA and ATPase assay Mlh1-Mlh3 nicking activity was assayed on supercoiled pBR322 or pUC18 (Thermo Scientific). DNA (2.2 nM) was incubated in 20 μl reactions containing indicated amounts of Mlh1-Mlh3 and Msh2-Msh3 [97] in 20 mM HEPES-KOH pH 7.5, 20 mM KCl, 0.2 mg/ml BSA, 1% glycerol, and 1 mM MgCl 2 for 1 h at 37˚C. Reactions were quenched by incubation for 20 min at 37˚C with 0.1% SDS, 14 mM EDTA, and 0.1 mg/ml proteinase K (New England Biolabs) (final concentrations). Samples were resolved by 1% agarose gel with 0.1 μg/ml ethidium bromide run in 1X TAE buffer for 50 min at 95 V. All quantifications were performed using GelEval (FrogDance Software, v1.37). The amount of nicked product was quantified as a fraction of the total starting substrate in independent experiments. bkg indicates that amount of nicked product was not above background levels established by negative controls. ATPase assays were performed as described [16].

Genetic map distance analysis
Diploids from the SK1 congenic strain background EAY1112/1108 [5][6][7] were used for genetic map distance analyses. EAY1112/2413 (MLH3/mlh3Δ::NATMX) and EAY1848/2413 (mlh3Δ:: KANMX/mlh3Δ::NATMX) were used as wild-type and null controls respectively (S2 Table). mlh3 alleles of interest were integrated into EAY3712 (same as EAY1112 but mlh3Δ::URA3) using standard techniques [89]. The resulting haploid strains (EAY3713-EAY3724) were mated to EAY2413 (mlh3Δ::NATMX) giving rise to diploids carrying markers suitable for genetic distance measurements (S2 Table). Two independent transformants were analyzed per allele. Diploids were selected on media lacking the appropriate nutrients and maintained as stable strains. Diploids were sporulated as described [7]. Tetrads were dissected on synthetic complete media and germinated at 30˚C after an incubation of 2-3 days. Spore clones were then replica-plated on various selective media to be scored after 1 day of incubation at 30˚C. Chromosome behavior was analyzed using the recombination analysis software RANA to measure genetic map distances and spore viabilities [7]. Genetic map distances ± SE were calculated using the formula of Perkins [98] through the Stahl Laboratory Online Tools portal (http://molbio.uoregon.edu/~fstahl/).
Genome wide mapping of meiotic recombination events in the S288c/ YJM789 hybrid Genomic DNA was extracted from spore colonies of four viable spore tetrads of the mlh3 point mutants, isogenic wild type, and isogenic mlh3Δ mutants as described previously [64]. Whole genome sequencing on the Illumina Hi-Seq 2500 platform was performed at Fasteris, Switzerland. Raw sequence reads were processed and SNPs genotyped as described in Krishnaprasad et al. [64] and Chakraborty et al. [70], with 4:0 marker segregations shared between tetrads removed from the analysis. Analysis of genome wide recombination events was performed as described in Oke et al. [34] (S5 to S9 Tables). Custom R scripts were used to generate the segregation file (input file for the CrossOver program (v6.3) of the ReCombine [99] program suite (v2.1)). The output of the CrossOver program was used as input for groupEvents (v15). Median crossover associated tract lengths were calculated from E2 events with non-zero tracts [34]. For all count data, the t-test was used to examine the difference in the mean between samples. For tract lengths, the Wilcoxon Rank-Sum test was used to test the difference in the median between samples.
The segregation files, CrossOver program output files, groupEvents output files and the custom R scripts are available online at the Dryad digital repository (http://dx.doi.org/10.5061/ dryad.bb702). Sequence data are available from the National Centre for Biotechnology Information Sequence Read Archive (Accession numbers SRP096621 and SRP110341).
Overexpression of Sgs1 and sgs1-hd SGS1 (native promoter, ORF, and termination sequence) was PCR amplified from SK1 genomic DNA obtained from NKY730 (MATa/alpha, ura3Δ::hisG/ura3Δ::hisG, leu2::hisG/leu2:: hisG, lys2/lys2) and cloned into the high copy vector pRS426 (2μ, amp R , URA3) to make pEAM266 (S3 Table). The correct DNA sequence was confirmed by Sanger sequencing the entire insert. sgs1-K706A the helicase defective variant of SGS1 (sgs1-hd) was made by sitedirected mutagenesis of pEAM266 (New England BioLabs Q5 kit). The correct sgs1-hd, 2μ, and URA3 sequences were confirmed by Sanger sequencing. Primer sequences used to make these constructs are available upon request. The high copy vector (with or without the SGS1 or sgs1-hd insert) was transformed into stable diploids of EAY3534-35/EAY3468 (mlh3-23/ mlh3Δ), EAY3552-53/EAY3486 (mlh3-32/mlh3Δ), EAY3819-20/EAY3486 (mlh3-D523N/ mlh3Δ), EAY3252/EAY3486 (MLH3/mlh3Δ), and EAY3255/EAY3486 (mlh3Δ/mlh3Δ) (S2 Table). Meiosis was induced as described in Argueso et al. [7] and vector selection was maintained by growing the diploid strains on minimal media lacking uracil prior to sporulation. In addition, sporulation media lacked uracil. For spore viability measurements, tetrads were dissected on synthetic complete media and germinated at 30˚C after an incubation of 2-3 days. At least two independent transformants were analyzed per high copy vector. Differences in spore viability were assessed for significance using the χ 2 test.  (Fig 3B). Mismatch repair was measured using the lys2-A 14 reversion assay [55] and crossing over was measured using the assay [56] depicted in Fig 3A. Bars represent the median reversion rates and percent tetratype normalized to MLH3 (1X). The vertical bars indicate the approximate position of the mlh3 mutations analyzed in this study with the height of each bar corresponding to the phenotype relative to MLH3 (1X). Red indicates a null phenotype, blue indicates wild-type (WT), and green indicates intermediate. For mismatch repair (top), bars represent reversion rates of at least 10 independently tested cultures from two independently constructed strains presented here normalized to MLH3 median rate 1X = 1.43x10 -6 (n = 140). For crossing over (bottom), bars represent percent tetratype of at least 250 tetrads from two independently constructed strains presented here normalized to MLH3 percent tetratype 1X = 36.7% (n = 226). Blue and red dotted lines represent MLH3 and mlh3Δ respectively. Black star indicates separation of function mutants (Table 1 and S2 File). (TIFF) S2 Fig. Crossover and noncrossover distribution on chromosomes for wild-type, mlh3-23,  mlh3-32, mlh3-D523N and mlh3Δ. A. and C. Scatter plot of average crossover (CO, E2+E3) and noncrossover (NCO, E1) counts per chromosome against chromosome size [34]. B. and D. Bar plot of average crossover and noncrossover counts per chromosome. The asterisk symbol ( Ã ) marks chromosomes that have significant difference (two tailed t-test for difference in mean; P<0.05) in crossover / noncrossover counts compared to wild-type. Chromosomes are arranged by size from left to right. Error bars are mean ± standard error (S2 File).  wild-type, mlh3-23, mlh3-32, mlh3-D523N, and mlh3Δ. A. Distribution of gene conversion tract lengths associated with simple NCO (E1) and CO (E2) events (S2 File). B. Distribution of gene conversion tract lengths associated with simple NCO (E1) and discontinuous NCO (E4) events. Ã p 0.05 comparing mlh3Δ to mlh3-32. In A and B the minimum, first quantile, median, third quantile, and maximum count are indicated in the box plot. were incubated with 2.2 nM supercoiled pBR322 DNA, analyzed in agarose gel electrophoresis, and endonuclease activity was quantified (average of 6 independent experiments presented +/-SD) as described in the Methods (S2 File). C. ATPase assays were performed as described in Rogacheva et al. [16], but contained the indicated amounts of Mlh1-Mlh3 and Mlh1-mlh3-6 incubated with 100 μM 32 P-γ-ATP. Reactions were performed in duplicate for two separate purifications of each, and the average values, +/-SD, are presented (S2 File). (TIFF) S1 Table. Yeast strains used in this study. EAY3252, EAY3255 and derivatives, and EAY3486 are SK1 strains that contain spore autonomous fluorescent markers described in Thacker et al. [56]. EAY1112 and EAY2413 contain chromosome XV markers described in Argueso et al. [7]. KTY strains were used for whole genome recombination mapping as described in Krishnaprasad et al. [64] and Oke et al. [34]. (PDF) S2 Table. Diploid strains used to measure % tetratype, spore viability, meiotic progression, genetic map distances and for whole genome recombination mapping. The indicated haploid strains were mated to form the diploids with the relevant genotype shown. (PDF) S3 Table. Plasmids used in this study. A. All MLH3 mutagenesis plasmids are derived from pEAI254, a 7.8 kb MLH3 SK1 ::KANMX integrating vector. pEAI254 was mutagenized by QuickChange to create the alleles listed. The DNA sequence of the entire ORF, and 70 bp upstream and 150 bp downstream, were confirmed by DNA sequencing using primers EAO318, EAO319, EAO1778 and EAO321. B. For the two-hybrid analysis, pEAM105 contains the entire MLH1 gene derived from the SK1 strain inserted immediately after the lexA binding domain in pBTM116. All GAL4 activating domain-mlh3 plasmids are derived from pEAM234, which contains DNA sequence encoding SK1 MLH3 amino acids 481 to 715 inserted immediately after the GAL4 activating domain in pGAD424. C. SGS1 was cloned into pRS426 to make pEAM266 as described in the Methods. Site directed mutagenesis of pEAM266 was performed to make pEAM270 carrying the helicase defective sgs1 allele as described in the Methods. (PDF) S4 Table. Genetic map distances for mlh3 separation of function mutants on chromosome XV from single spores and tetrads. All mlh3 mutants are isogenic derivatives of EAY1112/ EAY2413 (S2 Table; Methods). For single spores, recombination frequencies (recombinant spores/total spores) were multiplied by 100 to yield genetic map distances (cM). For tetrads, genetic distance in centimorgans (cM) was calculated using the RANA software without considering aberrant segregants (Argueso et al. [7]). The Stahl Laboratory Online Tools website (http://molbio.uoregon.edu/~fstahl/) was used to calculate standard error (SE) around the genetic distance for tetrads. n; number of single spores, N; four spore viable tetrads analyzed; Par, parental single spores; Rec, recombinant single spores. (PDF) S5  mlh3-23, mlh3-32, mlh3-D523N and mlh3Δ lengths in mlh3Δ, wild type, mlh3-23, mlh3-32 and  mlh3-D523N mutants. E1, E2, E4 are classified by groupEvents [34]. P-values show statistical significance of difference in median CO and NCO gene conversion tract lengths of wild-type and mlh3 alleles compared to mlh3Δ using the Wilcoxon rank sum test (Methods; S2 File). (PDF) S9 Table. Summary of all event categories, as defined in Oke et al. [34], in wild-type, mlh3-23, mlh3-32, mlh3-D523N and mlh3Δ. (PDF)