A High Throughput Genetic Screen Identifies New Early Meiotic Recombination Functions in Arabidopsis thaliana

Meiotic recombination is initiated by the formation of numerous DNA double-strand breaks (DSBs) catalysed by the widely conserved Spo11 protein. In Saccharomyces cerevisiae, Spo11 requires nine other proteins for meiotic DSB formation; however, unlike Spo11, few of these are conserved across kingdoms. In order to investigate this recombination step in higher eukaryotes, we took advantage of a high-throughput meiotic mutant screen carried out in the model plant Arabidopsis thaliana. A collection of 55,000 mutant lines was screened, and spo11-like mutations, characterised by a drastic decrease in chiasma formation at metaphase I associated with an absence of synapsis at prophase, were selected. This screen led to the identification of two populations of mutants classified according to their recombination defects: mutants that repair meiotic DSBs using the sister chromatid such as Atdmc1 or mutants that are unable to make DSBs like Atspo11-1. We found that in Arabidopsis thaliana at least four proteins are necessary for driving meiotic DSB repair via the homologous chromosomes. These include the previously characterised DMC1 and the Hop1-related ASY1 proteins, but also the meiotic specific cyclin SDS as well as the Hop2 Arabidopsis homologue AHP2. Analysing the mutants defective in DSB formation, we identified the previously characterised AtSPO11-1, AtSPO11-2, and AtPRD1 as well as two new genes, AtPRD2 and AtPRD3. Our data thus increase the number of proteins necessary for DSB formation in Arabidopsis thaliana to five. Unlike SPO11 and (to a minor extent) PRD1, these two new proteins are poorly conserved among species, suggesting that the DSB formation mechanism, but not its regulation, is conserved among eukaryotes.


Introduction
Organisms that reproduce sexually have acquired a specialized type of cell division, called meiosis, which allows the production of haploid gametes from a diploid mother cell. During prophase I of meiosis, homologous recombination occurs and can lead to the reciprocal exchange of genetic material between homologous chromosomes also called crossovers (COs). COs establish physical links between homologous chromosomes, allowing their correct segregation during the first meiotic division.
As a consequence, no overall view of the DSB-forming proteins is available in higher eukaryotes, and only a few players of this recombination step have been described. Mei-P22, for example, was isolated in a large screen for P-element insertions in D. melanogaster [33,34]. The Mei-P22 mutant shows a drastic decrease in c-H2Av foci (marker of DSBs) which can be partially restored by X-Ray treatment indicating that Mei-P22 is required for DSB formation [35]. The Arabidopsis protein AtPRD1 (for Arabidopsis thaliana Putative Recombination initiation Defects 1) and its mouse homologue Mei1 were found by phenotype-based screens for mutations causing infertility as a consequence of meiotic defects [36][37][38] and both were found to be necessary for meiotic recombination initiation [36,37,39].
We therefore screened a saturated collection of Arabidopsis thaliana mutants with the aim of isolating Atspo11-like mutants. With this screen, we successfully identified two types of genes according to their function during meiotic recombination: mutants that repair DSBs using sister chromatids such as Atdmc1 [40] or mutants that are unable to make DSBs like Atspo11-1 [8].
In the latter population, we identified two new DSB-forming proteins, AtPRD2 and AtPRD3, and thus increased the number of proteins necessary for DSB formation in Arabidopsis to five, including AtSPO11-1, AtSPO11-2 and AtPRD1. In addition, we provide new data concerning the Arabidopsis Hop2 homologue (AHP2, [41]) and the meiosis specific cyclin SDS [42] by demonstrating that these proteins are required for homologous chromosome driven DSB repair, probably through a DMC1 dependent pathway.

Results
A genome-wide screen for genes required for early meiotic recombination Mutants defective in meiotic DSB formation have been described in Arabidopsis thaliana [8,15,36]. They display a typical ''asynaptic'' phenotype, characterised by an absence of synapsis during prophase I and a drastic reduction in chiasmata at metaphase I ( Figure 1E-1H). In order to identify the entire set of

Author Summary
During fertilisation, paternal and maternal gametes meet to form the next generation zygote. The zygotic cells therefore contain two sets of chromosomes, paternal and maternal, called homologues. Gamete production depends on the completion of meiosis, during which the chromosome number is divided by two. For this to happen, homologous chromosomes associate into pairs called bivalents, where each chromosome is linked to its homologue by one or several chiasmata. These chiasmata reflect the formation of crossovers, one of the manifestations of the exchange of genetic material occurring during homologous recombination. Meiotic recombination is initiated by the formation of DNA double-strand breaks that are repaired using the homologous chromosome as a template, allowing the stable interactions between them. Although these events are strongly conserved among species, the molecular players are not. In order to investigate recombination in higher eukaryotes, we carried out a high-throughput mutant screen in the model plant Arabidopsis thaliana. The results obtained and presented in this paper describe the isolation and the characterisation of mutants defective in early steps of meiotic recombination from a collection of 55,000 lines, providing the first general view of the molecular players involved in early meiotic recombination in a higher eukaryote. Figure 1. Screening procedure. A mutant collection of 55,000 lines was screened for fertility defects on the basis of reduced silique elongation (arrows, compare: (A) wild type to (B) a low fertility line). In a second step, DIC microscopy on developing pollen mother cells was used to analyse male meiotic products. While normal meiosis produces four haploid cells encased in a single callose wall forming a tetrad of microspores (C), meiotic defects lead to the generation of irregular tetrads or polyads (D). DAPI staining of male meiocytes was then undertaken on meiotic mutants (E-H). In wild type, homologous chromosomes synapse at prophase (E) and five bivalents are easily observed at metaphase I (G). Mutant lines showing synapsis defects (F) and a shortage of chiasma at metaphase I (H) were selected. For each of these, genetic and molecular analyses allowed the construction of complementation groups and the cloning of the mutated genes. For each complementation group, a test for meiotic DSB formation was performed. doi:10.1371/journal.pgen.1000654.g001 proteins necessary for meiotic recombination initiation, we set up the screen presented in Figure 1. Firstly, a saturated Arabidopsis thaliana mutant collection (available at http://www-ijpb.versailles. inra.fr/en/sgap/equipes/variabilite/crg/) was screened for fertility defects. While wild-type Arabidopsis plants reproduce by selffertilisation to produce 50-70 seeds per fruit (silique, arrow Figure 1A), reproduction defects are associated with seed set decrease, and can be scored on the basis of short siliques ( Figure 1B, arrow). Next, differential interference contrast (DIC) observation of male meiotic products in the low fertility lines was undertaken to identify those impaired at meiosis. In wild type, the four male meiotic products forming a tetrahedric structure encased in a single callose wall can be observed just after meiosis ( Figure 1C), whereas meiotic defects result in irregular tetrads or polyads containing a variable number of daughter cells ( Figure 1D). Following this step, we recovered 80 meiotic mutants out of the 55,000 lines of the collection. We then further investigated the meiotic defects of the selected lines by DAPI staining after chromosome spreading ( Figure 1E-1H), and found 28 mutant lines showing the typical ''asynaptic'' phenotype we were looking for (Table 1). The 28 mutant lines (Ws-4 accession) isolated in this study are presented as well as additional alleles (asterisk) obtained in another accession (Col-0). Each of the mutations is presented within its complementation group together with the name of the mutated gene. Chiasma count at metaphase I is provided as well as the nature of the mutation. Full sequence information is provided in Table S1. ** Described in [50], nd: not determined, n: number of cells. doi:10.1371/journal.pgen.1000654.t001 The mutant collection used in this study was generated by T-DNA transformation [43]. Genetic analyses were therefore used to identify mutations that were tagged by the T-DNA insert, which was the case for seven out of the 28 mutants. For these, the mutated genes were recovered by T-DNA border isolation and sequencing (see Materials and Methods). In order to identify the genes in cases where the mutations were not linked to the T-DNA, the semi-sterile mutant plants (Ws-4 background) were crossed to wild-type Col-0 accession. For each F2 populations, 31 mutant plants were selected and genotyped using SNPlex technology (Applied Biosystems, http://www.appliedbiosystems.com) for 48 Ws-4/Col-0 polymorphic markers spanning the whole genome (Text S1). This rough mapping allowed the quick identification of putative allelic mutations. These were then tested by a direct complementation test with referenced mutations or by candidate gene sequencing (see Materials and Methods). For loci that did not colocalise with a putative candidate gene, fine gene mapping was carried out using additional semi-sterile plants that were genotyped for microsatellite markers in the selected genomic region.
Taken together, all these results allowed 9 complementation groups to be defined, which contained 2 to 4 alleles (Table 1 and  Table S1). Five mutations could not be cloned because they were found to map with two unlinked regions of the genome, suggesting that a major translocation had occurred, probably as a consequence of the T-DNA mutagenesis. These lines were not investigated further (Table 1).

Identification of proteins required for DSB formation versus proteins necessary for IH-DSB repair
Among the 9 loci identified in this study, AtSPO11-1, AtSPO11-2 and AtPRD1 were previously shown to be necessary for meiotic DSB formation [8,15,36]. This is not the case, however, for AtDMC1, also identified in our screen. In the Atdmc1 mutant, meiotic DSBs are formed as in wild type but repaired via a RAD51 dependent pathway so that chiasmata are not formed, which explains the presence of ten intact univalents at metaphase I instead of five bivalents [36,40,45]. As Dmc1 is one of the major players involved in inter-homologue (IH) bias observed during meiotic recombination, it is very likely that in Atdmc1 mutants, repair occurs using the sister chromatid rather than the homologous chromosome as the template [40,46].
In order to discriminate between these two situations (DSB formation defects or IH-bias repair defects) we introgressed each of the isolated ''asynaptic'' mutations into mutants defective for meiotic DSB repair such as Atmre11 or Atrad51. As shown previously, in both the Atmre11-3 and Atrad51-1 mutants, DSBs are formed but abnormally processed, leading to pronounced chromosome fragmentation during anaphase I ( [22,47] and Figure 2B and 2C). This DNA fragmentation was abolished in Atspo11-1Atmre11-3 and Atspo11-1Atrad51-1 double mutants ( [22,47]; Figure 2E and 2F) whereas it persisted in both Atdmc1Atmre11-3 Figure 2H) and Atdmc1Atrad51-1 ( [36,45]; Figure 2I), demonstrating that DSBs are absent in Atspo11-1 but present in the Atdmc1 mutant. As shown in Figure 3, chromosome fragmentation in Atmre11 or Atrad51 was also alleviated by of Atprd2-1 or Atprd3-1 mutations but not by crosses with the asy1, sds or ahp2-2/hop2-2 mutants. Thus, our results show that the screen we conducted identified two functional families.
The first family contains AtSPO11-1, AtSPO11-2, AtPRD1, AtPRD2 and AtPRD3, all essential for meiotic DSB formation. To confirm that Atprd2 and Atprd3 were defective in early recombination processes, we analysed the nuclear distribution of the DMC1 protein. Its appearance on meiotic chromosomes during prophase is thought to reflect the progression of recombination repair [48]. Dual immunolocalisation of ASY1 and DMC1 was performed to follow the dynamics of recombination during early prophase I of meiosis. DMC1 foci appear at the leptotene stage of meiosis ( [49]; Figure 4) and then decrease throughout meiotic prophase I until they completely disappear at pachytene stage (not shown). DSB deficient mutants Atspo11-1 and Atprd1 show a total absence of DMC1 foci indicating that localisation of recombinase DMC1 onto chromosomes is dependent upon DSB formation [36]. Likewise, DMC1 staining was completely missing in Atprd2-1 ( Figure 4D-4F) and Atprd3-1 mutants ( Figure 4G-4I). Taken together these data show that no meiotic DSBs are formed in Atprd2 and Atprd3 mutants.
The second functional class isolated in this screen contains AtDMC1, ASY1, AtHOP2/AHP2 and SDS. Our study shows that all four are required for normal levels of chiasma formation and that this is not a consequence of a major defect in the level of meiotic DSB. However, although a weak effect on the meiotic DSB level would probably not have been detected in our experiments and cannot be ruled out, it could certainly not explain the major CO level decrease observed. Since absolutely no trace of chromosome fragmentation was observed associated with the shortage of chiasma in any of the isolated alleles, DSB repair must occur very efficiently in these backgrounds.
The meiosis specific cyclin SDS is required for DMC1mediated DSB repair using the homologous chromosome SDS encodes a meiotic cyclin-like protein, distinct from all other known Arabidopsis cyclins [42]. Here we found that SDS is not required for meiotic DSB formation but is necessary for meiotic DSB repair via the homologous chromosome. In the absence of SDS, DSB repair acts efficiently (since chromosome fragmentation was never observed) but probably onto the sister chromatid, explaining the absence of CO formation. A similar phenotype was previously described for the Atdmc1 mutation and, to a lesser extent, for asy1 ( [40,50], see discussion). Our data therefore suggest that SDS could represent a new factor necessary for meiotic DMC1 driven homologous chromosome bias. In order to test this hypothesis, we analysed DMC1 and RAD51 foci formation in this background.
We used antibodies that recognise either AtDMC1 [36,49] or both AtRAD51 and AtDMC1 (RECA-like, originally raised against the tomato RAD51 protein, [51]). We verified that when we used these anti-RECA like antibodies, foci were still observed in both Atdmc1 ( Figure 5D) or Atrad51-1 (not shown), but had completely disappeared in Atdmc1Atrad51-1 ( Figure 5F), confirming that these antibodies recognize both AtDMC1 and AtRAD51 proteins. Both AtDMC1 and RECA-like foci disappear completely in the DSB defective Atspo11-1 mutant ( Figure 5G and [36]), confirming that they represent sites where DSB are initiated. In the present study, results with both the antibodies in the wild type were very similar to those of the previous studies [36,49,51,52]. Numerous foci appeared at leptotene (Figure 5A and 5D) and decreased progressively until mostly disappeared at pachytene (not shown).
Interestingly, the behaviour of DMC1 localisation in the sds-2 mutant was completely different to that of wild type. No DMC1 chromatin associated foci were observed in sds mutant cells ( Figure 5B and 5C), whereas RECA-like foci still remained ( Figure 5H and 5I), suggesting that mislocalisation of AtDMC1, but not of AtRAD51, is responsible for the recombination defects observed in this mutant.

Characterisation of the two new DSB-forming proteins, AtPRD2 and AtPRD3
Atprd2 and Atprd3 mutants did not show any vegetative growth defects, but showed reduced silique elongation, indicative of fertility defects ( Figure S1). The sterility of Atprd2 and Atprd3 mutants was correlated with abortion of the male and female gametophytes ( Figure S2). Furthermore, examination of earlier stages of male gametophyte development revealed aberrant meiotic products ( Figure S2) correlated with synapsis failure and a total absence of chiasmata as detailed in Figure S3.
The AtPRD2 gene (At5g57880) is 2119 bp long, contains ten exons and nine introns (Accession number FN356233 EMBL-EBI, Figure 6A) and encodes a protein of 385 amino acids (aa, molecular weight 44 kDa, pKi 5.5) with no obvious functional domains ( Figure 7A). Network protein analysis (http://www. predictprotein.org/ and http://npsa-pbil.ibcp.fr/) predicted the presence of several a-helices throughout the protein and three coiled-coil motifs in the middle of the protein (aa 43 to 82, aa 154 to 181 and aa 332-359), suggesting that AtPRD2 is a globular protein [53]. We also observed several S/T-P-X-X and S/T-S/ T-X-X motifs and a helix-turn-helix motif (10 to 31 aa) (http:// npsa-pbil.ibcp.fr/), which are abundant in many DNA-binding proteins [54]. Database searches, using the BLASTP program (Blosum 45), for proteins similar to AtPRD2 produced the highest scores for proteins from other plant species such as Oryza sativa, Populous trichocarpa, Vitis vinifera and Physcomitrella patens (for accession numbers see Materials and Methods). A multiple sequence alignment of AtPRD2 homologues (Bioedit software version 7.0.9.0) revealed overall sequence conservation over the entire length of the protein ( Figure 7A). However, despite using several PSI-blast iterations (Blosum 45) with AtPRD2 and its plant homologues, no significant similarities were found with proteins outside the plant kingdom. Test for meiotic DSB formation in the asynaptic mutants. In wild-type pollen mother cells, the ten Arabidopsis chromosomes are associated into five bivalents. Anaphase I in wild type (A) separates the homologous chromosomes in two equilibrated groups while Atmre11 (Atmre11-3 in B) and Atrad51 (Atrad51-1 in C) mutants show severe chromosome fragmentation at the first meiotic division. In Aspo11-1 or Admc1 mutants, ten intact univalents are observed at metaphase I that segregate randomly at anaphase I (shown for Atspo11-1-1 in D and Atdmc1-1 in G). DNA fragmentation of Atrad51 and Atmre11 is abolished in an Atspo11-1 mutant background (shown for Atspo11-1-1Atmre11-3 in E and for Atspo11-1-1Atrad51-1 in F) but not in an Atdmc1 context (Atdmc1Atmre11-3 in H and Atdmc1Atrad51 in I). Scale bar, 10 mm. doi:10.1371/journal.pgen.1000654.g002 The AtPRD3 gene (At1g01690) is predicted to contain 14 exons, encoding a predicted 1350 bp cDNA for a 449 aa protein (50 kDa, pKi: 10.3). Isolation and sequencing of the full-length AtPRD3 cDNA from flower buds of both Ws and Col-0 accessions, however, identified different PCR products ( Figure 6B). Moreover, the differences we observed between the predicted cDNA sequence and the sequence of the PCR products from AtPRD3 cDNA were at the exon/intron junctions, suggesting that the different cDNA products arose from alternative splicing. These splicing variants are likely to be functional given that the open reading frame is preserved in all the variants sequenced. However, further experiments will be needed to determine whether these variations play a regulatory role. BlastP analysis (Blosum 45) found a significant level of similarity (29% identity and 46% similarity) with the rice meiosis protein PAIR1 sequence (homologous pairing aberration in rice 1) [55]. The rice pair1 mutant exhibits a marked defect in synapsis associated with a total loss of meiotic chiasmata [55]. This is similar to the phenotype observed in Arabidopsis prd3, suggesting that AtPRD3 and OsPAIR1 are orthologues. Sequence similarities were also observed with proteins from other plant species, but not in the basal plant Physcomitrella patens. No AtPRD3 homologues were found outside plants using PSI-BLAST analysis (Blosum 45). Multiple alignment of AtPRD3 homologues revealed the overall conservation of these proteins, with the largest number of conserved residues in the N-terminal section of AtPRD3 ( Figure 6B).
Structural analysis of AtPRD3 and OsPAIR1 proteins failed to reveal any recognizable domains, but we did observe an abundance of a-helixes and several coiled-coil motifs in the centre of the protein ( Figure 7B and [55]).

Screen overview
In budding yeast, initiation of meiotic recombination is catalysed by the Spo11 protein but it is also dependent on numerous additional proteins [1]. Moreover, the behaviour of these proteins (immunolocalisation and physical interactions) revealed that they are interconnected [17,25,[56][57][58][59], suggesting that DSB formation is mediated by a large protein complex. Nevertheless, the existence of a ''DSB complex'' in higher We performed a large scale genetic screen to identify meiotic mutants among 55,000 Arabidopsis T-DNA mutant lines. Eighty mutants affected in meiosis were recovered and 28 showed synapsis defects associated with a drastic reduction in chiasmata formation, defined as a ''spo11-like'' class of mutants. Meiotic genes responsible for the mutations that were cloned were found to be distributed in nine loci. Therefore we can conclude from this screen that at least nine genes are required for normal levels of CO and synapsis in Arabidopsis thaliana. Nevertheless, this number may still be underestimated since genes present as multicopies in the genome are not isolated by T-DNA mutagenesis.

At least five proteins are necessary for DSB formation in Arabidopsis thaliana
Previous studies demonstrated that AtSPO11-1, AtSPO11-2 and AtPRD1 are required for meiotic DNA DSB formation in Arabidopsis [14,22,36,47]. Here, we showed that AtPRD2 and AtPRD3 belong to this same functional family. Thus we now know that at least five proteins are absolutely required for DSB formation in Arabidopsis. In budding yeast, nine proteins in addition to Spo11 are essential for DSB formation, and in S. pombe six have been described (the ScRec114 orthologue Rec7, the Ski8 orthologue Rec14, Mde2, Rec6, Rec15, and Rec24, [30,31,60]) indicating that DSB formation is a complex mechanism which requires the function of numerous proteins.
AtPRD3 is a protein with no recognizable domains and appears to be the orthologue of the -rice PAIR1 [55]. Structural analysis of rice PAIR1 and AtPRD3 shows a cluster of a-helical coiled-coil motifs that are often present in dimeric proteins. However, we did not observe see self-association of AtPRD3 in a yeast two hybrid assay (not shown). Nevertheless, we cannot rule out that selfinteraction occurs specifically during meiosis following meiotic specific post-translational modifications of the protein. This is the case for Mer2 phosphorylation which was shown to be necessary for protein-protein interactions with itself and other ''DSB proteins'' in S. cerevisiae [16,17]. AtPRD2 encodes a putative nuclear protein of 385 aa containing an a-helical coiled-coil structure (Figure 7). The S/T-P-X-X or S/T-S/T-X-X motifs, which are present in many DNA binding proteins [54], are abundant in the N-terminus of AtPRD2 suggesting that AtPRD2 could act on chromosomes to promote DSB formation. Immunolocalisation of AtPRD2 and AtPRD3/AtPAIR1 in wild type and the different context of DSB mutants, as well as an investigation of their physical interaction with other ''DSB proteins'', would clarify their roles during meiotic recombination initiation.
No AtPRD2 or AtPRD3 homologues were found outside of plant eukaryote protein databases, suggesting that they are either too divergent to be recognized, or that they are not conserved across kingdoms. Such a degree of divergence was also observed in other organisms. In budding yeast, for example, the majority of ''DSB proteins'' (Rec102, Rec104, Rec114, Mer2, Mei4) are apparently not conserved in higher eukaryotes. Moreover, even among species closely related to S. cerevisiae, most of the ''DSB proteins'' are weakly conserved, indicating that meiosis specific ''DSB proteins'' have rapidly diverged even in the Saccharomyces genus [2,61]. Indeed, we observed only a weak level of conservation between AtPRD3 and its rice homologue. Furthermore, we could not identify a homologue within the basal plant Physcomitrella patens genome, suggesting that rapid divergence of ''DSB proteins'' also occurred inside the plant kingdom. The six ''DSB proteins'' in S. pombe (Mde2, Rec6, Rec15, Rec24, Rec25 and Rec27) have no clear homologues in any other species. By contrast, Spo11 is relatively well conserved among eukaryotes, suggesting that although the catalytic activity of DSB formation is well conserved, the regulation of this process is more divergent.   AtHOP2/AHP2 and SDS could act with ASY1 to drive DMC1-mediated DSB repair using the homologous chromosome Our screen also led to the isolation of other asynaptic mutants, including mutations in DMC1, ASY1, HOP2 and SDS. No DSB formation defects were detected in these mutants but it appears more likely that DSB repair processes are affected. During meiosis the repair machinery is biased toward inter-homolog (IH) events. Rad51 is needed for both inter-homologue (IH) and inter-sister (IS) repair, while Dmc1 seems to be the important player for mediating IH repair [62] which is consistent with the observed phenotype following DMC1 disruption in Arabidopsis [40]. Similarly for ASY1, recent functional data demonstrated its role in coordinating the activity of the RECA homologues to create a bias in favour of IH repair [50]. The most likely explanation is that in the absence of ASY1, most of the DMC1mediated repair is prevented mostly leading to IS repair [50]. It should be noted, however, that in all the isolated alleles of asy1, the CO level was never below 15% of the wild-type level ( Table 1), showing that the requirement for ASY1 in IH-repair is not absolute. Interestingly, Arabidopsis ASY1 is a HORMA domain protein that exhibits similarity with the S. cervisiae Hop1 [44]. In S. cerevisiae, the absence of the meiosis specific recombinase Dmc1 leads to cell arrest in prophase due to unrepaired DSBs [63,64]. However, an efficient repair of the breaks via inter-sister recombination is observed when Hop1, Red1 or Mek1 is mutated together with Dmc1 [65][66][67][68]. Therefore, it has been proposed that these axis proteins, Hop1, Red1 and the protein kinase Mek1, form a complex that mediates the interhomologue bias in S. cerevisiae by suppressing Dmc1-independent strand invasion during meiosis [66,69,70].
Our study also identified SDS as a new player of IH-bias in plants. SDS is a plant-specific protein, with a C-terminal region possessing strong similarity with Arabidopsis cyclins [42]. Furthermore, it was shown to interact with the Arabidopsis CDKs, Cdc2a (CDKA;1) and Cdc2b (CDKB;1), suggesting that SDS is a cyclin that functions with CDKs in the regulation of meiosis [42]. Our study showed that meiotic DSBs are formed in the sds background, but are repaired efficiently without IH CO formation, probably using the sister chromatid as a template as is the case for Atdmc1 or asy1. We demonstrated that residual RECA-like foci are formed in an Atdmc1 context, suggesting that the repair of DSBs via intersister recombination is RAD51-dependent. We also showed that mislocalisation of AtDMC1 (but probably not of AtRAD51) is associated with the sds meiotic phenotype. These data suggest that in Atdmc1, sds and asy1, DSBs are repaired via an AtRAD51dependent pathway using sister chromatids as a template, showing that both SDS and ASY1 are necessary for DMC1-driven interhomologue repair.
Our study revealed that another locus may be involved in IHbias control: the Arabidopsis Hop2 homologue AHP2 [41]. Interestingly Hop2 together with its partner Mnd1 were proposed to support the activity of the recombinase Dmc1 to promote IH repair [71]. Nevertheless, hop2 knock-out mutants in most species exhibit meiotic DSB repair defects (similar to mutations in the gene encoding its heterodimeric partner Mnd1), suggesting a total absence of DSB repair in these backgrounds. Similarly in Arabidopsis thaliana the ahp2 knock out line shows a total absence of DSB repair [41] whereas the two alleles isolated in this study, probably representing a partial loss of function, appear to be extremely efficient in IS repair. Further investigations will be necessary to understand how this uncoupling of HOP2 function is achieved.

Conclusive remarks
This study provides, for the first time in a higher eukaryote, a global view of the molecular players involved in meiotic DSB formation on one hand and in the choice of homologous partner for DSB repair on another hand. Their studies will undoubtedly provide important information on how these two steps of meiotic recombination are regulated.

Mutation cloning
Linkage between the T-DNA insert (kanamycin resistance marker) and the meiotic phenotype was checked as described in [8] for all Table 1 mutants. When tagged, the mutation was isolated by cloning T-DNA flanking sequences either using TAIL PCR [73] or kanamycin rescue [74]. Untagged mutations were roughly mapped by SNP genotyping as described in Text S1. This rough positioning of the mutation allowed the quick identification of putatively allelic mutations and the identification of candidate genes. For loci which did not colocalise with a putative candidate gene, fine gene mapping was carried out using additional semisterile plants that were genotyped for markers in the selected genomic region (see Table S2 for marker list). The microsatellite and indel markers were amplified by PCR and the length polymorphisms were revealed by agarose gel electrophoresis as described by [75]. CAPS markers were PCR amplified and then digested by the suitable restriction enzyme before agarose gel electrophoresis. This fine mapping reduced the candidate intervals to 56, 418 and 371 kb for AtPRD2, AtPRD3 and AtHOP2 respectively. Then sequencing of the candidate genes was undertaken until mutations were identified. Gene identification was confirmed by sequencing the selected candidate gene in other available alleles recovered by the screen, and by confirming the mutation in T-DNA mutants obtained from NASC (http://nasc. nott.ac.uk/) (see Plant Material section).

Complementation tests
We tested for allelism between the mutations that mapped to similar locations, by crossing heterozygous plants for each of the mutations. When allelic, a quarter of the F1 plants were semisterile.

Double mutant generation
Double mutants were obtained by crossing plants heterozygous for each of the mutations. The resulting hybrids were selfpollinated. PCR screening was used to select the plants homozygous for both mutations among the semi-sterile plants in the F2 progeny.

PCR genotyping
See Text S1.

Microscopy
Comparison of early stages of macrosporogenesis and the development of pollen mother cells was carried out as described in [8]. Preparation of prophase stage spreads for immunocytology was performed according to [76], with the modifications described in [52]. The ASY1 and ZYP1 polyclonal antibodies [76,77] were used at a working dilution of 1:500. The AtDMC1 and AtDMC1/ AtRAD51 antibodies were described in [49,51,52] and used at a working dilution of 1:20 and 1:50 respectively. All observations were made using a Leica DMRXA2 microscope; photographs were taken using a CoolSNAP HQ (Roper) camera driven by Open (D-E) DIC observation of mature ovules. In a wild type-ovule (D) some of the seven cells of the mature embryo sac can be observed (black arrows) whereas, at the same stage of development no embryo sac has developed in mutants (E-F), and only degenerated cells can be seen (white arrows). (G-I) DIC observation of the product of male meiosis. In wild-type (G), the four meiotic products are observed encased in a callose wall forming a regular tetrad of microspores (three out of the four cells can be seen). In mutants (H-I), irregular tetrads and polyads are observed.