All T-DNA insertion mutants analyzed belong to the Columbia accession (Col). Seeds were obtained from the Salk Institute Genomic Analysis Laboratory (SiGnAL, http://signal.salk.edu/cgi-bin/tdnaexpress; Alonso et al., 2003), and provided by the Nottingham Arabidopsis Stock Centre (NASC). The mutants studied were ago2-1 (SALK_003380), ago3-2 (SALK_005335), ago5-4 (SALK_050483), ago8-1 (SALK_139894), and ago9-1 (SALK_127358). Plants were cultivated on a soil mixture of vermiculite and commercial soil (3:1) and grown in a greenhouse under a 16-h light/8-h dark photoperiod, at 18–20°C with 70% humidity. Genotyping of each mutant was performed by PCR using a combination of three primers, one T-DNA specific primer, LBb1.3 (5′-ATTTTGCCGATTTCGGAAC-3′), and two specific primers for the corresponding gene: LPAGO2 (5′-CCATTGTAGGGCTGAGTATGC-3′) and RPAGO2 (5′-CGTTTCCCTGTGGCCTGAACA-3′); LPAGO3 (5′-CGATAGTCCCGACTGACTCTG-3′) and RPAGO3 (5′-AAACAGAGAGACAGTGGACGC-3′); LPAGO5 (5′-CTACCCATCAGGGAGCTAAGG-3′) and RPAGO5 (5′-TTTCTGGACCATATCACAGCC-3′); LPAGO8 (5′-TCCCTGTTTTGGTTCCTTTTC-3′) and RPAGO8 (5′-TCCTGTTCCTGTTTCCATGAC-3′); LPAGO9 (5′-GGGATACATCCATCCCAACA-3′) and RPAGO9 (5′-AGGCCGTATCTTACCACACC-3′).

Fixation of flower buds, slide preparations of PMCs, and fluorescence in situ hybridization (FISH) were carried out according to Sánchez-Morán et al. (2001). The DNA probes used comprise ribosomal DNA 5S and 45S loci to identify individual bivalents (Gerlach and Bedbrook, 1979; Campell et al., 1992), and a 180 bp repeat sequence (pAL1) to detect centromeres (Martínez-Zapater et al., 1986). These probes were labeled with either biotin-dUTP or digoxygenin-dUTP, using a nick translation kit (Roche). For 5-methylcytosine immunodetection, the slide preparations were dried at 60°C (30 min), fixed in 1% formaldehyde (10 min), dehydrated through an ethanol series (70, 90, 100%) and then dried at room temperature. After denaturation using HB50 at 80°C (2 min), slides were rinsed in ice-cold 2×SSC and incubated in 1% BSA (1 h). The preparations were then washed three times in PBS and one in TNT (5 min each). For the immunodetection, slides were incubated with anti-5-methylcytosine raised in mouse (1:50, Eurogenetec) during 30 min and washed three times in TNT (5 min). Finally, a FITC-conjugated anti-mouse antibody (1:500, Abcam) was used.

To test the sensitivity to γ-rays, sterilized seeds were kept in sterile water at 4°C for approximately 24 h and then were exposed to 100, 200, 300, 400, and 500 Gy doses (2.94 Gy/min) from a ¹³⁷Cs source (IBL 437C; CISBIO Bioassays). After irradiation seeds were germinated on MS agar medium. For mitomycin C (MMC, Duchefa) assay, 4-day-old seedlings were transferred to liquid GM with different MMC concentrations (3, 6, 9, and 12 μg/ml). For cisplatin [cis-diamminedichloroplatinum (II), CDDP, Sigma] assay, seeds were germinated on MS agar medium with different CDDP concentrations (15, 30, 50, and 75 μ M). The effects of the individual DNA damage agents on plant growth were evaluated 12 days (MMC) or 14 days (γ-rays, CDDP) after sowing.

We compared different stages of meiotic progression after 4′,6-diamidino-2-phenylindole (DAPI) staining of PMCs from WT and mutant plants. Our observations indicate that in all mutants the crucial processes of pairing, synapsis and HR at first meiotic division occur normally, and also the second division until tetrad formation (Figure 1).

In WT meiosis, unrelated chromosomes are occasionally entangled with one another, usually at zygotene. This fact leads to the production of interlocks (ILs). In such situations, either a bivalent (type I IL) or one unsynapsed chromosome (type II IL) can be trapped between two aligned homologous chromosomes, held in place by regular synapsis to either side (Wang et al., 2009). Thus, ILs are always associated with unsynapsed regions, indicating that they can prevent or hinder normal synaptonemal complex (SC) formation. In Arabidopsis, observations of surface spread meiocytes under electron microscope revealed a mean value of around one IL per nucleus (in cells with more than 70% of synapsis) (López et al., 2008). Later on, by mid-late pachytene, no ILs were observed. Thus, WT meiosis must include a mechanism for their resolution during zygotene-pachytene. In the absence of AGO9, and in striking contrast to WT, 40% of the mid-late pachytene nuclei (n = 43) showed at least one IL (Figures 2A–I). In addition, we have also observed a new IL configuration (type III), only previously reported in Sordaria (Storlazzi et al., 2010), where two homologous chromosomes show nonsynapsed regions (variable in length) not involved in ILs (Figure 2F). These authors have suggested that these configurations could represent cases of incomplete resolution and/or regions that are unable to synapse after IL resolution. It is likely that this high frequency of ILs in ago9-1 can be responsible for a delay in the duration of meiosis at prophase I since the number of diplotene nuclei observed was substantially higher than in WT. Even we observed ILs during this stage (Figure 2J) and onwards (Figures 2K,L). The occurrence of bridges at anaphase I could be consequence of difficulties in IL resolution (Figure 2M). However, the second division was completely normal in ago9-1 PMCs (Figures 2N,O).

Maize AGO104, an ortholog of Arabidopsis AGO9, appears to be required for heterochromatic CHG and CHH methylation (Singh et al., 2011). Also, Havecker et al. (2010) have reported the involvement of AGO9 in the siRNA-directed maintenance of the silencing state of several classes of repetitive DNA, and Durán-Figueroa and Vielle-Calzada (2010) have pointed out that AGO9 is necessary in the ovule to inactivate transposable elements located in the pericentromeric regions of all five chromosomes of Arabidopsis. Taking into account these results and, since centromeric and pericentromeric regions of ago9-1 chromosomes appeared slightly different after DAPI staining respect to WT (Figures 3A,D), we decided to look for changes in 5-methylcytosine distribution, since this modification is mostly associated with heterochromatic regions in plants (Mathieu et al., 2002). No differences between the mutant and WT were found (Figures 3C,F–I). Likewise, no changes in the size and intensity of a signal corresponding to a centromeric probe were observed (Figures 3B,E,G–I).

Chiasma frequency estimations in Arabidopsis are best cytologically assessed at metaphase I where, despite the disadvantage of maximum chromosome condensation there is no risk of confusing chiasmata and relational twists. On the other hand, condensation represents an advantage when FISH is applied. The chromosomes of Arabidopsis can be individually distinguished for the purpose of chiasma counting following FISH with probes for 45S and 5S rDNA (Sánchez-Morán et al., 2001, 2002; López et al., 2012).

Chiasma frequencies were recorded from 41 to 69 FISH-labeled metaphase I PMCs per WT and each of the mutants, according to the criteria established previously by Sánchez-Morán et al. (2001). Data were collected from two or three plants per genotype. Since there were no significant differences in the mean cell chiasma frequencies per cell between them, individual plant data were grouped. In all of the analyzed cells, the five chromosome pairs always form five bivalents, either rods or rings (Figure 4). Rods are bound by chiasmata in one arm, whereas rings have both arms bound by chiasmata. The mean chiasma frequencies per cell, per bivalent, and per bivalent arm are summarized in Table 1 and Supplementary Table S1. ago2-1 showed a significant increase of the mean cell chiasma frequency per cell respect to WT, mainly due to an excess of chiasmata in the long arm of chromosome 4. In contrast, no significant differences for this parameter were observed between WT and ago3-2, ago5-4, ago8-1, and ago9-1. However, except in ago3-2, there was an increase in the chiasma frequency of the short arm of chromosome 4 in all of these mutants.

We have explored whether AGO2 and AGO9 could play a global role in DNA repair by using γ-irradiation and the cross-linking (CL) agents MMC and CDDP. Unfortunately, we were unable to perform this kind of experiments in ago5-4 due to its delay in germination time. We verified that ago2-1 and ago9-1 plants are hypersensitive to γ-rays because after a 100 Gy dose and beyond, they showed a reduction in the percentage of leaf number per plant with respect to WT (Figure 5, Supplementary Table S2), which was also accompanied with a parallel decrease in the fresh weight per plant (Supplementary Table S2). The results obtained also indicated that ago2-1 is more sensitive to γ-irradiation than ago9-1. Further, ago2-1 and ago9-1 plants are hypersensitive to MMC and, again, ago2-1 resulted to be more sensitive to this reagent than ago9-1. In the former the decrease of the parameters analyzed took place at a concentration of 6 μg/ml, while in the latter was at 9 μg/ml (Figure 6, Supplementary Table S2). By contrast, both ago mutants are not hypersensitive to CDDP, even they turned to be more resistant to CDDP than WT from 30 μM concentration and beyond (Figure 7, Supplementary Table S2).

AGO2 localizes to both the nucleus and the cytoplasm (Takeda et al., 2008). Although several publications have reported the involvement of this protein in the viral siRNA pathway (Blevins et al., 2006; Harvey et al., 2011; Wang et al., 2011), its implication in the nat-siRNA pathway is not completely understood. nat-siRNA are originated by the co-expression of overlapping sense/antisense transcripts which triggers the formation of dsRNAs. Once processed into 24-nt nat-siRNAs, they are probably loaded onto AGO2 (and/or AGO3) to conduct the cleavage of homolog transcripts, producing 21-nt-nat-siRNAs (Jin et al., 2008; Jamalkandi and Masoudi-Nejad, 2009). This kind of endogenous siRNAs have been related to environmental stress responses and to the sperm function control during double fertilization (Borsani et al., 2005; Ron et al., 2010; Yao et al., 2010). Further, Wei et al. (2012) detected the production of sRNAs, named diRNAs (DSB-induced sRNAs), in both plant and human cells, generated by DNA damage from the sequences around DSB sites. In Arabidopsis, diRNAs are recruited by AGO2 to mediate DSB repair and their production is dependent on the protein kinase ATR, which has an essential role during DNA damage response (Culligan and Britt, 2008). Wei et al. (2012) also reported that diRNAs probably function downstream of ATM/ATR-dependent H2AX phosphorylation, one of the earliest events after DSB formation (Amiard et al., 2010). Previously, Culligan et al. (2006) demonstrated that AGO2 can be induced by γ-irradiation and Wei et al. (2012) confirmed that the expression of AGO2, but not the expression other AGOs, was highly induced in plants upon γ-irradiation at both mRNA and protein levels. Our results corroborate the role of AGO2 in DSB repair because ago2-1 plants not only are hypersensitive to γ-rays (Figure 5) but also to MMC (Figure 6). However, we obtained opposite results by using CDDP because ago2-1 plants were more resistant to this agent than WT plants (Figure 7).

While γ-irradiation is a powerful DSB-inducing agent, MMC mainly produces inter-strand CLs on DNA (Rink et al., 1996) and CDDP preferentially forms intra-strand CLs (Eastman, 1985). Both types of CLs induce DSBs during DNA synthesis which are mostly repaired by HR. However, some DNA polymerases can bypass intra-strand CLs, making this type of lesion less toxic than that produced by inter-strand CLs. In G1 cells nucleotide excision repair (NER) can remove a subset of CLs (Deans and West, 2011). The results obtained here indicate that AGO2 (and diRNAs) is likely more involved in pathways for reparing inter-strand CLs than in those for repairing intra-strand CLs.The opposite situation was observed in Arabidopsis recq4 plants (defective for the helicase RECQ4) which showed high sensitivity for CDDP but no sensitivity against to MMC (Mannuss et al., 2010), reinforcing the idea that both agents induce at least partially different kinds of DNA damage. In this sense a large proportion of the genes induced by γ-irradiation and CL agents overlaps, but there is a specific subset of genes induced by each alone (Chen et al., 2003; West et al., 2004; Culligan et al., 2006). On the other hand, the resistant phenotype of ago2-1 in response to CDDP might be explained by the fact that HR blocking (and hence the generation of toxic intermediates) could improve the efficiency of other DNA repair pathways, in this case related to the elimination of intra-strand CLs. Indeed, a recent publication has demonstrated that in mammals, AGO2 forms a complex with the recombinase RAD51, indispensable during HR, and promotes its recruitment to DSB sites (Gao et al., 2014).

In Arabidopsis, meiotic HR, initiated by programmed DSBs, is required for proper synapsis of homologous chromosomes and meiotic progression. Hence, and taking into account the role of AGO2 in DSB repair efficiency, we decided to analyze the chromosome behavior in PMCs of ago2-1 plants. At cytological level, meiosis appeared normal (Figures 1, 4) with the only mention that mean cell chiasma frequency was higher than WT (Table 1). Defects in HR and increases in chiasma frequency have also been reported in plants with mutations in FANCM (Fanconi anemia complementation group M), which encodes for a DNA translocase required for the repair of DNA inter-strand CLs (Crismani et al., 2012; Knoll et al., 2012). This finding is particularly interesting per se, but it is also worth of mention that chromosome 4, especially its long arm, is the only responsible for the increase in chiasma frequency. In this context, it has also been reported that a particular chromosome may differ in chiasma frequency among either different mutants or accessions (Sánchez-Morán et al., 2001; Perrella et al., 2010; López et al., 2012).

AGO2 and AGO3 are included in the same Arabidopsis clade. The corresponding genes are very similar to each other, probably they arose from a recent duplication event and are adjacent to one another in the genome, suggesting the two proteins have similar activities and/or redundant functions (Baumberger and Baulcombe, 2005; Vaucheret, 2008). However, since the mean cell chiasma frequency of ago3-2 was similar to that displayed by the WT (Supplementary Table S1), it is tempting to speculate that the functions of both proteins, at least on meiotic HR, could be different.

AGO5, a close paralog of AGO1, directs sRNA-mediated gene expression regulation for all currently characterized Arabidopsis miRNAs (Vaucheret et al., 2004; Baumberger and Baulcombe, 2005). However, in contrast to AGO1, AGO5 expression profile is highly specific to reproductive tissues (Schmid et al., 2005) and accumulates in sperm cell cytoplasm, mature pollen, and growing pollen tubes (Borges et al., 2011). It is also expressed around reproductive cells during megagametogenesis, ago5-4 plants being defective in the initiation of this process (Tucker et al., 2012). The rice genome contains at least five paralogs (MEL1 group) of Arabidopsis AGO5 (Kapoor et al., 2008), but only one exhibits an expression profile similar to AGO5, MEL1. In the seed-sterile mel1-1 mutant, PMCs are arrested at early meiosis I (only ~5% of PMCs escaped meiotic arrest) and meiotic chromosomes present condensation defects, although axial elements are formed (Nonomura et al., 2007). These authors have proposed that MEL1 is involved in the organization of nucleolar organizing regions (NORs) because in the mutant the chromosome axis protein PAIR2 associates with the nucleolus, a situation never observed in WT cells.

Despite the decrease in fertility observed in homozygous ago5-4 plants, meiosis in PMCs was cytologically normal (Figures 1, 4). Although no differences in chiasma frequency respect to WT were observed (Table 1), NOR-bearing chromosomes (2 and 4), and also chromosome 3 displayed a significant increase in the number of chiasmata.

AGO9 is a close paralog of AGO4 and is involved in the ra-siRNA pathway, participating in the maintenance of the silencing state of several classes of repetitive DNA element (Havecker et al., 2010). Like AGO2 and AGO3, AGO8 and AGO9 are almost adjacent each other, suggesting they have arisen from a recent gene duplication event, although AGO8 is probably a pseudogene (Takeda et al., 2008; Vaucheret, 2008). Initially characterized by the absence of any obvious developmental defect (Havecker et al., 2010), a closer examination of ago9 mutant revealed the differentiation of multiple gametic cells from somatic companion cells that are able to initiate gametogenesis (Olmedo-Monfil et al., 2010). However, any publication has analyzed a possible role of AGO9 (and ra-siRNAs) in DNA damage repair. The results obtained were quite similar to those reported for ago2. Thus, ago9-1 was hypersensitive to both γ-irradiation and MMC, although in minor extent that ago2-1, and also was more resistant against CDDP than WT (Figures 5–7; Supplementary Table S1), revealing a possible role for ra-siRNAs and RdDM in the repair or inter-strand CLs and DSBs.

Mutations in maize AGO104 (related to AGO9) produced unreduced female gametes generating an apomixes-like fertilization-independent seed production phenotype. Also, the absence of AGO104 generate severe defects during male meiosis: abnormal condensation during diakinesis-metaphase I, absence of a bipolar spindle at anaphase I and dyads with partial chromosome condensation and irregular spindle formation, which lead to the formation of triads and microspores with multiple nuclei (Singh et al., 2011). However, our results have revealed normal meiosis in PMCs of ago9-1 (Figures 1, 4) with the exception of the high frequency of ILs observed at pachytene, some of them maintained until metaphase I (Figure 2). Since ILs have been observed in many organisms, the exceeding low rate of meiosis failure in WT meiocytes implicates the existence of specific mechanisms to deal with them (Zickler and Kleckner, 1999). One model proposed for their resolution suggests movement of the interlocked chromosomes out through the ends of the constraining bivalents (Scherthan et al., 2007; Conrad et al., 2008; Koszul et al., 2008). However, we have not observed apparent differences on centromere and telomere dynamics at prophase I between ago9-1 and WT PMCs. Another model invokes DNA topoisomerase II in axis breakage and reformation (Von Wettstein et al., 1984). In addition, Storlazzi et al. (2010) have found that MLH1, a central player in HR, is required for IL resolution, suggesting that this process requires elimination of constraining DNA connections formed by the recombination process. Thus, the hypersensitivity to γ-rays and MMC could be related with a defect in HR, revealing an involvement of AGO9 in this process, that could also be the reason of the ILs observed. Further studies will be necessary to comfirm this hypothesis.

AGO9 is important for heterochromatic silencing (Havecker et al., 2010). Even, Durán-Figueroa and Vielle-Calzada (2010) have provided evidence that AGO9 predominantly targets transposable element in the pericentromeric regions of all five chromosomes in the ovule. Also, AGO104 has a strong effect on centromeric repeats and controls non-CG DNA methylation at centromeric heterochromatin (Singh et al., 2011). Although these facts could be related with the different chromatin conformation observed in pericentromeric regions of ago9-1 chromosomes respect to the WT, this is not apparently the case (Figure 3). Additional experiments at molecular level will be required to determine the reason of these observations.

Finally, although mean cell chiasma frequency in ago9-1 plants was not different to that of WT (Table 1), chiasma formation was again significantly increased in short arms of NOR-bearing chromosomes 2 and 4. The same also occurs in ago8-1 (Supplementary Table S1).

Summing up, the results obtained in this work have confirmed that AGO2 has a role in DNA repair, but also that AGO9 could be involved in this process. The fact that both mutants are hypersensitive to both γ-irradiation and MMC, and resistant to CDDP respect to WT constitutes an exciting result because it opens a door to the possibility that these proteins may be involved in a specific pathway of DNA repair, related to the resolution of inter-strand CLs (and/or DSBs). On the other hand, the increase of chiasma frequency, especially in NOR-bearing chromosomes, observed in the analyzed mutants is intringuing and merits further investigation. Finally, it is important to mention that the absence of a clear meiotic phenotype in these mutants could be due to overlapping functions of AGO proteins. The analysis of double or triple mutants, in the case they were viable, could contribute to obtain more information about the function of these proteins.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.