Somatic small RNA pathways promote the mitotic events of megagametogenesis during female reproductive development in Arabidopsis

Unlike animals, in which male and female gametes are derived from germline stem cells, plants produce their gametes from somatic cells that are directed onto specialised pathways based on their position within the reproductive organ, rather than their lineage (Huala and Sussex, 1993; Nonomura et al., 2007). Formation of the female gametophyte (FG) in Arabidopsis ovules begins with megasporogenesis when a single sub-epidermal somatic cell differentiates from nucellar tissue located at the distal tip of the immature ovule, and is specified as a megaspore mother cell (MMC; Fig. 1A,B). The diploid MMC subsequently initiates meiosis to produce four haploid megaspores (Fig. 1C). The most proximal of these, termed the functional megaspore, initiates megagametogenesis and undergoes three rounds of syncytial mitosis, eventually giving rise to the mature female gametophyte (Fig. 1A).

Transcriptome profiling in Arabidopsis has identified pathways acting in the megaspore mother cell (Schmidt et al., 2011) and in dividing or mature female gametophytes (Yu et al., 2005; Johnston et al., 2007; Jones-Rhoades et al., 2007). Little is known, however, about pathways involved in the transition from megasporogenesis to megagametogenesis. Mutations in the INNER NO OUTER and WUSCHEL (WUS) transcription factors induce somatic defects in ovule tissues that compromise early stages of gametophyte development, suggesting that inter-regional signalling is important (Elliott et al., 1996; Lieber et al., 2011). Mutations in components of small RNA (sRNA) silencing pathways also influence gametophyte initiation. Ovules lacking function of ARGONAUTE9 (AGO9), RNA-DEPENDENT RNA POLYMERASE6 (RDR6) or SUPPRESSOR OF GENE SILENCING3 produce multiple functional megaspore-like cells in pre-meiotic ovules that express the pFM1:GUS marker (Olmedo-Monfil et al., 2010). AGO9 protein binds 24 nt small interfering RNAs (siRNAs) and accumulates in nucellar epidermal cells, suggesting that it promotes a non-cell-autonomous signal that restricts gametophyte identity in sub-epidermal cells (Olmedo-Monfil et al., 2010).

In this study, Arabidopsis ovules were dissected by laser capture microdissection (LCM) and profiled to identify pathways involved in the transition from megasporogenesis to megagametogenesis. Analyses revealed genes upregulated in distal parts of the ovule, including the nucellus and megaspores, relative to other ovule tissues. Characterisation of a unique insertion allele for one of the genes, ARGONAUTE5 (AGO5), suggested that sRNA pathways acting in somatic nucellar cells promote the initiation of megagametogenesis in the functional megaspore. This was supported by tissue-specific expression of viral RNAi suppressor proteins, and was shown to be independent of AGO9. These results indicate that at least two somatic sRNA pathways contribute to gametophyte development in Arabidopsis. One pathway restricts reproductive potential to the functional megaspore and another promotes the initiation of megagametogenesis in this cell.

Plants were grown as described previously (Goetz et al., 2006). The ago1-27, ago5-1, ago5-2, ago5-4 (SALK_050483) (Alonso et al., 2003), ago9-2 and ago10-1 alleles are in the Columbia background, whereas rdr6-3 is in Landsberg erecta.

Flowers were fixed in 3:1 ethanol:acetic acid and embedded in butyl-methyl-methacrylate (Rasheed et al., 2006). Sections (5 μm) were placed on membrane slides (Leica), treated with acetone and dissected using an AS-LMD (Leica). NUC2 and OV2 samples were captured from the same slides in three independent replicate collections. RNA was extracted using a Picopure kit (Acturus) and amplified twice using MessageAmp III (Applied Biosystems), similar to previous methods (Schmidt et al., 2011). RNA was labelled using a ULS aRNA labelling kit (Kreatech) and hybridised to Agilent 4x44K arrays (Institute for Molecular Bioscience, Brisbane, Australia).

Data was analysed in R (http://www.r-project.org/) using bioconductor (http://www.bioconductor.org) packages Agi4x44PreProcess and limma (Smyth, 2004). For each probe, background median intensities were subtracted from mean probe signal and subtracted values less than 0.5 were set to 0.5. Values were normalised across samples using quantile normalisation and log2 transformed. Agilent quality flags were used to identify probes robustly detected in all replicates of at least one sample group. This requires the difference between probe signal and local background to be more than 1.5 times the local background noise with acceptable spot image morphology. Replicates showed good correlation as shown by comparison of the median coefficient of variation from all probes, which was 2.8% across all NUC2/OV2 samples decreasing to 1.9% within sample groups. In total, 20,732 probes passed quality filtering and were tested for differential expression using the limma package, allowing for the paired nature of the experimental design. Finally, 1725 probes, corresponding to 1524 gene identifiers showed evidence of differential expression (P≤0.01, FC≥2). Probes were annotated through alignment to TAIR10. Heatmaps were generated in R using the heatmap.2 function in the gplots package.

RNA was extracted from frozen tissue using the RNeasy Kit (Qiagen). Quantitative PCR was performed on Superscript III-generated cDNA (Invitrogen) and normalised using eIF4a. Primer sequences are listed in supplementary material Table S6.

Upstream promoter sequences from At2g43150, At5g54570, At1g21670, At3g05460, At5g53250, At3g21000, At5g40730, At2g27880 (AGO5), At5g14010 (KNU) and At2g17950 (WUS) were cloned with restriction adapters from Columbia genomic DNA into pCR4-TOPO. XFP genes including YFPer, nlsYFP, 3xnls-vYFP, YFP and nlsCFP were subcloned from ML877, ER030, ML873, HB042 and HB411, respectively, and kindly provided by Dr Michael Lenhard and Dr Holger Breuninger (University of Potsdam, Germany). Constructs were assembled in pBlueMLAPUCAP (ML939) (Adamski et al., 2009). Recipient pAGO5:3′AGO5 (MT361), pWUS:3′WUS (MT310) and pKNU:3′KNU (MT311) constructs were designed with a central BamHI site to integrate subcloned XFP, P1/HcPro-HA, P19-HA, AGO5, AGO5-4, YFP-AGO5-4 and YFP-AGO5 fragments as BamHI or BamHI/BglII inserts. All clones were introduced into the pBARMAP binary vector (Adamski et al., 2009) via AscI digestion. At least ten transgenic lines were examined for each construct.

Ovules were dissected from flowers and cleared in 0.25% chloral hydrate, 30% glycerol for phenotypic analysis or 10% glycerol for fluorescence analysis on a Zeiss M1 Imager as described previously (Tucker et al., 2008).

To identify genes involved in the transition from megasporogenesis to megagametogenesis, cells from the distal tip of developing ovules undergoing megaspore selection were captured using LCM (stage 2-V, Fig. 1C-E) (Schneitz et al., 1995). The laser system used could not be sufficiently focused to capture developing functional megaspores. However, by identifying genes highly expressed in the ‘nucellus’ region, comprising the nucellar epidermis, developing megaspores and flanking sub-epidermal nucellar cells (NUC2), relative to other ovule tissues (OV2), components of both cell-autonomous and non-cell-autonomous pathways promoting megagametogenesis might be identified. Three biological replicates of both captured regions were profiled and 1524 genes were differentially expressed between the two samples (FC≥2.0, P≤0.01; supplementary material Fig. S1A and Table S1). The NUC2 subset (563 genes) was enriched for transcriptional regulators (agriGO, supplementary material Table S2) (Su et al., 2010), including positive controls WUS (Fig. 1F-G) and SPOROCYTELESS (SPL) (Yang et al., 1999), and comprised diverse gene expression clusters indicative of multiple developmental functions (supplementary material Fig. S1B). Qualitative comparisons of the NUC2 subset were made to lists of genes downregulated in sporocyteless ovules lacking functional nucelli (Johnston et al., 2007), genes expressed in the integuments (Skinner and Gasser, 2009) and genes upregulated in a dividing female gametophyte (FG3-4) sample generated in this study (supplementary material Table S3). Approximately 26% of the NUC2 genes were present in the spl and/or FG3-4 datasets, whereas only 2% were detected in the integument dataset (Fig. 1F, supplementary material Table S4), indicative of minimal integument contamination. The high proportion (∼70%) of unique NUC2 genes possibly reflects the different annotation versions and array systems used to generate the datasets, but also the ability of the LCM method to enrich for RNAs from specific cell types in complex tissues.

To assess whether these transcriptome data were indicative of expression dynamics in planta, promoter:fluorophore fusions were generated for multiple NUC2 candidates and examined in transgenic plants (supplementary material Table S5). Diverse expression patterns were conferred by the promoter fragments in ovules, including combinations of expression in the nucellar epidermis, functional megaspore, inner integument and female gametophyte (Fig. 1G-P). In most cases, the expression dynamics fit the criteria of being upregulated in the distal tip of the ovule relative to other ovule tissues during the transition from megasporogenesis to megagametogenesis. The LCM approach utilised was therefore considered to be successful in defining a unique set of genes and developmental markers expressed at this stage.

Expression of At2g27880 (AGO5; Fig. 1K) in somatic cells surrounding the MMC and megaspores (supplementary material Figs S2, S3) was reminiscent of AGO9, which accumulates in nucellar epidermal cells and inhibits gametophyte development in sub-epidermal cells (Olmedo-Monfil et al., 2010). Conserved in plants and animals, AGO proteins bind sRNAs and induce repression of complementary RNA targets (Mallory and Vaucheret, 2010). T-DNA insertion lines were examined to investigate whether AGO5 might be involved in gametophyte formation. Two ago5 insertion alleles (supplementary material Fig. S4A), ago5-1 (Katiyar-Agarwal et al., 2007) and ago5-2 (Takeda et al., 2008), showed no defects in ovule or gametophyte development. A third line, SALK_050483, containing an insertion closer to the 5′ end of AGO5 (supplementary material Fig. S4A) showed defects in seed development and is hereafter referred to as ago5-4. Heterozygous ago5-4/+ plants produced siliques containing aborted ovules (51±6%), white seeds (1±1%) and green seeds (48±6%; Fig. 2A, n=563), compared with 95±5% green seeds in wild-type (WT) siblings (Fig. 2A, n=543). Homozygous ago5-4/– plants were not identified, possibly owing to defects in embryogenesis (supplementary material Fig. S4B).

Emasculated flowers from ago5-4/+ plants showed different ovule phenotypes at anthesis. Approximately 51% of the ovules contained a mature WT-like female gametophyte containing an egg and central cell nucleus (Fig. 2B,C). In the remainder, a one-nucleate female gametophyte (FG1) was observed (Fig. 2D), indicating that FG development had terminated prior to the first mitotic division. Although a 1:1 WT:mutant ratio is characteristic of gametophytic-effect mutations, reciprocal crosses between ago5-4/+ and WT suggested that ago5-4 is not a typical gametophytic mutation. Approximately 36% (n=150) of the F1 plants produced from ago5-4/+ females crossed with WT males inherited the mutation, indicating that the presence of ago5-4 in the female gametophyte only slightly diminishes fecundity.

Expression of marker genes in the megaspore mother cell, megaspores (pKNU:nlsYFP, pKNU:nlsGUS) (Payne et al., 2004) and functional megaspore (pFM1:GUS) (Acosta-Garcia and Vielle-Calzada, 2004) indicated that ovule development was normal in ago5-4/+ until FG1 abortion (Fig. 2E-J). By contrast, a marker for the mature female gametophyte (pMYB64:GFP) (Wang et al., 2010) was not detected in aborted ago5-4/+ ovules (Fig. 2K,L). This suggests that ago5-4/+ FG1 abortion is not caused by an obvious change in MMC or megaspore identity. Moreover, ago5-4/+ does not form extra megaspore-like cells, and thus is phenotypically distinct from ago9 mutants (Olmedo-Monfil et al., 2010).

Sequencing of the ago5-4 allele showed that the T-DNA is inserted immediately after the coding sequence for the sRNA-binding PAZ domain (supplementary material Fig. S4). Quantitative-PCR indicated that AGO5 mRNA levels are only slightly reduced (∼89%) during early stages of ago5-4/+ fruit development, with 20% of expression originating from the ago5-4 allele compared with <2% in ago5-1 and ago5-2 (Fig. 3A-C). To test whether the ago5-4/+ phenotype might be due to expression of transcript from the ago5-4 locus, the predicted AGO5-4 cDNA fragment was cloned, fused to the AGO5 promoter and transformed into plants. Multiple (5/10) transgenic lines containing pAGO5:AGO5-4 showed similar defects in ovule development to ago5-4/+ and aborted at the FG1 stage (Fig. 4A,B) with frequencies ranging from 27% to 74% (Fig. 3D).

Comparable results were obtained for a pAGO5:YFP-AGO5-4 gene, which induced FG1 abortion and accumulated YFP-AGO5-4 protein in the cytoplasm of nucellar epidermal and inner integument cells (Fig. 4O,P). Notably, YFP-AGO5-4 was absent from the megaspores, similar to YFP-AGO5 (Fig. 4M,N) and pAGO5:YFPer (supplementary material Figs S2, S3). This is consistent with the somatic effect of the ago5-4/+ mutation and suggests that the ago5-4 phenotype is unlikely to result from movement of AGO5-4 into the MMC or megaspores. Supporting this, restricted expression of pWUS:AGO5-4 in somatic nucellar epidermal cells (Fig. 4E) induced FG1 abortion in 25-51% of ovules (Fig. 3D, Fig. 4F), and increased the frequency of FG1 abortion in ago5-4/+ to 63±5% (n=460), indicative of a dose-dependent effect on a somatic target. By contrast, expression of AGO5-4 in the MMC and megaspores via the KNUCKLES (KNU) promoter (Fig. 4I) did not alter gametophyte abortion in WT or ago5-4/+ plants (Fig. 3D, Fig. 4J).

To address whether ago5-4/+ competes with residual AGO5 activity, extra copies of AGO5 were transformed into ago5-4/+ plants. However, this failed to alleviate ago5-4/+ defects (Fig. 3D). Moreover, removal of WT AGO5 protein in trans-heterozygous ago5-4/ago5-1 and ago5-4/ago5-2 plants failed to alter the frequency of FG1 abortion (Fig. 3D). Similar results were obtained for double mutants with ago1-27 and ago10-1 (Fig. 3D), which encode ovule-expressed AGO5 homologues with affinity for microRNAs (Mallory et al., 2009).

Collectively, these results suggest that the ago5-4 insertion generates a semi-dominant AGO5 gene that acts in somatic cells to inhibit pathways required for the initiation of megagametogenesis. Although a complex level of redundancy cannot be discounted, the finding that ago5 knockouts have no obvious effect on FG development in WT or ago5-4 background suggests that the unique ago5-4 insertion compromises pathway(s) not normally targeted by AGO5. One possibility is that ago5-4 generates a truncated, tissue-specific AGO5 protein with altered sRNA binding efficiency, which induces de-repression of complementary sRNA-targets and disrupts a novel pathway required for the transition to megagametogenesis.

The predicted structure of AGO5-4, retaining the AGO5 PAZ domain but lacking the MID-domain and catalytic PIWI-domain, is reminiscent of viral suppressor proteins, such as P1/Hc-Pro and P19, that bind sRNAs and inhibit AGO-dependent pathways (Kasschau et al., 2003; Dunoyer et al., 2004). Although both proteins have a negative effect on miRNA function, P19 also inhibits function of 21 nt and 22 nt siRNAs (Voinnet et al., 2003), whereas P1/Hc-Pro inhibits function of 21 nt and 24 nt siRNAs (reviewed by Li and Ding, 2006).

To examine whether viral suppressor proteins can impair ovule development similarly to ago5-4/+, the P1/Hc-Pro polyprotein sequence from Turnip Mosaic Virus and P19 sequence from Tomato Bushy Stunt Virus were expressed from the tissue-specific AGO5, WUS, KNU (Fig. 4) and constitutive 35S promoters. Transgenic plants expressing p35S:P1/Hc-Pro or p35S:P19 showed vegetative defects similar to previous reports (Kasschau et al., 2003; Dunoyer et al., 2004). The other lines appeared indistinguishable from WT, but an examination of ovule development showed that the pAGO5:P1/Hc-Pro and pWUS:P1/Hc-Pro genes induced FG1 abortion similar to ago5-4/+ (Fig. 3D, Fig. 4C,G). By contrast, the pAGO5:P19, pWUS:P19 and reproductive cell-specific pKNU:P1/Hc-Pro and pKNU:P19 suppressor constructs had no effect on ovule or gametophyte development (Fig. 3D, Fig. 4D,H,K,L). These results suggest that specific, somatic sRNA pathways inhibited by P1/Hc-Pro but not P19, are required for initiation of megagametogenesis.

Based on their structural and phenotypic similarities, it is likely that P1/Hc-Pro and AGO5-4 inhibit similar pathways. Consistent with the inhibitory activity of P1/Hc-Pro, AGO5 predominantly binds 24 nt siRNAs and some 21 nt and 22 nt sRNAs (Mi et al., 2008; Takeda et al., 2008). Interestingly, AGO9 also acts through 24 nt siRNAs, but preferentially binds those with a 5′ terminal adenosine (Havecker et al., 2010) compared with the 5′ terminal cytosine preferred by AGO5 (Mi et al., 2008). Neither ago9 nor rdr6 mutations altered the frequency of FG1 abortion in ago5-4/+ plants, which might be expected if they competed for targets or acted in the same pathway (Fig. 3D). Furthermore, the frequency of ago9 ovules containing extra megaspore-like cells was not altered by ago5-4/+, although their subsequent development was inhibited irrespective of their origin (supplementary material Fig. S5).

Collectively, these data suggest that at least two sRNA pathways act in somatic cells flanking the MMC and megaspores to regulate female reproductive development in Arabidopsis (Fig. 4Q). One pathway depends on AGO9 and 24 nt siRNAs to prevent sub-epidermal cells from adopting megaspore-like identity (Olmedo-Monfil et al., 2010). A second, independent sRNA pathway is required to promote the transition to megagametogenesis in the functional megaspore. This pathway is inhibited by the semi-dominant ago5-4 mutation and P1/Hc-Pro polyprotein and might also act through 24 nt siRNAs. Notably, silencing pathways inhibited by P19, P1/Hc-Pro and ago5-4 either do not act or have limited cell-autonomous functionality in the early reproductive cells examined here. This highlights the key role played by somatic ovule cells in regulating the events of early female gametophyte development.

We thank Jim Carrington, Jean-Philippe Vielle-Calzada, Yuichiro Watanabe, Gary Drews, Olivier Voinnet, Hailing Jin, Adam Wells, Gwen Mayo, Paul Boss and the AMK laboratory for materials and assistance.