MutS homologue 4 and MutS homologue 5 Maintain the Obligate Crossover in Wheat Despite Stepwise Gene Loss following Polyploidization1[CC-BY]

Stuart D. Desjardins,a Daisy E. Ogle,a Mohammad A. Ayoub,b Stefan Heckmann,b Ian R. Henderson,c Keith J. Edwards,d and James D. Higginsa,2,3 Department of Genetics and Genome Biology, University of Leicester, Leicester LE1 7RH, United Kingdom Independent Research Group Meiosis, Leibniz Institute of Plant Genetics and Crop Plant Research, Gatersleben, D–06466 Stadt Seeland, Germany Department of Plant Sciences, University of Cambridge, Cambridge CB2 3EA, United Kingdom University of Bristol, Bristol BS8 1TQ, United Kingdom

Meiosis is a specialized cell division required to accurately segregate chromosomes into haploid gametes. During meiosis I, homologous chromosomes pair, synapse, and recombine, thus ensuring formation of the obligate crossover (CO) necessary to tether chromosome pairs for correct alignment on the metaphase I (MI) plate and subsequent segregation at anaphase I. Homologous recombination is initiated by programmed DNA double-strand breaks (DSBs) catalyzed by Spo11 and MTOPVIB (Keeney et al., 1997;Fu et al., 2016;Robert et al., 2016;Vrielynck et al., 2016) that are repaired as either COs by reciprocal exchange of DNA or noncrossovers when DNA is repaired either using the sister chromatid as a template or via the homologue, which can lead to gene conversion. COs are genetically and cytologically (chiasmata) more likely to be spaced apart than by random chance through a phenomenon known as interference (Jones and Franklin, 2006;Berchowitz and Copenhaver, 2010). The mechanism of interference is poorly understood, but in budding yeast (Saccharomyces cerevisiae) it is imposed by the meiosisspecific ZMM proteins (Zip1, Zip2, Zip3, Zip4, Mer3, MSH4, and MSH5) that also ensure formation of the obligate CO (Osman et al., 2011;Pyatnitskaya et al., 2019). Zip1 forms the synaptonemal complex (SC) transverse filaments (Sym et al., 1993). Zip2 (orthologous to SHOC1 in plants) and Zip4 mediate a molecular switch of recombination intermediates to MER3-MSH4/MSH5 complexes from association with the chromosome axis to SC central element components (Dubois et al., 2019). Zip3 (orthologous to HEI10 in plants) is an E3 ligase potentially involved in posttranslational modification of proteins by SUMOylation or ubiquitylation (Agarwal and Roeder, 2000). Mer3 is a DNA helicase with 39 to 59 activity that binds to D-loops in vitro and may unwind or migrate recombination intermediates in vivo to promote CO formation (Duroc et al., 2017).
MSH4 and MSH5 are meiosis-specific MutS homologues of the bacterial mismatch repair proteins that bind and stabilize recombination intermediates (Snowden et al., 2004). In budding yeast, MSH4 and MSH5 form heterodimers that promote the formation of class I interference-sensitive COs, thus ensuring formation of the obligate CO (Ross-Macdonald and Roeder, 1994;Hollingsworth et al., 1995;Pyatnitskaya et al., 2019). In Caenorhabditis elegans, recombination is virtually eliminated in the him14 (MSH4 orthologue) mutant, indicating that all COs are mediated by the class I pathway (Zalevsky et al., 1999). Recent evidence suggests that MutSg (MSH4 and MSH5) protects the nascent recombination intermediates from dissolution by the orthologue of the Bloom's syndrome RECQ helicase, thus promoting resolution into COs (Woglar and Villeneuve, 2018). In mouse (Mus musculus), no chiasmata were identified at diakinesis in msh5 GA mutants, suggesting a similar requirement for MutSg in promoting all COs, as in C. elegans (Milano et al., 2019). In Arabidopsis (Arabidopsis thaliana), MutSg ensures the obligate CO and is required for ;85% of chiasmata (Higgins et al., 2004(Higgins et al., , 2008b. The residual ;15% of chiasmata observed in mutSg mutants are interference insensitive and form via the class II pathway, which is partially dependent on the MUS81 endonuclease (Berchowitz et al., 2007;Higgins et al., 2008a). Based on a bioinformatics interactome analysis, MutSg in Arabidopsis is likely to function similarly to that in C. elegans in protecting recombination intermediates from dissolution by RECQ helicases (AbdelGawwad et al., 2019). In rice (Oryza sativa), MutSg promotes formation of the obligate CO and is required for 78% to 90% of chiasmata (Luo et al., 2013;Zhang et al., 2014;Wang et al., 2016). In addition, a direct physical association between OsMSH4 and OsMSH5 as well as between OsMSH5 and HEI10 was demonstrated using yeast two-hybrid and pull-down assays (Zhang et al., 2014), as were interactions between OsMSH5 and replication protein A family members (OsRPA1a, OsRPA2b, OsRPA1c, and OsRPA2c; Wang et al., 2016) and OsSHOC1 . In Arabidopsis and rice mutSg mutants, immunolocalization of the synaptonemal complex transverse filament proteins appeared normal, indicating a minimal effect on synapsis during prophase I (Higgins et al., 2004;Luo et al., 2013;Zhang et al., 2014). The only ZMM gene to be studied in temperate cereals by knockdown is barley (Hordeum vulgare) ZYP1. Barley ZYP1 RNAi lines exhibited severely impaired synapsis and did not maintain the obligate chiasma (Barakate et al., 2014). Chiasmata were reduced by 73%, but as the ZYP1 RNAi lines were knockdowns, it could not be determined if this ZMM protein was required for class I only or class I and class II COs.
Allohexaploid wheat (Triticum aestivum) is a domesticated species, originating from two hybridizations (Matsuoka, 2011). Evidence suggests that Triticum urartu (AA) first hybridized with an unknown Aegilops species (BB) ;500,000 years ago to form tetraploid wheat (Triticum dicoccoides; 2n 5 4x 5 24; AABB), followed by domestication into Triticum dicoccum and Triticum turgidum. T. turgidum then hybridized with Aegilops tauschii (DD) ;10,000 years ago to form allohexaploid wheat (2n 5 6x 5 42; AABBDD; Matsuoka, 2011). Pairing homeologous1 (Ph1) has been characterized as the major locus controlling chromosome pairing and homologous COs in polyploid wheat while preventing homeologous COs (Griffiths et al., 2006). It has been fine-mapped to a novel duplicated copy of the ZMM gene ZIP4 and verified using mutants that can be exploited to introgress DNA from distantly related germplasm (Rey et al., 2017). Ph1-ZIP4 on chromosome 5B originated from chromosome 3B, and the ancestral homeologous ZIP4 copies on 3A, 3B, and 3D are still present and expressed (Griffiths et al., 2006;Alabdullah et al., 2019). Therefore, increased ZIP4 gene dosage may bias recombination toward homologous chromosomes over homeologous chromosomes. A bioinformatics analysis revealed that in certain polyploid plant lineages, meiotic recombination genes are the fastest to return to a single copy, and it was proposed that this is a rapid response for adapting meiotic recombination post whole-genome duplication (Blanc and Wolfe, 2004;Lloyd et al., 2014;Sidhu et al., 2017). This is the inverse of ZIP4, which has gained a novel, dominant copy. Recently, it was demonstrated that reducing MSH4 copy number prevents meiotic COs forming between homeologous chromosomes in allohaploid Brassica napus plants generated by microspore culture (Gonzalo et al., 2019). A further bioinformatics analysis revealed that MSH4 systematically returns to a single copy in numerous plant species following independent polyploid events, suggesting an adaptive role during the diploidization of meiosis (Gonzalo et al., 2019). In polyploid species, diploidization refers to the process whereby recombination evolves a bias toward homologues rather than homeologues. A coexpression analysis in wheat revealed mostly balanced homeologous gene expression and a lack of significant meiotic gene loss following polyploidization (Alabdullah et al., 2019). However, as both ZIP4 and MSH4 act in processing recombination intermediates in the class I CO pathway, these opposing data on gene dosage may be complementary rather than antagonistic.
In this study we demonstrate that MutSg ensures formation of the obligate chiasma in tetraploid wheat even with a minimum gene copy number (msh5a 2/1 / msh5b 2/2 ). In addition, the functional MSH5 gene from the B subgenome was rapidly mutated posthybridization in the tetraploid lineage and a substantial section of MSH4D was deleted following the formation of hexaploid wheat. In a genetic background of largely unchanged meiotic recombination genes, functional MSH4 and MSH5 copies have decreased in a stepwise manner posthybridization while maintaining the obligate CO between homologues. This may have occurred by chance due to gene redundancy or may represent an adaptation to allopolyploidy by modulating recombination.
In Ensembl, MSH5B appears to be intact in tetraploid and hexaploid wheat species; however, on closer inspection, it was apparent that TtMSH5B exhibits a 5.4kb deletion compared with Ae. speltoides, a relative of the unknown B-genome progenitor (Supplemental Fig.  S1A). To confirm that the deletion was not just an artifact from an incomplete database sequence, the region in question was amplified from T. turgidum 'Kronos' using flanking primers. The deletion was experimentally confirmed and is predicted to remove exons 2 to 13 (861 bp) from the coding region and to remove 287 amino acid residues from the N terminus of the protein (Supplemental Fig. S1B). This would result in the loss of the entire MutSII domain and a section of the MutSIII domain. MutSII domains bind Holliday junctions and D-loops in vitro (Fukui et al., 2008), so the truncation is likely to lead to a nonfunctional MSH5B protein, indicating that TtMSH5B has degenerated into a pseudogene. The presence of the 5.4-kb deletion in MSH5B appears to have occurred early in the evolution of polyploid wheat, as it was also detected in the ancient, wild tetraploid wheat T. dicoccoides, and has been preserved throughout domestication into modern hexaploid wheat T. aestivum ( Fig. 1; Supplemental Fig. S1).
The wheat MSH4 gene models reflect a common ancestry with Arabidopsis, containing 24 exons and 23 introns. The coding regions are similar in size, with 2.3 kb for AtMSH4 and 2.4 kb for both TaMSH4A/B. Gene lengths for MSH4 are larger in wheat (TaMSH4A, 10.6 kb; and TaMSH4B, 10.5 kb) compared with Arabidopsis Figure 1. Stepwise gene loss of MSH5B and MSH4D during the evolution of polyploid wheat. Functional MSH5B was naturally mutated following allopolyploidization in early tetraploid wheat T. dicoccoides, and the functional MSH4D was subsequently mutated following the formation of hexaploid wheat T. aestivum. Approximate dates are given, and the appearance of the novel ZIP4 duplicate in the Ph1 locus on chromosome 5B is also shown. Functional gene copies are highlighted in blue and pseudogenes in red.
(5.3 kb), due to increased intron sizes. In Ensembl wheat databases, TaMSH4D does not possess a gene model (Alabdullah et al., 2019) due to an 8-kb deletion relative to Ae. tauschii (the D-genome donor species), presumably disrupting a prediction from the gene annotation software (Supplemental Fig. S2, A and B). The 8-kb deletion is predicted to remove exons 16 to 24 (986 bp) from the coding region and to remove 328 amino acids from the C terminus of the protein (Supplemental Fig. S2C). This would result in the partial loss of the MutSIII domain and the entire MutSV domain, including the helix-turn-helix motif (residues 724-742) required to dimerize with MSH5 (Obmolova et al., 2000). The disruption of these highly conserved MutS domains suggests that TaMSH4D has also degenerated into a nonfunctional pseudogene.

Wheat MSH4 and MSH5 Possess Disordered Low-Complexity N Termini
Tetraploid and hexaploid wheat MSH4A primary amino acid sequences only differ from the T. urartu ancestral diploid A genome sequence by two residues (808/810; Supplemental ). Therefore, consensus sequences were created for wheat MSH4 and MSH5 for further analysis. Wheat MSH4 shares 75% amino acid identity with AtMSH4 but is distinguished by a low-complexity Gly-rich disordered N terminus (MEEGAAGGGGGGGGGGVAVA) that is absent in Arabidopsis (Supplemental Fig. 5A). Wheat MSH4 shares 27% amino acid identity with Saccharomyces cerevisiae MSH4 but possesses a considerably shorter N terminus (Supplemental Fig. S5B). Wheat MSH5 shares 67% amino acid identity with AtMSH5 and contains a low-complexity disordered N terminus (MDEDEEEQLEEEEEVAETGID) that is highly acidic and absent in Arabidopsis (Supplemental Fig. 6A). The S. cerevisiae MSH5 protein shares 27% amino acid identity with wheat MSH5 and possesses 84 more residues, of which 35 constitute the N terminus (Supplemental Fig. 6B).
In Ttmsh4 and Ttmsh5 null mutants, the MutSg complex failed to localize to the chromosome axes. Observed background signals formed nonspecific aggregates, particularly around the nucleolus (Figs. 2B and 3B). The cause of these fluorescent bodies in the null mutants is unclear, but the antibodies may be staining MSH4/MSH5 aggregates that are misfolded and failing to load onto the chromosome axes. For MSH5, the antibody targets the C terminus of the protein (Higgins et al., 2008b), which is downstream of the predicted stop codon on the functional A genome. Therefore, the MSH5 antibody is unlikely to interact with mutant truncated forms but could conceivably target the remnant of MSH5B. For MSH4, the antibody is designed to the middle of the protein (Higgins et al., 2004), which targets regions upstream of the stop codons in the A and B subgenomes, so the antibody may bind to truncated forms of the proteins that do not specifically localize to the chromosome axes.
TtMSH4A and TtMSH4B Are Functionally Redundant TtMSH4A and TtMSH4B are functionally redundant in formation of the obligate chiasma in tetraploid durum wheat (T. turgidum). We observed that single mutants for the A (Ttmsh4a-1 and Ttmsh4a-2) and B (Ttmsh4b) subgenomes are fully fertile and indistinguishable from the wild type at meiotic MI ( Fig. 4; Supplemental Table S1). There was no significant difference between single Ttmsh4a/b mutants and the wild type for number of bivalents (P . 0.05) or the frequency of chiasmata (P . 0.05). However, double mutants (Ttmsh4ab) were sterile and did not maintain the obligate chiasma, indicated by the presence of univalents ( Fig. 4; Supplemental Table S1). Only 3.74 6 0.18 (n 5 72) mean bivalents per cell were observed in Ttmsh4ab-1 compared with 13.9 6 0.04 (n 5 59) in the wild type (P , 0.01). Furthermore, Ttmsh4ab-1 MIs predominantly contained rod bivalents (93%) compared with the wild type, which contained mostly rings (82%). The marked increase in univalents and a decrease in ring bivalents coincided with an 85% reduction in the mean chiasmata per cell, from 26 6 0.24 (n 5 59) in the wild type to 4 6 0.21 (n 5 72) in Ttmsh4ab-1. No significant differences were observed between Ttmsh4ab-1 and Ttmsh4ab-2 at MI ( Fig. 4; Supplemental Table S1).
The gene dosage of TtMSH5 appears to be in excess of that required for maintaining wild-type levels of chiasmata, even in the absence of a functional TtMSH5B. The heterozygous mutant Ttmsh5a-1 2/1 has the minimum dosage of TtMSH5 through a null mutation. Ttmsh5a-1 2/1 plants remained fully fertile and meiotic division appeared unperturbed ( Fig. 5; Supplemental Table S1). At MI, there was no significant difference in the mean number of bivalents formed per cell between the wild type and Ttmsh5a-1 2/1 (13.9 6 0.04, n 5 59 versus 13.96 6 0.03, n 5 52; P . 0.05) or in the mean number of total chiasmata per cell in the wild type and Ttmsh5a-1 2/1 (26.22 6 0.24, n 5 59 versus 25.81 6 0.21, n 5 52; P . 0.05). The number of COs between homologous chromosomes, therefore, appears unaffected by the minimum dosage of MSH5.

Early Recombination Events and Synaptonemal Complex Formation Appear Normal in Ttmsh4ab and Ttmsh5a
Even though the obligate chiasma is lost in Ttmsh4ab-1 and Ttmsh5a-1, earlier meiotic stages involving axis formation and synapsis appeared unperturbed (Supplemental Fig. S7). Immunolocalization of the axisassociated protein ASYNAPSIS1 (ASY1; Armstrong et al., 2002) and synaptonemal complex transverse filament protein ZYP1 (Higgins et al., 2005) appeared indistinguishable from the wild type (Sepsi et al., 2017). In the wild type, Ttmsh4ab-1, and Ttmsh5a-1, ASY1 formed a linear signal along the unsynapsed chromosome axes at leptotene but was depleted during zygotene in the synapsed regions (Supplemental Fig. S7). ZYP1 was initially detected in the wild type, Ttmsh4ab-1, and Ttmsh5a-1 during late leptotene at synapsis initiation sites that extended throughout zygotene until a complete linear signal was observed at pachytene (Supplemental Fig.  S7). Furthermore, the number of RAD51 foci that mark early recombination events at leptotene was not significantly different between the wild type (1,408 6 28, n 5 5), Ttmsh4ab-1 (1,346 6 58, n 5 5), and Ttmsh5a-1 (1,395 6 4, n 5 5; P . 0.05; Supplemental Fig. S8), indicating that early recombination events are unaffected in the mutSg mutants. Vegetative growth and floral development also appeared normal in Ttmsh4ab and Ttmsh5a mutants, indicating that they do not cause observable somatic defects.

Class II COs Are Unaffected by Loss of mutSg
The 84% to 85% decrease in chiasmata observed at MI in the Ttmsh5a and Ttmsh4ab mutants is consistent with loss of the class I pathway (Higgins et al., 2004(Higgins et al., , 2008b, but the class II pathway appears unaffected. Class I and class II COs were monitored cytologically by immunofluorescence using antibodies raised against barley HEI10, a class I CO-specific marker (Chelysheva et al., 2012), and TaMUS81, a class II COspecific marker ( Fig. 6; Higgins et al., 2008a). HEI10 localized as a linear signal along the unsynapsed chromosome axes during zygotene in the wild type and the mutSg mutants and then depleted to form discrete foci during synapsis (Fig. 6A). At pachytene, the mean number of HEI10 foci per meiocyte was 28.8 6 0.64 (n 5 21) in the wild type but only 2.7 6 0.26 (n 5 23) in Ttmsh4ab-1 and 3.1 6 0.28 (n 5 23) in Ttmsh5a-1 (P , 0.01). The residual HEI10 foci in the mutSg mutants were generally smaller, fainter, and not associated with the chromosome axes, but a small minority (17.6%, 0.5 6 0.12 foci per cell) appeared similar to those in the wild type (Fig. 6C). However, the mean number of MUS81 foci per meiocyte at pachytene was not significantly different between Ttmsh4ab (3.82 6 0.38, n 5 27) and Ttmsh5a (4 6 0.37, n 5 21) compared with the wild type (4.03 6 0.32, n 5 40; P . 0.05; Fig. 6, B and D).
In the wild type, chiasma frequency was tightly distributed around the mean and deviated significantly from a Poisson-predicted distribution [x 2(21) 5 64.71, n 5 59, P , 0.01]. However, in the Ttmsh4ab (Ttmsh4ab-1 and Ttmsh4ab-2) and Ttmsh5a (Ttmsh5a-1 and Ttmsh5a-2) mutants, the frequency of residual chiasmata was randomly distributed and did not deviate from a Poisson-predicted distribution [Ttmsh4ab, x 2(10) 5 6.71, n 5 127, P . 0.05; Ttmsh5a, x 2(11) 5 1.85, n 5 186, P . 0.05] (Fig. 7, A-C). This indicates that the number of chiasmata per cell is numerically random in Ttmsh4ab and Ttmsh5a, typical of class II COs (Higgins et al., 2004(Higgins et al., , 2008b. Although the number of class II COs was numerically random, the physical chromosomal location of chiasmata was significantly more distal in TtmutSg mutants (80% 6 1.25%, n 5 310) than in the wild type (74.6% 6 1.23%, n 5 59; P , 0.01; Fig. 7D). It follows that class II COs appear to have a slight but significant predisposition to form at the distal ends of chromosomes than do class I COs in tetraploid wheat. However, whereas class II COs are more likely to form in the distal regions, they are not confined there, as 20% of interstitial/proximal chiasmata are observed in TtmutSg mutants.

Hexaploid Wheat Maintains the Obligate Chiasma Despite Loss of TaMSH5B and TaMSH4D
Despite pseudogenization of both TaMSH5B and TaMSH4D, meiosis proceeded normally in hexaploid wheat T. aestivum 'Fielder' and the obligate chiasma was maintained (Supplemental Fig. S9; Supplemental Table S1). The mean number of bivalents per cell was 20.97 6 0.03 (n 5 36) and the total number of chiasmata per cell was 40 6 0.28 (n 5 36), which are predominantly distally distributed (76.5%). This is consistent with a computational analysis of 13 recombinant inbred mapping populations, which gave values of 40.8 to 51.9 COs per line, typically clustered toward the ends of chromosomes (Gardiner et al., 2019). DISCUSSION We have demonstrated that MSH4 and MSH5 (MutSg) are essential for the class I meiotic recombination pathway in tetraploid wheat, accounting for ;85% of meiotic COs as well as ensuring the obligate chiasma. This complements data from Arabidopsis, rice, tomato (Solanum lycopersicum), and B. napus, where MutSg is required for ;85% of COs and the obligate chiasma, indicating that this is most likely the major meiotic recombination pathway in the plant kingdom Figure 5. TtMSH5A is required to form the obligate chiasma. A, TtMSH5 coding regions for the A and B subgenomes with TILLING mutations indicated (red/blue) and predicted MutS domains highlighted in yellow. B, Representative 49,6-diamino-2phenylindole-stained male meiotic MIs from tetraploid wheat T. turgidum 'Kronos'. Bars 5 10 mm. C, Mean number of rings, rods, and univalents per male meiocyte. D, Box plot of chiasmata frequency per male meiocyte. n.s., Not significant. Asterisks indicate significant difference by pairwise Wilcoxon rank sum test (*** P , 0.01). (Higgins et al., 2004(Higgins et al., , 2008bLuo et al., 2013;Anderson et al., 2014;Wang et al., 2016;Gonzalo et al., 2019). The remaining class II COs (;15%) are random in number and are predominantly distally distributed, except for 20% that are observed interstitially. The proportion of class I to class II COs is consistent across Arabidopsis, rice, tomato, and Brassica spp., despite large disparities in chromosome number, genome size, and DSB number. For example, in wild-type tetraploid wheat, ;1,400 RAD51 foci were counted at leptotene as a proxy for DSBs, which mature into 29 HEI10 and four MUS81 foci at pachytene per cell. This closely matches chiasmata numbers in the wild type (26) and mutSg (four) mutants. HEI10 and MUS81 foci account for 2% and 0.3% of RAD51 foci, respectively. In Arabidopsis, ;215 DSBs initiate recombination and nine chiasmata form, of which 1.1 to 1.5 are dependent on the class II pathway (Higgins et al., 2004(Higgins et al., , 2008a(Higgins et al., , 2008bChoi et al., 2013). Therefore, ;2.3% of DSBs mature into chiasmata in tetraploid wheat and ;4% in Arabidopsis, so that the 85%:15% class I:class II proportion remains constant, even though the class II chiasmata fit a Poisson distribution. In addition, results from tomato demonstrated that the two pathways are not independent because of observed interference between class I and class II COs (Anderson et al., 2014). These data raise the intriguing possibility that the class I and class II CO pathways are intimately associated through an unknown mechanism, possibly by patterning of early recombination intermediates.
In allopolyploid wheat, the obligate chiasma is maintained despite the number of functional copies of MSH5 and MSH4 reducing in a stepwise manner posthybridization. The functional MSH5 from the B subgenome was mutated following the formation of allotetraploid (AABB) wheat, and MSH4 from the D subgenome was mutated following the formation of allohexaploid (AABBDD) wheat. The 5.4-kb deletion in Figure 6. HEI10 foci are reduced in number in Ttmsh4 and Ttmsh5, whereas MUS81 is unaffected. A, Representative immunostained pachytenes from tetraploid wheat T. turgidum 'Kronos' for class I marker HEI10 (left) and class II marker MUS81 (right). Bars 5 10 mm. B, Quantification of foci number per male meiocyte. Mean values with SD are presented, whereas individual counts are represented as dots. n.s., Not significant. Asterisks indicate significant difference by pairwise Wilcoxon rank sum test (***P , 0.01. MSH5B is predicted to result in the loss of the entire MutSII domain and a section of the MutSIII domain. MutSII domains bind Holliday junctions and D-loops in vitro (Fukui et al., 2008), so the truncation is likely to lead to a nonfunctional MSH5B protein, providing evidence that TtMSH5B is a pseudogene. Cytological evidence supports this prediction, as a reduction of 85% chiasmata was observed in msh5a mutants, but there was no additive effect in the msh5ab double mutants. MSH4D contains an 8-kb deletion that is predicted to remove 328 amino acid residues from the C terminus of the protein, resulting in the partial loss of the MutSIII domain and the entire MutSV domain, suggesting that TaMSH4D is also a nonfunctional pseudogene. Even in the absence of MSH5B and MSH4D, the obligate chiasma is still maintained in hexaploid wheat.
Allopolyploid cells possess multiple sets of homeologous chromosomes that must pair, recombine, and synapse to ensure accurate chromosome segregation during meiosis to preserve genome stability and reproductive success. In neoallopolyploids, a bias for meiotic recombination to occur between homologues rather than homeologues may not be preadapted, so diploidization will be under strong selection pressure. The mechanisms underlying meiotic stabilization in newly formed allopolyploids are poorly understood, but a recent study in B. napus demonstrated that reducing MSH4 copy number prevented meiotic COs forming between homeologous chromosomes, whereas homologous COs were unaffected (Gonzalo et al., 2019). Furthermore, MSH4 and MSH5 have been shown to systematically reduce to a single copy in numerous polyploid plant lineages, which is more likely due to convergent selection than by chance (Lloyd et al., 2014;Gonzalo et al., 2019). This is consistent with a potential role for MSH5B/MSH4D gene losses in the diploidization of polyploid wheat.
The functional MSH4 and MSH5 homeologues are highly conserved between wheat species at the primary amino acid level. Wheat MSH4 and MSH5 possess unique low-complexity disordered N termini that are not present in Arabidopsis or S. cerevisiae. The S. cerevisiae MSH4 N terminus destabilizes the protein and is targeted for degradation (He et al., 2020), whereas the disordered wheat MutSg N termini do not contain Ser residues and may stabilize the proteins, although functional studies would be required to test this. A stable MutSg complex would be advantageous in a diploid background but may promote homeologous recombination in a neoallopolyploid. In S. cerevisiae, MutSg stabilizes single-end invasions as well as Holliday junction recombination intermediates (Lahiri et al., 2018). The ability of MutSg to stabilize early recombination intermediates comprising divergent heteroduplex sequences in wheat may have been negated by deletions in MSH5B and MSH4D. The class I CO pathway is regulated in a dosage-dependent manner in mouse and Arabidopsis by RNF212 and HEI10, respectively, as well as by MSH4 in B. napus (Reynolds et al., 2013;Ziolkowski et al., 2017;Gonzalo et al., 2019). Homologous COs appear unaffected by gene duplicate loss in wheat and can maintain the obligate chiasma, despite possessing only one functional MSH5 copy in Ttmsh5a-1 2/1 , the minimum dosage through a null mutation. The dosage of MutSg is therefore in excess of that required for normal levels of homologous COs, but this may not be the case for homeologous COs, where it is potentially a limiting step. However, this will need to be confirmed experimentally by reconstituting functional MSH5B in tetraploid wheat and MSH5B and MSH4D in hexaploid wheat.
The novel duplicated copy of ZIP4 on chromosome 5B in the Ph1 locus prevents homeologous COs forming in hexaploid wheat crossed with distantly related species (Rey et al., 2017). As a constituent of the ZMM complex, ZIP4 may interact with either MSH4 or MSH5 directly or indirectly in wheat. In S. cerevisiae, MSH5 and ZIP4 physically interact in yeast two-hybrid assays, Figure 7. Chiasmata are random in number and located predominantly distally in TtmutSg mutants. A to C, Observed and Poisson-predicted distributions of chiasma frequency per cell. A, The observed wild-type distribution deviates significantly from a Poisson-predicted distribution [x 2(21) 5 64.71, P , 0.01]. B, The observed Ttmsh4ab distribution does not deviate from a Poisson-predicted distribution [x 2(10) 5 6.71, P . 0.9]. C, The observed Ttmsh5a distribution does not deviate from a Poisson-predicted distribution [x 2(11) 5 1.85, P . 0.99]. D, The mean proportion of chiasmata (percent) that are distally located. Error bars represent SE. Asterisks indicate significant difference by pairwise Wilcoxon rank sum test (***P , 0.01).
suggesting that this interaction mediates the association of the MutSg dimer with ZIP2-ZIP4-Spo16 complexes to process recombination intermediates from D-loops into stable single-end invasions and Holliday junctions (De Muyt et al., 2018). It has been suggested that high levels of expression of ZIP4 in hexaploid wheat prevent homeologous COs (when crossed with a closely related ancestor) in a dosage-dependent manner by overcoming a threshold (Rey et al., 2017). Based on data from Gonzalo et al. (2019), we hypothesize that higher levels of MutSg expression may stabilize recombination intermediates between divergent heteroduplex sequences in wheat (such as homeologues and distantly related chromosomes) but lower levels of MutSg are sufficient to stabilize intermediates between homologues. However, it may be a coincidence that following tetraploidization, wheat evolved an extra copy of ZIP4 while it lost a functional copy of MSH5 followed by mutation of MSH4D in the hexaploid.

CONCLUSION
In summary, MutSg promotes the formation of class I COs in wheat (;85% of all COs) and maintains the obligate chiasma despite stepwise pseudogenization of MSH5B and MSH4D following polyploidization. Loss of MSH5B and MSH4D does not perturb formation of the obligate chiasma and may even play an adaptive role in meiotic recombination in allopolyploid wheat.

PCR
Total genomic DNA was isolated from T. turgidum 'Kronos' leaf material using the DNeasy Plant Mini Kit (Qiagen). A large, previously unidentified deletion in TtMSH5B was verified experimentally by nested PCR with flanking primers designed to amplify over the predicted gap. The first reaction used external primers TtMSH5B_del_F_ext (59-TGGATGACGACGAGGAGGAG-39) and TtMSH5B_del_R_ext (59-AGATGCTCTGATATTCCTACCTCG-39), and the successive reaction used internal primers TtMSH5B_del_F_n (59-AGGAGG AGGTGGCCGAGAC-39) and TtMSH5B_del_R_n (59-CTGAAACAAGAGAA GCGGGC-39). Cycling conditions were annealing temperature 5 60°C and extension time 5 2 min for both reactions. Amplicons were purified using the E.Z.N.A. Cycle Pure Kit (Omega Bio-tek). Sanger sequencing was performed by Eurofins.

Reverse Transcription PCR
Total RNA was extracted from T. turgidum 'Kronos' spikes using the ISO-LATE II RNA Mini Kit (Bioline), and cDNA was synthesized by reverse transcription using the Tetro cDNA synthesis kit (Bioline). The coding sequences were amplified using subgenome-specific primers (Supplemental Table S3), ligated into pDrive (Qiagen), and Sanger sequenced (Supplemental Fig. S10). The presence of mutations in the coding sequences was confirmed for the selected TILLING lines. Ttmsh4a-1 has a CAG-to-TAG mutation in exon 13, which results in a premature stop codon. Ttmsh4a-2 has a CAA-to-TAA mutation in exon 20, which results in a premature stop codon. Ttmsh4b has a GT-to-GA mutation at the splice donor site between exons 22 and 23, which causes the retention of intron 22 and the formation of a premature stop codon (TAA) 16 codons downstream. Ttmsh5a-1 has a CAG-to-TAG mutation in exon 8, which results in a premature stop codon. Ttmsh5a-2 has a CGA-to-TGA mutation in exon 34, which results in a premature stop codon. Ttmsh5b has a GT-to-GA mutation at the splice donor site between exons 27 and 28, which causes the retention of intron 27 and the formation of a premature stop codon (TGA) 14 codons downstream.

Antibody Production
Total RNA was extracted from T. aestivum 'Cadenza' spikes using the ISOLATE II RNA Mini Kit (Bioline), and cDNA was synthesized with the Tetro cDNA synthesis kit (Bioline). The wheat ASY1 D subgenome coding region (TraesCS5D02G294100) was used as a template to amplify nucleotides 1 to 696 with Q5 DNA polymerase (New England Biolabs) using primers ASY1F (59-A GCATATGGTGATGGCTCAGAAGACG-39) and ASY1R (59-TCCTCGAGGA CACTCTTAACCTTCAAAGC-39). The PCR products were cloned into pDrive (Qiagen) and confirmed by sequencing. The ASY1 fragment was digested by NdeI/XhoI using sites incorporated into the primers (highlighted in boldface), gel purified, and cloned into pET21b (Merck). pET21b-ASY1 was transformed into Escherichia coli Rosetta (DE3) cells expressing the HORMA domain (residues 1-232 of the ASY1 protein). Inclusion bodies were purified and sent for guinea pig antibody production (DC Biosciences). The wheat MUS81 B subgenome coding region (TRAES3BF061700170CFD) was used as a template to amplify nucleotides 895 to 1,728 with Q5 DNA polymerase (New England Biolabs) with primers MUS81F (59-AGCATATGGGTTCTGCTGAAAACTCT C-39) and MUS81R (59-AGCTCGAGTCCTTCAGCCCAGATGA-39). The PCR products were cloned into pDrive (Qiagen) and confirmed by sequencing. The MUS81 fragment was digested by NdeI/XhoI using sites (highlighted in boldface) incorporated into the primers, gel purified, and cloned into pET21b (Merck). pET21b-MUS81 was transformed into E. coli BL21 (DE3) cells expressing residues 299 to 576 of the MUS81 protein. Inclusion bodies were purified and sent for rat antibody production (DC Biosciences). The barley (Hordeum vulgare) full-length HEI10 coding sequence (HORVU6Hr1G040680) was amplified using Phusion Polymerase (New England Biolabs) from anther cDNA of two different varieties (cv Barke and Golden Promise) with oligonucleotides HEI10F (59-ATGAAGTGCAACGCTTGCTGG-39) and HEI10R (59-CTATAACGTGAACATTTGTGGACG-39). The amplicons were cloned into pJET1.2 Cloning Vector (Thermo Fisher Scientific). In all cases, sequencing of clones revealed an absence of the first 27 nucleotides of exon 7 compared with the predicted gene model. The identified HvHEI10 coding sequence was subcloned into an expression vector by Biomatik and used to produce a 294amino acid recombinant protein (without predicted residues 214-222). Recombinant HvHEI10 protein was used for immunization of rabbits and guinea pigs. Rabbit anti-HvHEI10 was affinity purified against recombinant HvHEI10 protein by Davids Biotechnologie.

Supplemental Data
The following supplemental materials are available.
Supplemental Figure S1. Multiple-sequence alignments of MSH5 comparing cultivated polyploid wheats with their diploid wild relatives.
Supplemental Figure S2. Multiple-sequence alignments of MSH4 comparing cultivated polyploid wheats with their diploid wild relatives.
Supplemental Figure S3. Functional MSH4 proteins in polyploid wheats are largely unchanged at the primary protein structure compared with diploid ancestors.
Supplemental Figure S4. Functional MSH5 proteins in polyploid wheats are unchanged at the primary protein structure compared with diploid ancestors.
Supplemental Figure S7. Axis formation and synapsis are unaffected in Ttmsh4 and Ttmsh5 null mutants.
Supplemental Figure S8. Early recombination protein RAD51 loading is unaffected in Ttmsh4 and Ttmsh5 null mutants.
Supplemental Figure S9. Hexaploid wheat maintains the obligate chiasma despite loss of TaMSH5B and TaMSH4D.
Supplemental Figure S10. TtMSH4 and TtMSH5 coding sequences from wild-type and mutant lines.
Supplemental Table S1. Severe reduction of chiasmata in Ttmsh4 and Ttmsh5 null mutants.