MTOPVIB interacts with AtPRD1 and plays important roles in formation of meiotic DNA double-strand breaks in Arabidopsis

Meiotic recombination is initiated from the formation of DNA double-strand breaks (DSBs). In Arabidopsis, several proteins, such as AtPRD1, AtPRD2, AtPRD3, AtDFO and topoisomerase (Topo) VI-like complex, have been identified as playing important roles in DSB formation. Topo VI-like complex in Arabidopsis may consist of subunit A (Topo VIA: AtSPO11-1 and AtSPO11-2) and subunit B (Topo VIB: MTOPVIB). Little is known about their roles in Arabidopsis DSB formation. Here, we report on the characterization of the MTOPVIB gene using the Arabidopsis mutant alleles mtopVIB-2 and mtopVIB-3, which were defective in DSB formation. mtopVIB-3 exhibited abortion in embryo sac and pollen development, leading to a significant reduction in fertility. The mtopVIB mutations affected the homologous chromosome synapsis and recombination. MTOPVIB could interact with Topo VIA proteins AtSPO11-1 and AtSPO11-2. AtPRD1 interacted directly with Topo VI–like proteins. AtPRD1 also could interact with AtPRD3 and AtDFO. The results indicated that AtPRD1 may act as a bridge protein to interact with AtPRD3 and AtDFO, and interact directly with the Topo VI-like proteins MTOPVIB, AtSPO11-1 and AtSPO11-2 to take part in DSB formation in Arabidopsis.

of Rec114 and Mei4 to axis sites [17][18][19] . RMM binds to chromosome axial sites, but DSBs occur at hotspots in chromatin loops 17,20 . The mechanism of tethering these loop sites to the axis may explain this paradox 17,20,21 . Recent findings showed that Spp1 (a component of the COMPASS complex) can recognize and bind to H3K4me2/me3 marks on chromatin loops to axis sites through its direct interaction with Mer2 22,23 . Finally, Mre11-Rad50-Xrs2 (MRX) complex is recruited during DSB formation and functions as a nuclease for DSB repair 24,25 .
In mammals, SPO11, MEI1, MEI4 and REC114 have been identified as functioning in DSB formation. Mutation in MEI1 causes sterility in mice and reduced γH2AX (a phosphorylated form of histone H2AX that is a marker for DSBs) signals. MEI1 encodes an unknown function protein and is meiosis-specific in DSB formation 26,27 . Recent findings showed that MEI1 is required for recruiting MEI4 to chromosome axes and that MEI4 directly interacts with REC114 28,29 .
In Arabidopsis, six proteins, namely AtSPO11-1, AtSPO11-2, AtPRD1, AtPRD2, AtPRD3 and AtDFO, were found as being involved in DSB formation [30][31][32][33][34] . AtPRD1 is homologous to MEI1 of humans and mice 26,27,32 . It has a conserved N-terminal region that can interact with AtSPO11-1 and itself 32 . AtPRD2 is homologous to Mei4 33,35 . AtPRD3 is a plant-specific protein for DSB formation and is the ortholog of the rice PAIR1 33,36 . AtDFO is also a plant-specific protein with conserved function-unknown domains 34 . Studies have shown that AtSPO11-1 and AtSPO11-2 are involved in DSB formation 30,31 , but how the Arabidopsis Topo VI-like complex interacts with other proteins involved in DSB formation remains unclear.
Recently, it was reported that TOPVIB-like proteins, which are homologous to archaeal Topo VIB proteins, are required for DSB formation in Arabidopsis (MTOPVIB) 37 , mice 38 and rice 39,40 . Here, we report our independent characterization of the MTOPVIB gene using the mutant alleles mtopVIB-2 and mtopVIB-3. mtopVIB-3 exhibited reduced fertility, polyads and abnormal tetrads. It had a weaker meiotic phenotype compared to mtopVIB-2. MTOPVIB could interact with AtSPO11-1 and AtSPO11-2. AtPRD1 could interact with the Topo VI-like complex proteins (MTOPVIB, AtSPO11-1 and AtSPO11-2). AtPRD1 may act as a bridge protein to recruit AtPRD3 and AtDFO to form a subcomplex, and may interact directly with Topo VI-like complex to promote meiotic DSB formation in Arabidopsis.
Mutation in MTOPVIB affects embryo sac development. The mtopVIB ovules at different stages ( Fig. 2a-h) were compared to those in the wild-type plant ( Fig. 2i-l) using laser scanning confocal microscopy (LSCM) 43 . At the developmental stage FG0, the mtopVIB-2 and mtopVIB-3 ovules appeared normal as in the wild-type plant (Fig. 2a,e,i). At the FG1 stage, the unique nuclei of functional megaspores appeared smaller in mtopVIB-3, and no nucleus was observed in the mtopVIB-2 embryo sacs (Fig. 2b,f), compared to those in the wild-type plant (Fig. 2j). At the FG2 stage, the two-nucleate embryo sacs in mtopVIB-2 were not observed, instead, strong auto-fluorescent signals found in the mutant embryo sacs (Fig. 2c). In mtopVIB-3, single-nucleate embryo sacs were observed (Fig. 2g), compared to the wild-type embryo sacs, which had two nuclei after the first nuclear division of the functional megaspore (Fig. 2k). At the developmental stage FG7, the embryo sacs in mtopVIB-2 and mtopVIB-3 did not have the typical nuclear and cellular organization/structure, and they had only strong auto-fluorescent signals instead (Fig. 2d,h), compared to a typical four-celled mature embryo sac in wild-type plants (Fig. 2l), indicating that the embryo sacs in mtopVIB mutants had stopped developing and were degenerated. In particular, the mtopVIB-2 and mtopVIB-3 mutants had only 2.80% (n = 283) and 42.5% (n = 246) normal embryo sacs, respectively, compared to 99.3% (n = 133) in wild-type plants. The results indicated that the mutations in MTOPVIB showed abnormal embryo sac formation.

mtopVIB-2 and mtopVIB-3 mutants are defective in meiosis.
To investigate whether the defect in tetrad formation was caused by abnormal meiosis in the mtopVIB-2 and mtopVIB-3 mutants, the chromosome behaviors in the mutant male meiocytes were examined by comparison with those of the wild-type plant. At the leptotene stage in the wild-type plant, the chromosomes appeared as single strands. The chromosomes then underwent synapsis from the zygotene to pachytene stages, which appeared as thick chromosome threads ( Supplementary Fig. S3a-c). In the mtopVIB-2 and mtopVIB-3 male meiocytes, the chromosome behaviors displayed no obvious differences at the leptotene and zygotene stages ( Supplementary Fig. S3e,f,i,j). At the pachytene stage, the chromosomes in the wild-type plant were shorter and thicker, whereas in the mtopVIB-2 mutants, the chromosomes appeared thinner, and in the mutant mtopVIB-3, the chromosome structure was found with a weaker phenotype than that in mtopVIB-2 ( Supplementary Fig. S3c,g,k). Five normal pairs of highly condensed chromosome bivalents were observed in the male meiocytes of the wild-type plant at the diakinesis stage. At metaphase I, wild-type meiocytes had five well-aligned bivalents at the equatorial plate leading to equal chromosome segregation at the anaphase I stage ( Supplementary Fig. S3d,m,n). In the mtopVIB-2 and mtopVIB-3 mutants, most chromosomes were appeared as univalents in the meiocytes at diakinesis. At metaphase I, compared to 5 bivalents (n = 89) in wild-type meiocytes, mtopVIB-2 showed no bivalent, but 10 univalents (n = 90), and mtopVIB-3 showed average number 3.2 of bivalents per meiocyte (n = 74), leading to improper chromosome segregation at anaphase I ( Supplementary Fig. S3h,l,r,w,s,x). At metaphase II, the meiocytes in the wild-type plant  had two sets of aligned chromosomes, and finally segregated to form four nuclei, each of which contained five chromosomes ( Supplementary Fig. S3o,p). In the mtopVIB-2 and mtopVIB-3 mutant meiocytes at the same stage, misaligning of chromosomes was observed, and chromosome segregations at the anaphase II stage were then abnormal ( Supplementary Fig. S3t,u,y,z). As a result, the mtopVIB-2 and mtopVIB-3 meiocytes generated four or more abnormal daughter cells and formed irregular tetrad and polyads after meiosis, instead of the normal tetrad ( Supplementary Fig. S3q,v,a2).
In female meiosis, the mutant chromosomes at the leptotene stage were as normal as those in the wild-type plant ( Supplementary Fig. S4a,e,i). At the pachytene stage, thick chromosome threads were not observed in mtopVIB-2 and mtopVIB-3 mutant ovules, as in the wild-type plant ( Supplementary Fig. S4b,f,j). Moreover, at the metaphase I stage, the five bivalents were aligned on the metaphasic plate in the wild-plant (n = 77). In contrast, no bivalents in mtopVIB-2 (n = 85) and average number 2.4 of bivalents per meiocyte in mtopVIB-3 female meiocytes (n = 83) were observed (Supplementary Fig. S4c,g,k). At the telophase I stage, chromosomes in the female mutant meiocytes were also segregated unequally like those in the male meiocytes ( Supplementary Fig. S4d,h,l). Taken together, these results indicated that the mutations in MTOPVIB drastically affected both male and female meiotic division.
Co-immunolocalization of ASY1 and ZYP1 was determined to investigate whether synapsis was defective in mtopVIB-2 and mtopVIB-3. ASY1 is associated with the chromosome axes 44 , while ZYP1 is a major component of the central element of the synaptonemal complex (SC) 45 . The ASY1 signals showed no obvious differences between the wild-type and mutant meiocytes at the pachytene stage ( Supplementary Fig. S5a,b,e,f,i,j). In contrast, no ZYP1 signals in mtopVIB-2 and weak ZYP1 signals in mtopVIB-3 were observed compared to those in wild-type meiocytes ( Supplementary Fig. S5c,d,g,h,k,l). The results showed that synapsis did not occur in mtopVIB-2 and was clearly reduced in mtopVIB-3 meiocytes.

mtopVIB-3 is caused by a point mutation in MTOPVIB.
Genetic analysis showed that the F1 progeny from the self-pollinated heterozygous mtopVIB-3 mutant plants segregated in an approximate ratio of 3 normal:1 abnormal (172:68), implying that mtopVIB-3 might be a single, recessive nuclear mutation. To identify the mutation point, the mtopVIB-3 mutant in Ler background was used as a female to cross with the wild-type Col plant. F2 plants with the sterile phenotype and without Ds insertion were selected and used for map-based cloning. mtopVIB-3 was initially mapped to the bottom arm of chromosome 1 and was then precisely mapped in between the BAC clones T13D8 and F8A5 ( Supplementary Fig. S6a). Sequencing of all the genes in the region between BAC T13D8 and F8A5 revealed three notable mutations, of which two, A693G and A800G (counting from the first putative start codon ATG), were the same as those in the wild-type Col genome in AT1G60450. The other, G1890A (counting from the first putative start codon ATG), was located at the last nucleotide of the eighth exon of AT1G60460 (MTOPVIB) in mtopVIB-3 ( Supplementary Fig. S6b), which was not found in the wild-type genome. Therefore, we speculated that G1890A in AT1G60460 might be related to the phenotype of mtopVIB-3.
RT-PCR assays using the RNAs from mtopVIB-3 and wild-type inflorescences showed that only the expression of AT1G60460 was detected as being changed among the ten genes in the region between BAC T13D8 and F8A5 of mtopVIB-3, compared to those in the wild-type plant ( Supplementary Fig. S6c). The real-time PCR assay showed that the expression level of mtopVIB-3 was reduced to 0.51-fold of the wild-type expression level ( Supplementary Fig. S6d). Monoclonal sequencing of the RT-PCR products indicated that the MTOPVIB RNAs were changed in mtopVIB-3. Seven wild type MTOPVIB transcripts showed two types: five normal transcripts and two transcription intermediates containing the 136-bp intron behind the eighth exon (Table 1; Supplementary  Fig. S6e). The sequences of the thirteen RT-PCR products from mtopVIB-3 could be catalogued into four different types (Table 1). First, five transcripts contained the 136-bp intron behind the eighth exon (the point mutation site), which could be the transcription intermediates like those detected in wild-type, might encode 493 aa (221 R-K ) ( Supplementary Fig. S6f). Second, four transcripts lacked the 43-bp sequence upstream of the mutation site. Third, one transcript had an extra 32 bp behind the mutation point. Fourth, two transcripts had an extra 11 bp behind the mutation point ( Supplementary Fig. S6e). In the last three cases, new GU-AG intron splicing sites were identified as being different from the primary splicing site of the eighth exon in the wild-type plant, and might cause early termination of translation. These results implied that the point mutation dramatically affected the splicing of the MTOPVIB RNAs in mtopVIB-3 mutant. The change in splicing sites that led to truncation of the MTOPVIB protein and might cause a leaky phenotype in mtopVIB-3 ( Supplementary Fig. S7). In mtopVIB-2, the T-DNA was inserted in the fifth intron of MTOPVIB. No normal MTOPVIB transcript was detected in the homozygous mtopVIB-2 plants (Supplementary Fig. S6g-i). The T-DNA insertion also disfunctioned the MTOPVIB protein ( Supplementary Fig. S7).

Plants Samples Sequencing of transcripts
To confirm that the mtopVIB-3 phenotype was caused by the point mutation (G1890A) in MTOPVIB, a 5.959-kb full-length genomic DNA fragment of MTOPVIB, including 2.382-kb promoter region, 3.577-kb coding region and 289-bp 3′-terminal region, was cloned into the pCAMBIA1300 vector (CAMBIA, Canberra, Australia) and then introduced into the mtopVIB-3 plants by Agrobacterium-mediated infiltration. In total, 14 independent transgenic plants were obtained, 12 of which had full seed set restored like that in the wild-type plant ( Supplementary Fig. S8a). All the siliques from these transgenic mtopVIB-3 plants were morphologically normal and had full seed set ( Supplementary Fig. S8b). The siliques from complemented mtopVIB-3 plants had a similar length as that of the wild-type plant (Supplementary Fig. S8b,c). In particular, the seed set rate in the complemented mtopVIB-3 plants was restored to 97.7% (n = 663), close to 97.9% (n = 512) of the wild-type plant, compared to 41.7% (n = 510) in the non-transgenic mtopVIB-3 plants (Supplementary Fig. S8d). This result showed that the cloned genomic fragment could encode the full functions of the MTOPVIB protein. Thus, the meiotic defect in the mtopVIB-3 plants was caused by the mutation in MTOPVIB.
Real-time PCR assays showed that MTOPVIB was expressed ubiquitously in different tissues, including roots, stems, leaves, seedlings, inflorescences, siliques and pollen grains, specially as a higher level in inflorescences ( Supplementary Fig. S9a). MTOPVIB protein localized on chromosomes as described by Vrielynck et al. 37 . By fusion of the MTOPVIB protein with the reporter GFP under the control of the cauliflower mosaic virus (CaMV) 35 S promoter, strong GFP-MTOPVIB signals were detected in the nuclei and weak ones in the cytoplasm (Supplementary Fig. S9e-g). The GFP signals in the control were distributed in the whole cell ( Supplementary  Fig. S9b-d). This result indicated that overexpression of MTOPVIB can lead to distribute GFP signals in nuclear and cytoplasm.
The mtopVIB mutants are defective in meiotic DSB formation. The F1 plants from crosses of mtopVIB-3 (Ler) with mtopVIB-2/+ (Col) were used to analyze the chromosome recombination frequency. Two markers (F3P11 and T16B24) on chromosome II and two markers (T26D22 and K6M13) on chromosome V were used to measure the recombination frequencies between each of the two pairs. The recombination rate was 10.9% for chromosome II and 14% for chromosome V in mtopVIB-2/mtopVIB-3, compared to 21.4% and 30.5% in mtopVIB-3/+ plant (control) ( Table 2), respectively, indicating that mtopVIB-2 and mtopVIB-3 had a significant reduction in recombination.

MTOPVIB interacts with AtPRD1.
To further investigate whether MTOPVIB is associated with DSB formation in Arabidopsis, assays for the interactions of MTOPVIB with the known proteins that are involved in DSB formation were performed. Yeast two-hybrid (Y2H) and bimolecular fluorescence complementation (BiFC) assays were performed. Western blot was firstly to detect that these proteins were expressed in yeast cells ( Supplementary Fig. S10a,b). Then using Y2H and BiFC to carry out protein interactions and found that MTOPVIB could interact with AtSPO11-1 and AtSPO11-2 (Supplementary Tables S2, S4; Fig. S11a-c). These results suggested that MTOPVIB (Topo VIB) could associate with AtSPO11-1 and AtSPO11-2 (Topo VIA) to form Topo VI-like complex.
AtPRD1 encodes a function-unknown protein with 1330 aa in Arabidopsis, and contains two potential coiled-coil structures in the C-terminal region (800-850 aa and 1280-1330 aa), which may be related to protein-protein interaction (https://npsa-prabi.ibcp.fr/cgi-bin/primanal_lupas.pl). To examine the roles of different parts of AtPRD1 in its interaction with MTOPVIB and other related proteins, the AtPRD1 protein was divided    Table S3; Fig. S11f-h), but AtPRD1-B, AtPRD1-C and AtPRD1-E could not interact with any of the proteins tested. These results implied that AtPRD1-A might be an important region for its interactions with MTOPVIB, AtSPO11-1, AtSPO11-2, AtPRD3 and AtDFO. The yeast three-hybrid (Y3H) assay was performed to examine the relationship of AtPRD1 with AtSPO11-1 and MTOPVIB using the combination of a prey construct expressing MTOPVIB and a bait construct expressing AtSPO11-1 and AtPRD1-A bridge protein (Fig. 5a). Alternatively, the assay was also performed using the combination of a prey construct expressing AtSPO11-1 and a bait construct expressing MTOPVIB and AtPRD1-A (1-388 aa) bridge protein (Fig. 5a). Under the conditions that AtPRD1-A (1-388 aa) was not expressed in the transgenic yeast cells grown in SD/-Leu-Trp-His-Ade medium, MTOPVIB could interact with AtSPO11-1. Under the conditions that induced the expression of AtPRD1-A (1-388 aa) in the transgenic yeast cells grown in SD/-Leu-Trp-His-Met medium, MTOPVIB also could interact with AtSPO11-1. The results suggested that AtPRD1-A (1-388 aa) expression did not affect the interaction of MTOPVIB with AtSPO11-1 in yeast (Fig. 5b).
The relationship of AtPRD1 with AtDFO and AtPRD3 was then examined using the combination of a prey construct expressing AtDFO and a bait construct expressing AtPRD3 and AtPRD1-A (1-388 aa) bridge protein, or the alternative combination of a prey construct expressing AtPRD3 and a bait construct expressing AtDFO and AtPRD1-A (1-388 aa) bridge protein (Fig. 5a). AtPRD3 and AtDFO could not interact with each other when the truncated AtPRD1-A was not expressed in the transgenic yeast cells grown in SD/-Leu-Trp-His-Ade medium, but they did when the truncated AtPRD1-A was expressed in the transgenic yeast cells grown in SD/-Leu-Trp-His-Met medium (Fig. 5b). Furthermore, the BiFC assay showed that AtPRD1 could mediate the interaction of AtPRD3 and AtDFO in onion epidermal cells (Fig. 5c). These results suggested that AtPRD1 did not influence the formation of Topo VI-like complex, but might act bridge interaction with AtPRD3 and AtDFO.

Discussion
DSB formation is an important step of meiotic chromosome recombination that is widespread among plant sexual reproduction. Several proteins have been identified as being involved in DSB formation in Arabidopsis, including MTOPVIB, which was reported recently by Vrielynck et al. 37 . However, the roles of these proteins in DSB formation remain unclear. Furthermore, little is known about how these proteins work to promote DSB formation in plants including Arabidopsis. In this study, we independently investigated the roles of MTOPVIB in DSB formation by characterization of mtopVIB mutants and the MTOPVIB protein. mtopVIB-3 was caused by point mutation in the last base pair (G1890) of the eighth exon in MTOPVIB. The mutation of G1890A caused alternated splicing of the related introns and disrupted the function of the gene, leading to significant reduction in fertility of the plants. However, some MTOPVIB proteins (493 aa (221) R-K ) might exist and function normally in mtopVIB-3, which lead to mtopVIB-3 showed leaky phenotype with some bivalents. The mtopVIB-2 was a  Atrad50-1 (c), Atrad51 (d) and Atmre11 (e) that exhibited chromosome fragmentations. (f-j) The meiocytes at metaphase I from mtopVIB-2 (f), mtopVIB-2 Atcom1-1 (g), mtopVIB-2 Atrad50-1 (h), mtopVIB-2 Atrad51 (i) and mtopVIB-2 Atmre11 (j) had ten univalents, in which no chromosome fragmentation was observed. Bars = 10 μm.
knockout allele which disrupted formation of bivalent. Genetic analyses showed that functional loss of MTOPVIB severely affected male and female gametophyte formation, leading to drastic reduction of seed set. The cytological characterization of the mtopVIB mutants suggested that MTOPVIB plays important roles in meiosis. These results are consistent with the data recently reported by Vrielynck et al. 37 .
This study demonstrated that MTOPVIB is involved in DSB formation instead of DSB repair. First, the mtopVIB mutants were found to be defective at the pachytene stage, implying that MTOPVIB is related to homologous chromosome synapsis and recombination. Furthermore, the assays using the synapsis marker proteins ASY1 (as lateral elements) and ZYP1 (as transverse elements) 44,45 showed that the mtopVIB mutant meiocytes were defective in synapsis, which is also similar to the characteristic of the mutants involved in DSB formation. In addition, the chromosome recombination frequencies were dramatically reduced in the mtopVIB mutants (Table 2), also suggesting that the MTOPVIB gene is related to DSB formation.
Our results also showed that Arabidopsis MTOPVIB could interact with AtSPO11-1 and AtSPO11-2, suggesting that they can interact with each other to form a Topo VI-like complex and catalyze DSB formation. A  previous study showed that AtPRD1 is involved in DSB formation and that the N-terminal region of AtPRD1 could interact with itself and AtSPO11-1 32 . Our results showed that AtPRD1 also could interact with MTOPVIB and AtSPO11-2.
Furthermore, as demonstrated by Y2H and BiFC assays, AtPRD1 also interacted with AtPRD3 and AtDFO. AtPRD1 is homologous to human MEI1 in amino acid sequence, with a conserved domain in its N-terminal region 27,32 . The structural analysis showed that AtPRD1 (46-576 aa) might contain putative armadillo repeats, which exhibited a strong similarity to β-catenin in human or importins in yeast (https://www.swissmodel.expasy. org/), and might mediate recognition and interaction with its partners. Furthermore, the truncation assays indicated that AtPRD1-A (1-388 aa) had a stronger affinity in the interaction with Topo VI-like complex, AtPRD3 and AtDFO, respectively. The AtPRD1 also contains coiled-coil domains that may be involved in protein-protein interactions. However, our truncation assays showed that coiled-coil-containing truncated fragments AtPRD1-C and AtPRD1-E could not interact with other proteins tested, suggesting that the coiled-coil domains in AtPRD1 may not be required for its interaction with other component proteins.
Y3H assay showed that AtPRD1 did not have any impact on the interaction of MTOPVIB and AtSPO11-1, suggesting that it did not influence the formation of Topo VI-like complex. Y3H and BiFC assays showed that AtPRD1 might be a bridge protein to interact with AtDFO and AtPRD3. Therefore, AtPRD1 may recruit AtPRD3 and AtDFO to form a subcomplex, and interact with Topo VI-like complex to promote DSB formation. In mice, MEI1 (the homologue of AtPRD1) could recruit MEI4 (the homologue of AtPRD2) to localize to axial sites 29 . However, in Arabidopsis, AtPRD1 did not interact with AtPRD2, but could directly interact with AtPRD3 and AtDFO and the Topo VI-like proteins. AtPRD1 may take part in recruiting these proteins and may affect the localization of these proteins. Nevertheless, more studies are required to verify the hypothesis.

Methods
Plant materials and growth conditions. All Arabidopsis thaliana materials used in this study are of Landsberg erecta (Ler), Columbia (Col-0) and Wassilewskija (Ws) backgrounds. The mtopVIB-3 mutant was isolated as described previously 41,50 , and GK-314G09-015819 (mtopVIB-2) was obtained from the Arabidopsis Biological Resource Center (ABRC). The mtopVIB-3 is backcrossed with wide-type plant (Ler) three times to generate the purified mtopVIB-3 without Ds insertion for further application. The other T-DNA insertion mutants used in this work, namely Atrad51 (SAIL_873_C08), Atrad50-1 (WiscDsLox402F12), Atmre11 (SALK_067823) and Atcom1-1 (SALK_061706C), were obtained from ABRC. All the Arabidopsis seeds were surface-sterilized and plated on Murashige and Skoog (MS) agar plates in cool (4-7 °C) conditions for 2-4 days. The plates were transferred to the growth room with 16 h light/8 h dark cycles at 22 °C for seed germination and further growth. The resulting 7-day-old seedlings were transplanted to soil for further growth under the same conditions. Phenotype analysis. The Arabidopsis wild-type and mutant plants were photographed with a Canon digital camera (Canon, Tokyo, Japan). Laser scanning confocal microscopy of ovules were performed as described previously 1 . The pollen grains and dissected tetrads were collected from wild-type and mutant plants, then stained with 4′,6-diamidino-2-phenylindole (DAPI) and Alexander solution as described previously 1 . Morphology of the pollen grains was observed using scanning electron microscopy (SEM, HITACHI, TM3000). Preparation of Arabidopsis meiotic chromosome spreads was performed as described previously 51 . The samples were stained with DAPI solution (0.1 M sodium phosphate, pH 7.0, 1 mM EDTA, 0.1% Triton X-100 and 0.25 mg/mL DAPI). The images of the chromosome spreads were captured using OLYMPUS CCDDP26 microscope (http:// olympus-imaging.cn) and Leica DM2500 microscope (http://www.leica.com).
Immunolocalization. Immunofluorescence was performed as previously described with minor modifications 52 . The primary antibodies were diluted at 1:200 (anti-ASY1) or 1:100 (anti-ZYP1) in blocking buffer (1 × PBS containing 0.1% Tween 20, 5% BSA and 1 mM EDTA) and added to the slides, followed by overnight incubation at 4 °C in a moisture chamber. The slides were washed with washing buffer (1 × PBS containing 0.1% Tween-20) three times (15 min each time). The secondary antibodies diluted at 1:400 in blocking buffer were added to slides, followed by incubation in a moisture chamber at 37 °C for 60 min in the dark. The slides were washed three times (15 min each) with washing buffer, and then stained with DAPI and observed using an OLYMPUS CCD DP26 fluorescence microscope.
Mapping cloning and complementation experiment. The purified mtopVIB-3 mutant plants were crossed with wild-type (Col) plants to generate hybrid plants. The individual F2 plants with sterile phenotype were selected from the 1500 mutant F2 plants and used for mapping cloning. Simple sequence length polymorphism (SSLP) markers (Supplementary Table S5) were used to map the mutation in MTOPVIB. The mutation was primarily located between the BAC clones F2J6 and T6C23 on chromosome 1. The further mapping using the markers of T30E16, T13D8, F8A5, F23C21, T7P1, F11P17 and F8K4 located the mutation position of mtopVIB-3 in the region between the BAC clones T13D8 and F8A5. Finally, the mutation region covering 10 candidate genes was sequenced and then compared to the corresponding wild-type sequences from the GenBank database.