mlh3 separation of function and endonuclease defective mutants display an unexpected effect on meiotic recombination outcomes

Mlh1-Mlh3 is an endonuclease hypothesized to act in meiosis to resolve double Holliday junctions into crossovers. It also plays a minor role in eukaryotic DNA mismatch repair (MMR). To understand how Mlh1-Mlh3 functions in both meiosis and MMR, we analyzed in baker’s yeast 60 new mlh3 alleles. Five alleles specifically disrupted MMR, whereas one (mlh3-32) specifically disrupted meiotic crossing over. Mlh1-mlh3 representatives for each separation of function class were purified and characterized. Both Mlh1-mlh3-32 (MMR+, crossover-) and Mlh1-mlh3-45 (MMR-, crossover+) displayed wild-type endonuclease activities in vitro. Msh2-Msh3, an MSH complex that acts with Mlh1-Mlh3 in MMR, stimulated the endonuclease activity of Mlh1-mlh3-32 but not Mlh1-mlh3-45, suggesting that Mlh1-mlh3-45 is defective in MSH interactions. Whole genome recombination maps were constructed for two mlh3 mutants with opposite separation of function phenotypes, and an endonuclease defective mutant. Unexpectedly, all three showed increases in the number of non-crossover events that were not observed in mlh3Δ. Our observations provide a structure-function map for Mlh3 that reveals the importance of protein-protein interactions in regulating Mlh1-Mlh3’s enzymatic activity. They also illustrate how defective meiotic components can alter the fate of meiotic recombination intermediates, providing new insights for how meiotic recombination pathways are regulated. Author Summary During meiosis, diploid germ cells that become eggs or sperm undergo a single round of DNA replication followed by two consecutive chromosomal divisions. The segregation of chromosomes at the first meiotic division is dependent in most organisms on at least one genetic exchange, or crossover event, between chromosome homologs. Homologs that do not receive a crossover frequently undergo non-disjunction at the first meiotic division, yielding gametes that lack chromosomes or contain additional copies. Such events have been linked to human disease and infertility. Recent studies suggest that the Mlh1-Mlh3 complex is an endonuclease that resolves recombination intermediates into crossovers. Interestingly, this complex also acts as a matchmaker in DNA mismatch repair (MMR) to remove DNA replication errors. How does one complex act in two different processes? We investigated this question by performing a mutational analysis of the baker’s yeast Mlh3 protein. Five mutations were identified that disrupted MMR but not crossing over, and one mutation disrupted crossing over while maintaining MMR. Using a combination of biochemical and genetic analyses to further characterize these mutants we illustrate the importance of protein-protein interactions for Mlh1-Mlh3’s activity. Importantly, we illustrate how defective meiotic components can alter the outcome of meiotic recombination events. They also provide new insights in our understanding of the basis of infertility syndromes.


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 mlh3Δ-like 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 and whose activity is enhanced by Msh2-Msh3 in vitro [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 non-crossovers (NCOs; [18][19][20][21][22][23]). In this pathway a DSB, which primarily forms on one chromatid of a homologous pair, is resected Al-Sweel et al. 5 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 create a single-end invasion intermediate (SEI). A second invasion event initiating from the SEI, known as second-end capture, can re-anneal and ligate to the other side of the DSB resulting in the formation of a double Holliday junction (dHJ). The dHJ can be acted upon by Holliday junction (HJ) resolvases to form CO and NCO products. In baker's yeast the majority of COs are formed through an interference-dependent CO pathway (class I COs) in which the vast majority of dHJs are resolved to form evenly spaced COs in steps requiring the ZMM proteins Zip1-4, Mer3, and Msh4-Msh5 as well as the Sgs1-Top3-Rmi1 (STR) helicase/topoisomerase complex, Mlh1-Mlh3, and Exo1 [8,[24][25][26][27][28][29][30][31]. These steps are biased to resolve the two junctions present in the dHJ in opposing orientations such that the resulting product is primarily a CO. A second interference-independent pathway was identified that accounts for a small (~10%) number of CO events (class II COs). In this pathway, which does not involve the ZMM proteins, the two junctions are resolved independently by the Mms4-Mus81 endonuclease, leading to a mixture of CO and NCO products [7,8,32,33]. Events that escape STR disassembly form unregulated joint molecules that are resolved by the structure selective nucleases (SSNs) as noncrossovers or class II crossovers.
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 showed that mlh1 and mlh3 mutants display approximately two-fold reductions in crossing over Al-Sweel et al. 6 [7,34,35]. 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 [5,7,8,36], with Yen1 and Slx1-Slx4 acting in cryptic or backup roles. When all four factors were removed, crossing over was reduced to nearly background levels; however in an mms4 slx4 yen1 triple mutant, in which Mlh1-Mlh3 is maintained, relatively high CO levels (~70% of wildtype 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 ,7, 8, 16, 29, 30, 31, 35, 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 CO-associated gene conversion patterns in yeast tetrads, suggested that Zip3, a SUMO E3 ligase, is required for biased cleavage [47]. 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 [47]. 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 [43]. Furthermore, cytological observations in mouse have shown Al-Sweel et al. 7 that MSH4-MSH5 foci appear prior to MLH1-MLH3 [37,44,48,49]. Consistent with these observations, MLH1 and MLH3 foci formation requires MSH4-MSH5 [49].
Additional support for the above model was obtained from analysis of the STR complex, which has been identified as a pro-CO factor in the ZMM pathway [8,30,31,46,50]. 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) [30,50]. Initially, the Sgs1 helicase was characterized as anti-CO because it facilitates unwinding of DSB repair intermediates. 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. 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 [51].
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 Al-Sweel et al. 8 analysis of these mutants revealed unexpected ways in which defective meiotic components can alter the fate of meiotic recombination intermediates.

Results
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 C-terminal Mlh1 interaction domain [52]. 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 [53,54], and has provided a comprehensive view of the functional organization of MLH1 [55]. As shown below, we identified mutations that caused defects in MMR but not crossing over, likely through disrupted interactions with Mlh1 and other MMR and meiotic CO factors. acid positions of charged-to-alanine substitutions presented in red on the primary sequence of Saccharomyces cerevisiae Mlh3. Each cluster of underlined residues represents one allele corresponding to the vertical bars in panel A. mlh3- 39, -40, -57, -58, and -59 are colored in red as in panel A. mlh3-60 represents the last 11 residues of Pms1 which constitute patch II of the heterodimerization interface of Mlh1-Pms1 [52]. C. Metal binding site of Pms1 (left panel) from [52] comprised of the five highlighted residues (H703, E707, C817, C848, and H850) were found to be highly conserved in Mlh3 (right panel) based on sequence alignment and structural modeling (H525, E529, C670, C701, and H703) and were targeted in the mutagenesis described in this study (alleles represented in red in A and B).

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; [56]). In this assay the median reversion rate of mlh3Δ is six-fold higher than wild-type (Fig 3B; Table 1; [5,6]). To measure meiotic crossing over we crossed mutant mlh3 strains to mlh3Δ strains 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 ( [57]; 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). 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 [57].  [57]. Percent tetratype at this interval in wild-type meiosis is 36.7%. B. Mismatch repair (top) and crossing over (bottom) phenotype of MLH3 (blue) vs mlh3Δ (red). Mismatch repair was measured using the lys2-A 14 reversion assay [56] and crossing over was measured using the assay depicted in   [58]. This work showed that conformational changes license MutS-MutL interaction and are essential for MMR.
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,  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  interactions between the C-terminal domain of Mlh1 and the endonuclease active site of Pms1 [52]. 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 [52]. 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 [52].
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 [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.
For this reason we tested whether the mutant complex displayed a defect in ATPase activity.
As shown in S1C Fig, Mlh1-Mlh3 and Mlh1-mlh3-6 displayed similar ATPase activities.   Table). Thus, these mutants were confirmed as separation of function alleles and are candidates for in-depth characterization and high-resolution recombination mapping.  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  As presented in Fig 1, the STR complex can act as both a negative and positive regulator of CO formation in meiotic prophase [8,30,31,46,50]. 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 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 [66].
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 structure-specific nucleases (SSNs), Mus81-Mms4 and Yen1.
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 (S2 Fig: 76% in mlh3Δ+2µ vs. 57% in mlh3Δ+SGS1-2µ) and conferred a more endonuclease activity [5,16]. Nonetheless, this mutant did not behave like mlh3Δ in response to Sgs1 overexpression. This finding encouraged us to explore roles for Mlh1-Mlh3 that appear independent of its enzymatic activity.
High-resolution recombination maps illustrate unexpected effects of mlh3 hypomorphs and the mlh3-D523N allele on resolving meiotic recombination intermediates.
We characterized two alleles with opposite separation of function phenotypes, mlh3-23 (MMR -, CO + ) and mlh3-32 (MMR + , CO -), by mapping recombination events genome-wide using the S288c/YJM789 hybrid [67]. We also analyzed the mlh3-D523N mutation described above [5,16]. The Mlh1 protein sequence has two amino acid differences between SK1 and YJM789 strains and three amino acid differences between SK1 and S288c strains. The SK1 Mlh3 protein has 11 amino acid differences with respect to S288c Mlh3 and seven with respect to YJM789 Mlh3. Therefore we analyzed the SK1 mlh3 mutations in the presence of SK1 MLH1 in the S288c/YJM789 hybrid to avoid genetic incompatibilities between Mlh1 and Mlh3.
As described below, high-resolution recombination mapping analysis indicated that mlh3 point mutants displayed genome-wide increases in NCO events that were not observed in wildtype or mlh3Δ. Consistent with classical tetrad analysis, mlh3-32 and mlh3-D523N displayed CO values similar to mlh3Δ. Importantly, median gene conversion tract lengths associated with COs and NCOs were longer in mlh3Δ compared to wild-type, mlh3-23, mlh3-32 and mlh3-D523N. There are multiple possible explanations for these phenotypes, but one possibility is that longer gene conversion tract lengths arise in mlh3Δ if the two Holliday junctions present in dHJ intermediates are separated by a longer distance as the result of entry into a pathway that uses a different processing and resolution mechanism (see Discussion). We did not obtain any evidence that the number of DSBs increased in mlh3 point mutants; such an increase would have provided a simple explanation for why an increase in NCO events was observed.

ii. Genome-wide increase in non-crossovers in
One explanation for the increase in NCO events seen in mlh3-23, mlh3-32 and mlh3-D523N mutants in the genome wide recombination analysis is that these mutants experience meiotic progression delays that result 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 [27,30,50,57,72]. To test this possibility we 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 S5 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), suggesting that the increase in NCO events in mlh3-32, mlh3-23, and mlh3-D523N cannot simply be explained due to a meiotic progression delay. Furthermore, as shown in Fig 7C,  iii. Non-exchange chromosome frequencies in mlh3-32 and mlh3-D523N are similar to mlh3∆. At least one non-exchange chromosome was observed in 43% and 40% of four viable spore tetrads in mlh3-32 and mlh3-D523N, respectively. This was comparable to that seen in mlh3∆ (47%; Fig 8A).  tetrads with zero, one, or more than one non-exchange chromosomes are shown.

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 wellcharacterized structure-selective endonucleases (e.g. Mus81-Mms4, Slx1-Slx4, and Yen1) in terms of homology and intrinsic behavior in vitro (reviewed in [51] pro-CO factors [16,17,51]. The identities of these factors are for the most part known, though it is not understood how they contribute to Mlh1-Mlh3's ability to nick DNA in the directed manner required to generate COs. Our analysis of two separation-of-function alleles, mlh3-32 (MMR + , CO -) and mlh3-45 (MMR -, CO + ), suggests that protein-protein interactions are critical for directing Mlh1-Mlh3 endonuclease activity (Table 2). Mlh1-Mlh3 has been shown genetically to act downstream of Msh4-Msh5 [40,41,44,48]; this order of events is analogous to steps in DNA MMR where MLH acts following MSH recognition [14,73]. 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 [51]

Does Mlh1-Mlh3 have a regulatory role in meiotic pathway choice?
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 [8,30,50]. In these models DSB intermediates that escape capture by ZMM proteins are resolved into class II COs or NCOs by structure selective nucleases. NCOs can also arise from the action of the STR complex through synthesis dependent strand annealing (SDSA; [30,50]; Fig 1).
We observed a genome-wide increase in NCOs in mlh3-23, mlh3-32, and mlh3-D523N that was not seen in mlh3Δ mutants or wild-type. In addition, an increase in tract lengths for gene conversions associated with COs and NCOs was observed in mlh3Δ compared to wildtype, mlh3-23, mlh3-32 and mlh3-D523N (S4 Fig; S8 Table).
One explanation for the above observations is that in the absence of Mlh1-Mlh3, DSB intermediates are readily available for processing by class II pathway SSNs. In this scenario, Mlh1-Mlh3, in concert with the ZMM proteins, protect recombination intermediates from Sgs1, Al-Sweel et al. 29 and thus limit heteroduplex extension. Consistent with this, Zip3 has been shown to limit gene conversion tract lengths by limiting heteroduplex extension driven by Sgs1 [47]. Also, mlh3Δ, zip3Δ, and msh4Δ show increased CO gene conversion tract lengths that were not seen in the mlh3 separation of function mutants, suggesting a possible commitment to resolution involving the ZMM proteins and endonuclease-independent functions of Mlh1-Mlh3 [23,47,64,67].
The data presented above can be explained by Mlh1-Mlh3 having an early structural role that is active in mlh3-23, mlh3-32, and mlh3-D523N mutants, resulting in an increase in NCO events due to the loss of biased resolution of dHJs into COs that is a hallmark of the ZMM pathway. Defects in subsequent steps could arise from altered interactions between mutant Mlh1-Mlh3 complexes and pro-CO ZMM factors, permitting structure specific nucleases to resolve dHJs into COs and NCOs (Fig 1). Recent studies suggested that some meiotic factors have earlier roles than first hypothesized; for example, Thacker et al. [75] identified a feedback pathway for the ZMM proteins Zip1, Zip3 and Msh5 that regulates DSB formation in meiosis. At the time this was considered surprising because the ZMM proteins were thought to act exclusively after DSB formation. In support of an early structural role for Mlh1-Mlh3 we found that Sgs1 overexpression decreased the spore viability of mlh3Δ strains but not MLH3 or mlh3-D523N strains; we also found that Sgs1 overexpression modestly increased the spore viability of mlh3-32 mutants (S2 Fig). Interestingly, the spore viability pattern seen in mlh3Δ strains overexpressing Sgs1 is consistent with a Meiosis I segregation defect, which might be expected if COs which do not display interference are produced through non-ZMM pathways.
Finally, recent work from Duroc et al. [76] provided evidence that another MLH complex, Mlh1-Mlh2, acts to limit the extent of meiotic gene conversion. In their studies 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. We observed more subtle increases in gene conversion tract It is equally plausible and perhaps simpler that the mutant Mlh1-mlh3 complexes analyzed here display a pathogenic behavior that prevents alternative dHJ resolution activities following ZMM entry. dHJ resolution by Mlh1-Mlh3 is thought to occur when the synaptonemal complex breaks down [9,77]. If Mlh1-Mlh3 is absent at this time one could imagine that dHJs become susceptible to the actions of the STR complex, resulting in the unwinding and the convergent migration of the two HJs until a single pair of crossing strands in a hemicatenane can be removed by the topoisomerase [30,31]. However, if defective Mlh1-mlh3 complexes remain bound to dHJs and prevent their dissolution by STR, SSNs or other resolvases could resolve dHJs into class II events (Fig 1). This can also explain the longer gene conversion tract lengths associated with COs and NCOs observed in mlh3Δ compared to wild-type and the mlh3 mutants. Physical assays (e.g. two-dimensional electrophoresis) that temporally measure recombination intermediates in meiosis will likely be useful to test this idea (e.g. [8]).
Lastly, it is possible is that delays in meiotic progression in mlh3 mutants 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 (S5 Fig). Also, an analysis of the density of CO and NCO events in our genome-wide recombination events suggest that DSB densities were not altered in mlh3-32, mlh3-23, and mlh3-D523N mutants. This is further supported by a decrease in CO:NCO ratios from 2.0 in the wild-type background to 1.3, 1.1, and 1.0 in mlh3-23, mlh3-32, and mlh3-D523N respectively, indicating that the total number of events does not change significantly (from Table 3). These observations suggest that the additional NCO events seen in the mlh3 mutants did not result from increased DSB formation.

Mlh3's linker arm is critical for its meiotic function.
Al-Sweel et al. 31 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 [55]. Previous work showed that the Mlh1-Pms1 heterodimer undergoes large global conformational changes in an ATP binding and hydrolysis cycle [78]. 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 [78], 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 [79]. 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 (Fig 3C), 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 [80] 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 DNA mismatch repair, 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 mutageneses of Mlh1 [55] 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 [79]; 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 However, similar to results seen for Mlh3 (Fig 3C), mutations were identified just before the unstructured linker in Mlh1 (253-312) that conferred strong mutator phenotypes [55]. Curiously, the corresponding region in MutL contains residues that have been linked through crystallographic analysis to DNA binding [81], 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. 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 [51]. 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 [82][83][84], 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 Al-Sweel et al. 33 recombination outcomes that was similar that seen in mlh3 mutants that conferred more severe defects (Fig 7) . 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). 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 non-disjunction; [57]) as well as the lys2::InsE-A 14 cassette to measure reversion to Lys + [56]. 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. [88]. 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.
The haploid strains described above were analyzed for reversion to Lys + as described in Tran et al. [56]. At least 10 independent cultures were analyzed for each mutant allele alongside wildtype or mlh3Δ controls. Analyses were performed for two independent transformants per allele.
Reversion rates were measured as described [89,90], 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 [91].

Al-Sweel et al. 35
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 [57]. 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. [36]  al. [5]. Expression of the LACZ reporter gene was determined by the ortho-nitrophenyl-β-Dgalactopyranoside (ONPG) assay as described in [93].

Purification of Mlh1-Mlh3 and mutant complexes from baculovirus-infected Sf9 cells.
Mlh1-Mlh3 and Mlh1-mlh3 mutant derivatives were purified from Sf9 cells infected with Bac-to-Bac baculovirus expression system using pFastBacDual constructs [16]. Mutant Mlh1-mlh3 complexes were purified using the same protocol developed to purify wild-type Mlh1-Mlh3. This promoter and the His 10 -mlh3-HA gene is downstream of the pPH promoter. The sequence of the restriction fragments inserted into pEAE348 were confirmed by DNA sequencing (Cornell Biotechnology Resource Center). Msh2-Msh3 was purified as described previously [95].
Endonuclease assay on supercoiled plasmid DNA and ATPase assay.
Mlh1-Mlh3 nicking activity was assayed on supercoiled pBR322 or pUC18 (Thermo Scientific 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].
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 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 Chakraborty et al. [70]. Analysis of recombination events, interference was performed using the CrossOver program (v6.3) in the ReCombine suite of programs (v2.1; [66]). Parameters for the CrossOver program were set as described in Krishnaprasad et al. [64]). Custom R scripts were used to generate the segregation file (input file for the CrossOver program), plots and to perform statistical tests. The raw recombination data files and the custom R scripts are available online at the Dryad digital repository (MLH3/mlh3Δ) and EAY3255/EAY3486 (mlh3Δ/mlh3Δ) backgrounds as controls (S2 Table).
Al-Sweel et al. 39 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. Two independent transformants were analyzed per high copy vector. Differences in spore viability were assessed for significance using the χ 2 test.  Genetic map distance (cM) % spore viability