A wide reprogramming of histone H3 modifications during male meiosis I in rice is dependent on the Argonaute protein MEL1

Argonaute (AGO) family proteins are highly conserved in eukaryotes. AGOs, in association with non-coding small RNAs, modulate both transcriptional and post-transcriptional gene silencing during various developmental events (reviewed by Vaucheret, 2008). AGO proteins are classified into two subfamilies, AGO and PIWI. AGO subfamily proteins are expressed ubiquitously, and associate with microRNAs (miRNAs) and small interfering RNAs (siRNAs). The PIWI subfamily proteins bind a distinct class of small RNAs, termed PIWI-interacting small RNA (piRNA; Kim et al., 2009). The piRNA-induced silencing complex plays important roles in germline development in many animals (Aravin et al., 2006; Girard et al., 2006; Lau et al., 2006).

In contrast, plant genomes lack PIWI-clade AGOs (Kapoor et al., 2008). However, several plant AGO proteins specifically acting in germ cell development have been reported. MEIOSIS ARRESTED AT LEPTOTENE1 (MEL1) is a rice AGO protein, which functions in germ-cell development and meiosis (Nonomura et al., 2007). MEL1-clade AGOs include Arabidopsis thaliana AGO5 (Kapoor et al., 2008). Of ten Arabidopsis AGOs, AGO2 and AGO9 are responsible for chiasma frequency and dissolution of meiotic chromosome interlock, respectively, but AGO5 is not required for meiotic events (Oliver et al., 2014). Thus, the meiotic roles of MEL1-clade AGOs are still largely ambiguous. MEL1 is expressed in premeiotic germ cells in a gender-neutral fashion; the mutant phenotypes confirm that the wild-type MEL1 function in faithful progression of premeiotic mitosis and meiotic homolog synapsis (Nonomura et al., 2007; Komiya et al., 2014). Recently, deep sequencing of MEL1-associated small RNAs has revealed that MEL1 prefers a unique class of 21-nucleotide (nt) siRNAs, which have a conserved cytosine residue at the 5′-terminus and are expressed abundantly in reproductive organs of rice (Komiya et al., 2014). The precursors of the MEL1-associating siRNAs (masiRNAs) are transcribed specifically in the reproductive phase from >1000 intergenic loci (Komiya et al., 2014), mostly corresponding to the 21-nt phased siRNA (phasiRNA) precursor (PHAS) loci, reported previously (Johnson et al., 2009). Both 21-nt masiRNAs and phasiRNAs supposedly target few transposons, but lack obvious target RNAs (Song et al., 2012; Komiya et al., 2014; Zhai et al., 2015). The miR2118, which targets the 22-nt conserved motifs in most of the 21-nt PHAS loci, is broadly conserved in angiosperms and gymnosperms (Zhai et al., 2011). In contrast to grass species, the dicot plants retain the PHAS loci as overlapping with protein-coding regions; the PPR family genes in Arabidopsis (Chen et al., 2007; Howell et al., 2007) and NB-LRR defense family genes in the legume Medicago (Zhai et al., 2011). Thus, the biological functions of the MEL1–masiRNA complex have largely remained elusive. Furthermore, the existence of thousands or more of 21-nt masiRNA and phasiRNA species has made it difficult to narrow down their target RNAs to date.

Dynamic chromatin reprogramming is indispensable for faithful progression of meiosis in most eukaryotes, including plants. For example, meiotic double-strand break (DSB) positions are highly correlated with specific chromatin marks, such as trimethylation of histone H3 lysine 9 (H3K9) in yeast (Pan et al., 2011) and mammals (Brick et al., 2012), and recruitment of histone H2AZ, an H2A variant, in Arabidopsis (Choi et al., 2013). In Arabidopsis megaspores, a highly dynamic chromatin reprogramming, coinciding with meiotic DNA replication, is required for the transition of cell fate from somatic to germline (She et al., 2013). Transcriptional inactivation, represented by rapid H3K9 deacetylation and removal of active RNA polymerase II (RNAPII) from meiotic chromosomes, takes place at early meiotic stages in mouse meiocytes (Page et al., 2012). During these stages, meiotic DSB initiation and repair, and homologous chromosome synapsis progress. However, the biological roles of the dynamic chromatin remodeling during plant meiosis have been ambiguous.

In the present study, we performed immunocytological analyses of Oryza sativa pollen mother cells (PMCs) in the mel1 mutant, and investigated the possibility of MEL1 being involved in the global control of histone modifications in meiotic chromosomes, prior to homologous recombination and pairing. The nucleolar organizing region (NOR), encoding hundreds of tandem-repeated 45S ribosomal RNA (rRNA) genes (Oono and Sugiura, 1980), was affected by the mel1 mutation differently from most chromosomal regions. This study provides new insights in the relationship between the AGO-mediated epigenetic control and the pivotal meiotic events in plants, such as pre-synaptic centromere association, meiotic DSB initiation and homologous chromosome synapsis.

Generally, commitment to meiosis, including meiosis-specific DNA replication, occurs during premeiotic interphase. Meiotic prophase I is divided into five substages; leptotene, zygotene, pachytene, diplotene and diakinesis. In flowering plants, programmed DSBs during meiosis are initiated and repaired, and homologous pairs are synapsed from zygotene to pachytene (Hamant et al., 2006). To determine the meiotic stages precisely, two synaptonemal complex components were used: PAIR2 as an axial component (Nonomura et al., 2006) and ZEP1 as a central transverse filament (Wang et al., 2010) (Fig. S1A, see Materials and Methods for the details).

In the mel1 mutants of Oryza sativa, the development of germ cells was arrested at the meiotic leptotene stage, despite the display of meiotic cell features, specified by their enlarged volume and meiosis-specific gene expression (Nonomura et al., 2007; Komiya et al., 2014). To get further insights into the mel1 phenotype, meiotic DSB formation was examined in PMCs by immunostaining of phosphorylated histone H2AX (γH2AX). γH2AX was transiently present in the DSB-containing chromatin stretches (Kuo and Yang, 2008). In leptotene and zygotene, the signal was widespread along the wild-type rice chromosomes (Fig. 1A,B), but almost disappeared during the stages from zygotene to pachytene (Fig. 1C). In contrast, it was hardly detectable in mel1 mutant PMCs (Fig. 1D,E), indicating that meiotic DSB initiation was largely impaired.

Centromere (CEN) association is another pivotal meiotic event. CEN association often occurs among the nonhomologous CENs, prior to homolog synapsis, and is required for proper homolog recognition and pairing in eukaryotes, including many plant species (Aragón-Alcaide et al., 1997; Martinez-Perez et al., 2001; Prieto et al., 2004; Zhang et al., 2013). As meiosis proceeds, CEN association is replaced by homologous CEN pairing by pachytene (Tsubouchi and Roeder, 2005; Bisig et al., 2012; Obeso et al., 2014). It was observed that the number of CEN foci was reduced in both the wild-type and mel1 PMCs with the progression from pre-leptotene to meiotic leptotene (Fig. 2A). This reduction took place at leptotene, prior to the homolog synapsis, and was defined as presynaptic centromere association. The CEN number seemed unchanged in the subsequent pachytene (Fig. 2A), although, from cytological observations, we judged that CEN associations were switching to the homologous CEN pairing in the wild type. In contrast, in the mel1 mutant, CEN association was unresolved and the CEN number further decreased from leptotene to pachytene (Fig. 2A). Unresolved CEN association frequently resulted in the appearance of extremely enlarged CEN foci (Fig. 2B). These results imply that MEL1 has some functions in establishing the meiosis-specific CEN structure that is essential for facilitating the dissociation of CEN associations.

Meiotic chromosome synapsis, represented by ZEP1 loading between the homologous pairs, was fully disrupted in the mel1 mutant (Fig. S1B), as reported previously (Komiya et al., 2014). Taken together, we conclude that MEL1 is required for pivotal meiotic events, such as centromere association, DSB initiation and homolog synapsis.

Loss of pivotal meiotic events indicated the possibility that the structure and modification of meiotic chromosomes was globally altered in the mel1 mutant. To confirm this, we focused on the H3K9 dimethylation (H3K9me2) and acetylation (H3K9ac). H3K9me2 is associated with gene silencing and heterochromatin structure (Soppe et al., 2002). H3K9ac is a histone mark often associated with active transcription and is antagonistic to H3K9 methylation (Rando, 2007).

During pre-leptotene, H3K9me2 was maintained at a low level in the wild-type PMCs (Fig. 3A). However, it was remarkably intensified in the late leptotene (Fig. 3B). The intense signal was also observed in pachytene and metaphase I (Fig. 3C,D) and eventually declined during anaphase and telophase I (Fig. 3E). The median of overall H3K9me2 intensity was significantly increased by 5.5-fold during the premeiosis-to-meiosis transition (Fig. 3F). These results were obtained with a monoclonal anti-H3K9me2 antibody product, mAbcam1220 (Abcam). A similar H3K9me2 dynamism in wild-type PMCs was reproduced with another antibody product, 05-1249 (Upstate) (Fig. S2A–E), although the overall intensity was weaker than with the mAbcam1220 antibody.

The H3K9ac staining pattern was in contrast to that of the H3K9me2. The H3K9ac intensity was maintained at a moderate level at pre-leptotene (Fig. 3G), but was extremely reduced in late leptotene and pachytene (Fig. 3H,I). Although the signal sometimes recovered (data not shown), it weakened again by metaphase I (Fig. 3J) and eventually recovered at telophase I (Fig. 3K). The transient signal recovery at diakinesis was seemingly because of the rapid chromosome compaction. The median of H3K9ac intensity was significantly reduced by 2.7-fold during the premeiosis-to-meiosis transition (Fig. 3L).

These results suggest that H3K9s were reprogrammed broadly, and dimethylated and deacetylated during the meiotic prophase I in the wild-type PMCs.

We next examined the large-scale meiotic alteration of H3K9 modifications in the mel1 mutant PMCs. As assessed by use of the antibody Abcam1220, the mel1 PMCs exhibited lower H3K9me2 levels at pre-leptotene (Fig. 4A), as did wild-type cells (Fig. 3A). In leptotene-like PMCs arrested by the mel1 mutation, unlike in the wild type, the reduced H3K9me2 levels continued even after meiotic entry, which was indicated by PAIR2 loading onto the chromosomes (Fig. 4B). H3K9me2 signals disappeared from most of the chromatin regions in the later stages (Fig. 4C). The intensity was significantly lower than that of the wild type in both pre-leptotene and early meiotic stages (Fig. 3F). The same tendency was also observed upon use of the 05-1249 antibody (Fig. S2). In contrast, unexpected hyperacetylation of H3K9 was frequently observed at pre-leptotene in the mel1 PMCs (Figs 3L and 4D), more than in wild-type PMCs. The hyperacetylation continued in cells at leptotene-like or later stages (Fig. 4E). The median of the H3K9ac intensity was not reduced during premeiosis-to-meiosis transition, but rather increased in the mel1 mutant (Fig. 3L). The above observations indicate that the reprogramming events are dependent on MEL1 function.

Crosstalk between phosphorylation and other modifications at the histone H3 N-terminal tail has been reported. In mammals and yeast, H3K9ac is promoted by the acetyltransferase recruited onto H3 phosphorylated at S10 (H3S10ph), which is adjacent to the K9 acetylation site, resulting in transcriptional activation and K9 hypomethylation (Cheung et al., 2000; Lo et al., 2001). In wild-type rice PMCs, H3S10ph was maintained at moderate levels during pre-leptotene to pachytene and was reduced during zygotene to metaphase I (Fig. 5A–E). In contrast, extremely intense H3S10ph signals were observed in the pre-leptotene mel1 PMCs (Fig. 5F), with a median significantly higher (2.3-fold) than in the wild-type pre-leptotene PMCs (Fig. 5I). Different from in the wild type, the H3S10ph levels in pre-leptotene PMCs did not decline, but rather increased at leptotene and pachytene in the mel1 mutant (Fig. 5G–I). This pattern was largely consistent with the H3K9ac pattern (Fig. 3G–L), suggesting that MEL1 functions to repress H3S10ph, as well as H3K9ac, in the context of the large-scale meiotic chromosome reprogramming observed above in PMCs.

Generally, H3K9me2 is known as a silencing mark in plants (Soppe et al., 2002). Thus, H3K9me2 accumulation on meiotic chromosomes suggests that transcription is inactivated widely in wild-type rice PMCs. However, a previous analysis using premeiotic flowers suggests that mRNA transcription patterns are almost comparable between the wild type and mel1 mutant (Komiya et al., 2014). In the meiotic anthers used in this study, no difference was detected in amounts of 18S and 25S ribosomal RNAs by quantitative RT-PCR (Fig. S3A, see legend for further details regarding quantitative RT-PCR). Given that flowers and anthers contain numerous somatic cells, transcriptional differences due to the mutation in PMCs, if any, could be masked by the larger amounts of unaffected transcripts from somatic cells. The results suggest that somatic anther cells are unaffected by the mel1 mutation, as expected.

Thus, the transcription status in meiotic PMCs was estimated by analyzing the immunostaining pattern of phosphorylated Rpb1, the largest subunit of RNAPII. Phosphorylation of Rpb1 at different residues in its C-terminal domain (CTD) correlates with the RNAPII transcriptional activity; phosphorylation at S5 (S5ph) is associated with initiation and elongation of RNAPII-dependent transcripts, and that at S2 (S2ph) with the termination of transcription (Komarnitsky et al., 2000). In cultured mammalian cells, numerous minute RNAPII foci are linked with active transcription, and enlarged speckles are associated with the inactive state (Bregman et al., 1995; Zeng et al., 1997).

In the wild-type rice PMCs, numerous minute signals of both S5ph and S2ph were scattered in the pre-leptotene nuclei (Fig. 6A; Fig. S3B). However, from leptotene to zygotene, the number of scattered foci was reduced, and many large speckle-like structures emerged in all ten PMCs observed (Fig. 6B-D; Fig. S3B). These signals disappeared by metaphase I (Fig. 6E; Fig. S3B). Few S5ph signals overlapped with S2ph signals during the meiosis I stages (Fig. S3B), suggesting that the two types of CTD phosphorylation are also associated with different transcriptional steps in rice cells. Taken together, the appearance of enlarged speckled-foci of S5ph and S2ph in the wild-type PMCs (Fig. 6C,J) suggests that large-scale meiotic alteration of histone modifications is accompanied by the inactivation of transcription in rice.

At pre-leptotene, the staining pattern of both S5ph and S2ph in the mel1 mutant PMCs was largely comparable with that of the wild type (Fig. 6F; Fig. S3B), except in the ribosomal DNA (rDNA) repeat regions (see below). However, unlike in the wild type, numerous minute foci of phosphorylated RNAPII continued to appear during meiosis I in all ten PMCs observed (Fig. 6G–I,K), although the signal eventually disappeared, as in the wild type (data not shown). The frequency of the appearance of enlarged S5ph speckles was slightly decreased in mutant PMCs (27.1% of all foci counted) compared to that in wild-type PMCs (13.2%) (Fig. 6L). In addition, the size and intensity of foci were significantly reduced in the mel1 mutant (Fig. 6M). These results indicate that the inactivation of transcription is accompanied by large-scale histone modifications during wild-type meiosis I, and that it is also dependent on the MEL1 function.

Nonomura et al. (2007) have demonstrated that there is an aberrant aggregation of a meiotic protein PAIR2 on the NOR, composed of the tandem-repeated ribosomal 45S RNA genes (rDNA). The present study reconfirmed the previous result (Fig. 4B,C). In addition to PAIR2 accumulation, excess amounts of RNAPII S5ph and S2ph were enriched on NORs in the mel1 mutant PMCs (Fig. 6H,I). The aberrant PAIR2- and active RNAPII-enriched NORs were often observed as adhering to the external periphery of the nucleolar region, which was detectable as a ‘black hole’ without 4′,6-diamidino-2-phenylindole (DAPI) staining (Fig. 4B,C; Fig. S3C). When using the 05-1249 antibody, an intense H3K9me2 signal was observed as overlapping PAIR2 and RNAPII signals on NORs (Figs S2F–H and S3C). Similar NOR signals were also observed upon use of the Abcam1220 antibody, but the intensity was much fainter than with 05-1249 (Fig. S3C), although the reason was unclear. It was clear that compared to wild-type PMCs, more H3K9s were dimethylated in rDNA regions in mel1 mutant PMCs, but further analyses will be necessary.

Interestingly, a hypermethylation at H3K9 was also detected at CEN regions in mel1 mutant PMCs (Fig. S2H). The reason is ambiguous, but the result might relate to the mutant phenotype of centromere dissociation in mel1 PMCs (Fig. 2).

The present study demonstrates that large-scale and meiosis-specific chromatin remodeling occurs during rice meiosis I. In the wild-type PMCs, most of the meiotic chromosomal regions except for the rDNA repeat regions were reprogrammed as indicated by increased H3K9me2, and decreased H3K9ac and H3S10ph (Figs 3 and 5). This is consistent with the results observed in the Arabidopsis female megaspore mother cell (MMC), in which cell fate transition was accompanied by large-scale chromatin reprogramming, including increased H3K9me2 and a reduction in active RNAPII (She et al., 2013). Hereafter, we designate the MEL1-dependent meiotic events demonstrated in this study as large-scale meiotic chromosome reprogramming (LMR). LMR would also be expected to occur in rice MMCs, because MEL1 functions in a gender-neutral fashion (Nonomura et al., 2007).

The rice LMR was concomitant with the alteration of the active RNAPII distribution pattern during the premeiosis-to-meiosis transition (Fig. 6), likely representing global transcriptional inactivation of meiotic chromosomes in PMCs. This finding is reminiscent of meiotic silencing of unpaired chromosomes (MSUC) in mice (Barchi et al., 2005; Bellani et al., 2005; Turner et al., 2005), which is accompanied by reduced H3K9ac and inactive RNAPII (Page et al., 2012). In male mice, unpaired regions of X and Y chromosomes are compartmentalized in the sex-body and are silenced (Turner, 2007). In addition, unpaired autosomal regions are accompanied by pseudo-sex body formation and are transcriptionally silenced by MSUC, leading to meiocytes undergoing apoptosis due to a ‘pachytene checkpoint’, which is widely present in the non-plant species (Roeder and Bailis, 2000). Pseudo-sex bodies are formed at pachytene, when meiotic transcription is resumed at the paired regions. Thus, MSUC might be a surveillance mechanism to remove meiocytes that could accidentally cause unpairing of the autosomes. In other words, global transcriptional reactivation at mid-pachytene exposes silenced MSUC regions to the checkpoint machineries.

However, the timing of relieving rice LMR (anaphase or telophase I; Figs 3 and 5) was different from that of mice MSUC (mid-pachytene) (Page et al., 2012). This distinction might be due to the difference in surveillance systems between plant and non-plant species. It is widely accepted that the pachytene checkpoints do not occur in plants (Caryl et al., 2003). The absence of the pachytene checkpoint might facilitate the survival of unreduced gametes and the long-term persistence of the polyploidy (Li et al., 2009), which is generally thought to be advantageous for inheritance of genetic diversity and heterosis in plant species (Comai, 2005). Rice LMR was observed almost throughout the meiosis-I stages (Figs 3 and 5), and, thus, plant meiocytes cannot be discriminated by the transcriptional state of the unpaired regions. Therefore, plants might have evolved a transcription-independent system to monitor the homologous chromosome synapsis for survival.

Another possibility is that LMR determines the proper positions of meiotic DSBs rather than controlling the transcriptional state. In Arabidopsis, meiotic crossover hotspots overlap with histone-H2AZ-enriched nucleosomes at the promoter regions of genes (Choi et al., 2013). H2AZ activates transcription by protecting the promoters from DNA methylation (Zilberman et al., 2008). Its depletion caused by the loss of ARP6, a nucleosome-remodeling complex, reduces the foci of meiotic proteins, such as RAD51, DMC1 and MLH1 during meiosis I (Choi et al., 2013). These studies suggest that H2AZ-enriched chromatins are required for both active transcription and meiotic DSB formation. H3K9ac is also found at the 5′-end of coding regions, in addition to H2AZ and H3K4me3, and their levels correlate with the transcription rate (Rando, 2007; Choi et al., 2013). Thus, it is possible that the MEL1-dependent LMR restricts transcriptionally active genes to determine the sites for meiotic DSB initiation. In this case, the increased H3K9ac level in the mel1 mutant PMCs might cause too many regions that can be preferred for the meiotic DSBs, thereby disturbing proper DSB formation (Fig. 1).

Different from most of the other chromosomal regions, the H3K9me2 level was exceptionally increased at the NOR in the mel1 mutant, in addition to the aberrant loading of PAIR2 and phosphorylated RNAPII (Figs 4 and 6; Figs S2 and S3B,C). The aberrant PAIR2 and RNAPII accumulations were never observed in somatic cells in mutant anthers (Fig. S4), indicating that MEL1-dependent reprograming events happened only in meiotic PMCs. In Arabidopsis, silenced and heterochromatic rDNA sequences are epigenetically allocated at the nucleoplasm, towards the outside of the nucleolar territory (Lawrence and Pikaard, 2004; Earley et al., 2006, 2010), as observed in this study (Fig. S3C). Thus, PAIR2-, RNAPII- and H3K9me2-enriched NORs in the mel1 mutant PMCs are supposed to be highly heterochromatinized and silenced. This situation seemed to be the opposite of that in the other chromosomal regions, in which the H3K9me2 level was extremely reduced in the mutant (Figs 3A–F and 4).

One possibility to explain this discrepancy is that the hyper-methylation at H3K9 in NORs in the mel1 mutant is a by-product to compensate for reduced H3K9me2 on most chromosomal regions. Another possibility is that the rDNA repeat region is prevented from undergoing LMR during wild-type meiosis I in a manner that is dependent on MEL1 function. In this context, concomitant aggregation of PAIR2 on the NOR is suggestive. Rice PAIR2 is a HORMA domain protein (Nonomura et al., 2004, 2006), a functional homolog of yeast HOP1 and Arabidopsis ASY1 (Hollingsworth et al., 1990; Sanchez-Moran et al., 2007), both of which are essential for establishment of homologous chromosome synapsis and chiasma formation. In yeast, HOP1 is prevented from invading nucleolar territory and NOR by the meiosis-specific AAA+ family ATPase PCH2 to suppress non-allelic homologous recombination (NAHR) between rDNA repeats (San-Segundo and Roeder, 1999; Vader et al., 2011). NAHR in repetitive regions results in structural or copy-number changes, such as deletion, duplication and inversion, leading to genome destabilization (Sasaki et al., 2010). Taken together with H3K9me2 accumulation resulting in heterochromatinization of NORs (Earley et al., 2006), keeping NORs away from the LMR or heterochromatinization during meiosis I might be required for excluding meiotic machineries from the NOR, probably resulting in NAHR inhibition and maintenance of genome stability, as well as for protecting rRNA genes from silencing to secure meiotic cell homeostasis.

A recent study deciphered that Dcr1, the sole Dicer protein in fission yeast, promotes termination of RNAPII-dependent transcription to avoid collision with DNA replication (Castel et al., 2014). RNA-interference-mediated knockdown of Dcr1 caused an alteration in RNAPII loading in broad genomic regions, including those for rDNA. The rDNA repetitive regions are known as ‘difficult-to-replicate’ regions, because passage of transcription complexes frequently stalls DNA replication (Alzu et al., 2012; Sabouri et al., 2012). Interestingly, in the dcr1 mutant, H3K9me2 is slightly enriched in rDNA, which is different from other genomic regions, and the rDNA copy number is extremely reduced through meiotic NAHR (Castel et al., 2014). In addition, RNAPII S5ph and S2ph are aberrantly accumulated on rDNA in the dcr1 mutants (Castel et al., 2014). Thus, the phenotypes of the yeast dcr1 mutant resemble those of the rice mel1 mutant, as demonstrated in this study. In fission yeast, the RNA silencing complex, including AGO1, induces H3K9 methylation in an siRNA-dependent manner, to create binding sites for SWI6, the heterochromatin protein 1 (HP1) homolog (Hutvagner and Simard, 2008). These findings prompted us to hypothesize that MEL1 is involved in the maintenance of genome stability and rDNA copy number through transcriptional control during rice meiosis. Although the effect of MEL1 in rRNA transcription could not directly be verified in this study (Fig. S3A), how MEL1 balances the transcriptional activity between NORs and non-NORs will be an important question.

Subcellular localization is important for understanding the molecular action of AGOs. Cytoplasmic AGOs are generally thought to function in post-transcriptional gene silencing through mRNA degradation and translational inhibition (Baumberger and Baulcombe, 2005; Leung et al., 2006; Lanet et al., 2009; Li et al., 2013). The cytoplasmic MEL1 AGO is also supposed to have similar functions (Komiya et al., 2014). If it is so, the meiosis-specific control of chromatin modifications, as demonstrated in this study, should be mediated by unidentified genes and proteins, downstream of the MEL1 pathway.

However, it is also possible that MEL1 is transported into the meiotic nuclei and directly engages with the dynamic control of meiotic chromatin modifications. In fact, nuclear localization of MEL1 has been reported in leptotene meiocytes, although infrequently (Komiya et al., 2014). Arabidopsis AGO4 associates with heterochromatic-siRNAs (hc-siRNAs) at the cytoplasm and mediates RNA-directed DNA methylation (RdDM), a nuclear process (Law and Jacobsen, 2010; Li et al., 2006; Pontes et al., 2006). The cytoplasmic Arabidopsis AGO4 assembly with hc-siRNAs is thought to induce conformational change in the AGO4 protein around the nuclear localization signal and promotes the nuclear import of the AGO4–hc-siRNA complex (Ye et al., 2012). The relationship of LMR to nuclear MEL1 function warrants further examination.

In conclusion, the present study highlights the importance of meiosis-specific chromatin remodeling, LMR, for pivotal meiotic events during rice meiosis I. LMR accompanies the dynamic alteration of chromatin modifications, such as H3K9me2, H3K9ac and H3S10ph, and transcription inactivation represented by the RNAPII-loading pattern in the meiotic nuclei. Cytological characterization of mel1 phenotypes suggests that LMR is a highly controlled epigenetic process and that the MEL1 protein is a key regulator of LMR. Furthermore, the impact of rRNA transcriptional control on meiosis is strongly suggested in rice. This finding implies the existence of unknown epigenetic mechanisms to protect repetitive rDNA regions from the deleterious NAHR during plant meiosis. This system might be shared with other genomic repeats, because meiotic behavior of centromeres was somewhat affected by the mel1 mutation (Fig. 2; Fig. S2H). MEL1 localizes in the cytoplasm in premeiosis, but is competent to transfer into the nucleus at the onset of meiosis (Komiya et al., 2014). Generally, cytoplasmic AGOs are engaged in translational control, whereas nuclear AGOs are involved in transcriptional silencing. Whether cytoplasmic or nuclear localization of MEL1 or both is more essential for LMR is an important question. Reproductive phasiRNAs might affect MEL1 subcellular localization, because the 21-nt and 24-nt phasiRNAs are expressed preferentially during premeiosis and meiosis, respectively (Zhai et al., 2015), and MEL1 is able to bind both classes (Komiya et al., 2014). Further studies on MEL1 will shed more light on the unidentified epigenetic mechanisms to assure faithful meiosis progression in plants.

The rice (Oryza sativa L. subspecies japonica) cultivar Nipponbare was used as wild-type plants. The mel1 homozygous mutants were selected from selfed progenies of Tos17-inserted heterozygous plants (Nonomura et al., 2007), which were backcrossed five times with cultivar Nipponbare. Genotyping in the MEL1 locus was performed as described by Nonomura et al. (2007). All plants were grown in a field in the city of Mishima, Shizuoka, Japan.

Rice flowers at premeiosis and meiosis were fixed and stored as described by Nonomura et al. (2006). The flower length and/or anther length were referred to for estimating meiotic stages of flowers and anthers, according to Itoh et al. (2005).

Rice proteins PAIR2 and ZEP1 were used as intracellular meiotic markers. PAIR2 begins to accumulate in the nucleoplasm just following meiotic DNA replication, and subsequently, at meiotic leptotene, associates with the axial element of the synaptonemal complex (Nonomura et al., 2006) (Fig. S1A). During zygotene and pachytene, axial PAIR2 proteins are gradually removed from chromosomes and replaced by ZEP1 loading between homolog pairs (Wang et al., 2010) (Fig. S1A). In this study, ‘premeiosis’ or ‘pre-leptotene’ is defined as the cellular stage at which PMCs retain decondensed chromatins and accumulate PAIR2 proteins through the nucleoplasm, but not on chromatin.

Subcellular localization of MEL1, PAIR2, ZEP1 and CenH3 was observed by indirect immunofluorescent staining of PMCs, as described previously (Nonomura et al., 2006, 2007; Komiya et al., 2014). The antibody against rice γH2AX was produced according to the method described by Miao et al. (2013) and used at 1:150. To examine histone H3 modification, mouse monoclonal anti-H3S10ph antibody (1:150, cat. no. 05-598, Upstate) and rabbit polyclonal anti-H3K9ac antibody (1:150, cat. no. 06-942, Upstate) were used. For H3K9me2 detection, two different products of the mouse monoclonal anti-H3K9me2 antibody were used; cat. no. 05-1249 (Upstate) and mAbcam1220 (Abcam) at 1:75 and 1:150, respectively. Anti-PIIS2ph antibody (1:150, cat. no. ab5095, Abcam) and anti-PIIS5ph antibody (1:150, cat. no. ab5408, Abcam) were used to stain the different phosphorylated forms of RNA polymerase II. Antibodies for PAIR2 and ZEP1 were used as meiotic stage markers (Nonomura et al., 2006; Wang et al., 2010). All immunofluorescent images were obtained by confocal laser scanning microscopy with a FV300 microscope (Olympus). To make the subcellular localization of immunofluorescent foci clearer, a photo of a single optical section was taken for each nuclear image, except for counting CenH3 foci (see below), and processed by the software Photoshop (Adobe).

For validation of the centromere association, stereo 3D images of a PMC were viewed through an Olympus FV1000-D microscope. PMCs were stained with antibodies against CenH3 and PAIR2. CenH3, a histone H3 variant specifically involved in centromeric chromatins, was used to monitor the centromere. PAIR2 was used as a marker for meiotic-stage progression. In Fig. 2A, the number of CenH3 foci was counted on a merged image of multiple optical sections from the same meiocyte. The largest CenH3 focus was selected in each meiocyte, and the pixel value of each focus was calculated by ImageJ (Schneider et al., 2012) and plotted in the graph shown in Fig. 2B.

The signal intensity of H3K9me2, H3K9ac and H3S10ph was quantified according to a previous method (She et al., 2013) with slight modifications. Briefly, PMCs were immunostained with antibodies for H3K9me2, H3K9ac or H3S10ph, and counterstained with 4′,6-diamidino-2-phenylindole (DAPI). Antibody against PAIR2 or ZEP1 was used as a marker for meiotic-stage progression. A nuclear midplane image was taken in each H3K9- and DAPI-stained PMC observed, and their pixel values were calculated by ImageJ. The intensity of H3K9me2, H3K9ac or H3S10ph was divided by that of DAPI for normalization, and plotted (Figs 3F,L and 5I). In the figures, each plot was obtained by combining results from two or three biological replicates together, because in all wild-type samples, the replicates showed a similar variance of the pixel values to each other in accordance with one-way ANOVA (data not shown).

To investigate RNAPII activity during meiosis I, PMCs were immunostained with anti-PIIS5ph or PIIS2ph antibodies and counterstained with DAPI. A nuclear midplane image was taken for each PMC observed. The images of RNAPII S5ph foci were binarized in ten PMCs each for the wild type and the mel1 mutant, to determine the areas to be analyzed, and the pixel density of each focus area was measured by ImageJ.

We thank Mitsugu Eiguchi (NIG) for great help in growing plant materials. We also thank Dr Robert A. Martienssen (Cold Spring Harbor, USA) for useful comments and discussions. We are grateful to Drs Takahiro Kusakabe (Kyushu U., Japan) and Katsutoshi Tsuda (NIG, Japan) for useful comments and reading the manuscript. The mel1 mutant seeds were kindly provided by National Bioresource Project (NBRP) Rice, Japan Agency for Medical Research and Development (AMED).