Redundant and Specific Roles of the ARGONAUTE Proteins AGO1 and ZLL in Development and Small RNA-Directed Gene Silencing

The Arabidopsis ARGONAUTE1 (AGO1) and ZWILLE/PINHEAD/AGO10 (ZLL) proteins act in the miRNA and siRNA pathways and are essential for multiple processes in development. Here, we analyze what determines common and specific function of both proteins. Analysis of ago1 mutants with partially compromised AGO1 activity revealed that loss of ZLL function re-establishes both siRNA and miRNA pathways for a subset of AGO1 target genes. Loss of ZLL function in ago1 mutants led to increased AGO1 protein levels, whereas AGO1 mRNA levels were unchanged, implicating ZLL as a negative regulator of AGO1 at the protein level. Since ZLL, unlike AGO1, is not subjected to small RNA-mediated repression itself, this cross regulation has the potential to adjust RNA silencing activity independent of feedback dynamics. Although AGO1 is expressed in a broader pattern than ZLL, expression of AGO1 from the ZLL promoter restored transgene PTGS and most developmental defects of ago1, whereas ZLL rescued only a few AGO1 functions when expressed from the AGO1 promoter, suggesting that the specific functions of AGO1 and ZLL are mainly determined by their protein sequence. Protein domain swapping experiments revealed that the PAZ domain, which in AGO1 is involved in binding small RNAs, is interchangeable between both proteins, suggesting that this common small RNA-binding domain contributes to redundant functions. By contrast, the conserved MID and PIWI domains, which are involved in 5′-end small RNA selectivity and mRNA cleavage, and the non-conserved N-terminal domain, to which no function has been assigned, provide specificity to AGO1 and ZLL protein function.


Introduction
Small RNA-directed gene regulation is a major process in plant development and viral defense [1,2]. A central component in these pathways is the activity of ARGONAUTE (AGO) proteins, which bind small RNAs and mediate repression of the complementary RNA targets [3,4]. In Arabidopsis, 10 AGO genes have been identified [5]. AGO1 [6] associates with numerous microRNAs (miRNAs) and short interfering RNAs (siRNAs) to mediate target repression via mRNA cleavage and inhibition of translation [3,4,7]. Binding of AGO1 to miR168, which targets AGO1 mRNA, establishes a homeostatic AGO1 regulatory loop [8,9]. AGO4 and AGO6 function in small RNA mediated chromatin regulation whereas AGO7 associates specifically with miR390 and directs cleavage of the non-protein coding TAS3 precursor RNA to generate trans-acting short interfering RNAs (tasiRNAs) [5]. Recently, ZLL was implicated in miRNA-directed translational inhibition [7] and repression of miR165/166 levels [10].
AGO1 and ZLL protein sequences are highly similar, including the PAZ and MID domains, which bind small RNAs in AGO1 [11], and the PIWI domain, which is required for target mRNA cleavage in AGO1 [3,4]. By contrast, their N-terminal domains do not display sequence similarities. Both genes differ in their expression patterns and developmental functions. AGO1 is expressed broadly during plant development, and ago1 loss-offunction mutants display pleiotropic defects in development and in virus defense [6,12]. Seedlings of the null allele ago1-1 form only a few finger-like leaves and about 10% of seedlings lack a shoot meristem. ago1 mutants are deficient in transgene posttranscriptional gene silencing (PTGS) of L1 35S:GUS, a standard reference for systemic sense transgene PTGS in Arabidopsis [13], the tasiRNA pathway, and cell autonomous miRNA-directed repression [5]. In contrast to ago1-1, the hypomorphic allele ago1-27, which expresses an AGO1 protein with reduced mRNA cleavage activity, displays more subtle developmental defects [12].
Expression of ZLL is limited to the provasculature and, weaker, to the adaxial (upper) sides of leaves, and ceases as tissue differentiation takes place [14,15]. In the Landsberg erecta (Ler) accession, zll mutant seedlings display differentiated cells or complete organs in place of the shoot meristem stem cells with allele specific penetrance [14][15][16]. Recent studies indicate that ZLL function in the provasculature is necessary and sufficient to maintain shoot meristem stem cells during embryogenesis [17]. Furthermore, ZLL acts in a sequential manner with AGO1 during embryogenesis to potentiate WUSCHEL (WUS) dependent signaling from the stem cell organizer to the stem cells in the developing shoot meristem primordium [17]. ago1 zll double mutants of strong alleles result in early embryo arrest, suggesting that both proteins also have redundant activities during early embryo development [14]. Recent findings demonstrated that both proteins function in miRNA-directed repression of Cu/Zn SUPEROXIDE DISMUTASE 2 (CSD2) and SCARECROW-LIKE 6 (SCL6-IV) mRNAs and proteins [7]. In contrast to ago1 mutants, however, L1 transgene PTGS is not compromised in zll mutants [12].
Here, we address specific and overlapping functions of ZLL and AGO1 in development and RNA silencing pathways. Our results indicate that in ago1 hypomorphic mutants, loss of ZLL function restores leaf development and siRNA and miRNA pathways and leads to increased AGO1 protein levels, implicating ZLL as a negative regulator of AGO1. Analyses of chimeric gene constructs indicate that the PAZ domain, which is thought to mediate small RNA binding, is exchangeable between both proteins, whereas the MID-PIWI and N-terminal domains appear to contribute to their specific functions.

zll suppresses leaf defects of ago1 hypomorphs
To study genetic interactions between ZLL and AGO1, we analyzed different mutant combinations. Since double mutants of strong zll and ago1 alleles in the Ler ecotype are embryo lethal [14], we analyzed mutant alleles in the Col ecotype, where ZLL loss of function alone does not greatly affect development ( Figure 1A and Figure S1), unlike in the Ler accession, where shoot meristem stem cells are defective [14,15]. Despite the reduced effect of zll mutations in Col compared with Ler, ago1-1 zll ago10-1 double mutant embryos also arrested at the late globular stage with defects in cell division, cell elongation, and expression of both WOX5 and WUS genes, which mark root and shoot stem cell niches, respectively ( Figure S2). None of these effects were observed in any single mutant, indicating redundant functions of ZLL and AGO1. To avoid embryo lethality obtained in double mutants with the null allele ago1-1 [6] and to enable the analysis of genetic interactions during postembryonic development, we used the hypomorphic ago1-27 mutant in combinations with zll ago10-1 and zll ago10-3 alleles. ago1-27 mutants are defective in small RNAdirected regulation [9,12] and, in contrast to the severe growth and developmental defects of ago1-1, display increased leaf margin serration, reduced leaf width, abnormal flower phyllotaxis, and reduced fertility compared to wildtype [12]. By contrast, seedlings of zll ago10-1 and zll ago10-3 single mutants did not display any noticeable leaf defects ( Figure 1 and Figure S1A) [18] and only infrequently a defective shoot meristem (0.2%, n.1000) [14,15]. Surprisingly, ago1-27 zll ago10-1 and ago1-27 zll ago10-3 double mutants revealed that both zll mutations partially suppressed the increased leaf margin serration of ago1-27 ( Figure 1B and Figure S1A), rather than enhancing it as we expected for two related AGO proteins involved in RNA silencing. By contrast, neither the phyllotaxis nor the fertility defects of ago1-27 were restored by the zll mutations (data not shown).
zll mutations restore transgene PTGS and miRNAmediated gene silencing in hypomorphic ago1 mutants To study ZLL and AGO1 interactions at the level of RNA silencing, we first analyzed PTGS of the L1 35S:GUS transgene. Our previous studies indicated that PTGS of the L1 35S:GUS transgene was compromised in sgs3, rdr6, hen1, and ago1 mutants but not in zll single mutants [12,19,20]. The newly identified ago1-40 EMS mutation causes an A to V amino acid change at position 863 of the protein, resulting in increased mRNA levels and protein activity and decreased siRNA levels for the L1 35S:GUS transgene ( Figure 2 and Table S1). Unlike previously identified ago1 mutations that impair L1 PTGS with 100% efficiency, about 50% of ago1-40 adult plants at each generation had triggered PTGS, allowing us to test whether zll mutations affected L1 PTGS in ago1-40. To avoid any potential interference between the 35S promoters embedded in the T-DNA of the available insertional zll mutants in Col and the L1 35S:GUS transgene [21], we backcrossed five times to L1 the EMS-induced zll-3 mutant, which was isolated in the Ler accession [15]. L1/zll-3 Col had similar GUS mRNA levels, protein activity and siRNA levels as silenced L1 controls ( Figure 2) [12].
GUS mRNA levels in L1/ago1-40 zll-3 Col double mutants were reduced in comparison to L1/ago1-40 mutants to nearly the level of silenced L1 controls ( Figure 2). This increase in L1 silencing in the double mutant correlated with increased levels of GUS siRNAs. Seven days after germination (DAG), GUS siRNA levels were more than 10-fold higher than in L1/ago1-40 mutants, reaching levels comparable to silenced L1 controls 7 DAG, and by 15 DAG even exceeding L1 control levels ( Figure 2). Thus, loss of ZLL function restored L1 gene silencing compromised in ago1-40.
To address whether zll mutations also were able to restore the miRNA pathway in ago1 hypomorphs, we analyzed miRNA levels and miRNA-regulated target genes in the ago1-27 zll ago10 double

Author Summary
In eukaryotes, short RNAs (21-24 nucleotides long) have broad effects on gene expression through the action of ARGONAUTE (AGO) proteins. The model flowering plant Arabidopsis thaliana contains ten AGO proteins, among which AGO1 and ZLL/PNH/AGO10 play a major role in regulating gene expression through small RNA-directed RNA cleavage and translational repression. Here, we address the common and specific effects of zll and ago1 loss of function in Arabidopsis. We show that zll mutations lead to increased AGO1 protein levels and suppress a subset of small RNA-directed gene regulatory defects of weak ago1 mutations. Although AGO1 and ZLL proteins are highly similar in sequence, we show that only the PAZ domain, which in AGO1 is involved in binding small RNAs, can be exchanged between the two proteins. By contrast, the PIWI domain, that is responsible for the RNA cleaving activity of AGO1, the MID domain, which is involved in 59 nucleotide selection of small RNAs, and the functionally uncharacterized N-terminal domain contribute to their individual functions during small RNA-directed gene regulation and development.
mutants. miR398 levels were reduced and CSD2 mRNA and protein levels were substantially elevated in ago1-27 compared to zll ago10-3 and zll ago10-1 single mutants and wildtype ( Figure 3A and Figure S3). By contrast, in both ago1-27 zll ago10-3 and ago1-27 zll ago10-1 double mutants, CSD2 mRNA and protein levels were reduced and miR398 levels were elevated, compared to ago1-27 alone ( Figure 3A and Figure S3). The zll ago10-3 mutation also restored miR164 accumulation and miR164-directed CUC2 silencing to wildtype levels in the ago1-27 background ( Figure 3B). To extend our investigation to the whole-genome level, a transcriptome analysis was performed using Col wildtype, ago1-27, zll ago10-1 and ago1-27 zll ago10-1 . Among 46 miRNA targets that were elevated in ago1-27 compared to wildtype but which were not affected in zll ago10-1 single mutants, 19 were reduced completely or partially to wildtype levels in the ago1-27 zll ago10-1 double mutant (Table S2). Taken together, loss of ZLL function restored L1 PTGS and silencing of approximately half of the miRNA targets deregulated in ago1- 27. zll mutations enhance AGO1 protein accumulation in hypomorphic ago1 mutants The suppression of developmental, L1 silencing and miRNA pathway defects in hypomorphic ago1 mutants by zll mutations raised the question whether ZLL might be a negative regulator of AGO1. To test this hypothesis, we compared AGO1 mRNA and protein levels in ago1, zll and ago1zll double mutants. AGO1 protein levels were increased in both ago1-27 zll ago10-3 and ago1-40 zll-3 Col double mutants compared to the corresponding ago1 single mutants ( Figure 4 and Figure S4). AGO1 mRNA and miR168 levels, however, were not significantly different ( Figure 4). This indicates that ZLL is a negative regulator of AGO1 at the protein level, consistent with the role of ZLL in translational inhibition [7]. Protein sequence and specific expression patterns determine the functional differences between AGO1 and ZLL To determine whether the specific effects of ago1 and zll mutations could be explained by the expression patterns of AGO1 and ZLL, we first compared the expression patterns of pZLL:YFP-ZLL and pAGO1:CFP-AGO1 reporter genes. Both reporter constructs rescued the corresponding mutants, indicating that the fusion proteins are functional (Table S3) [17]. YFP-ZLL and CFP-AGO1 proteins were detected in a largely overlapping punctuate pattern outside the nucleus of expressing cells ( Figure 5E-5J). As previously reported, pZLL:YFP-ZLL is initially expressed throughout the embryo, but becomes limited to provascular strands and the adaxial side of the cotyledons at about the globular stage ( Figure 5A) [17]. By contrast, pAGO1:CFP-AGO1 is expressed in the whole embryo with the strongest signal in the provascular cells from globular stage to early torpedo stage ( Figure 5C). Thus, ZLL and AGO1 expression patterns overlap partially, with the AGO1 expression pattern being broader than the one of ZLL, in agreement with mRNA localization results [14].
To evaluate the significance of the broad AGO1 expression pattern, we expressed AGO1 from the ZLL promoter and found that pZLL:AGO1 by and large restored development of ago1-1 (Table 1 and Figure S5) and ago1-27 (data not shown) mutants and also L1 PTGS in ago1-27 ( Figure 6A-6B). However, miR398 accumulation and CSD2 silencing were only partially restored in ago1-27/pZLL:AGO1 ( Figure 6C). These results suggest that limiting expression of AGO1 to the ZLL region is sufficient to provide most AGO1 functions in development and RNA silencing. Nevertheless, expression in cells outside the ZLL pattern is required to completely restore AGO1 activity.
Next, we addressed whether differences within ZLL and AGO protein sequences are responsible for differences in their functions by analyzing whether AGO1 could replace ZLL and vice versa. AGO1 expression from the ZLL promoter (pZLL:AGO1) rescued shoot meristem formation in the zll-1 mutant in the majority of cases (Table 2). By contrast, expression of ZLL from the AGO1 promoter (pAGO1:ZLL) in the strong ago1-1 allele resulted only in a slight reduction of leaf radialization compared to untransformed ago1-1 ( Figure S6), but did not rescue any other developmental defect. Furthermore, in the ago1-27 hypomorph, pAGO1:ZLL was unable to rescue altered flowering time, reduced rosette size ( Figure S7), L1 PTGS ( Figure 6A and 6B) or CSD2 regulation ( Figure 6C). Thus, whereas AGO1 can largely replace ZLL function in stem cell development, ZLL appears unable to efficiently replace the developmental, miRNA and PTGS functions of AGO1. Intriguingly, although pAGO1:ZLL did not restore CSD2 silencing in ago1-27, it fully restored miR398 accumulation to wildtype levels ( Figure 6C). These results suggest that the intrinsic differences of AGO1 and ZLL proteins determine their specific contribution to small RNA and development pathways.

The PAZ domain, but not the ZLL MID-PIWI-or Nterminal domains, is exchangeable between ZLL and AGO1 proteins
To address whether and if any ZLL and AGO1 protein domains have similar functions, we analyzed the ability of chimeric proteins composed of AGO1 and ZLL domains to rescue the respective mutant defects. As expected from the pZLL:AGO1 result, most chimeric ZLL AGO1 proteins (where one AGO1 protein domain was embedded in a ZLL protein backbone) driven from the ZLL promoter rescued shoot meristem formation of the zll-1 mutant ( Table 2). The marked exception was the AGO1 N-terminal domain (pZLL:ZLL AGO1 N9 ) that could not efficiently replace the corresponding ZLL N-terminal domain ( Table 2). This finding was unexpected since the complete AGO1 protein largely replaced ZLL, and might indicate that the function of the N-terminal domain is sensitive to the correct protein context.
On the converse, only the ZLL PAZ domain within the AGO1 backbone (pAGO1:AGO1 ZLL PAZ ) efficiently rescued developmental defects not only of the ago1-27 hypomorph ( Figure 7L and Figure  S7) but also of the null ago1-1 allele ( Figure 7E, Table 1, and Figure  S5). The ZLL PAZ domain also largely restored L1 PTGS and GUS siRNA accumulation, and CSD2 silencing and miR398 accumulation in ago1-27 ( Figure 6). PTGS restoration, however, was delayed compared to the developmental rescue ( Figures 6A and 7), consistent with previous findings that PTGS is more sensitive than development to compromised AGO1 activity [12]. By contrast, replacing the N-terminal or MID-PIWI domains of AGO1 with the corresponding ZLL regions (pAGO1:AGO1 ZLL N9 and pAGO1:A-GO1 ZLL MID-PIWI ) only restored bilateral leaf development but not sterility of ago1-1 mutants ( Figure 7C, 7F, and 7G, Table 1, and Figure S5), or any developmental defects of ago1-27 mutants ( Figure 7M and 7O and Figure S7). In addition, neither the Nterminal domain nor the MID-PIWI domains of ZLL were able to restore L1 PTGS and GUS siRNA accumulation or CSD2 silencing in ago1-27 ( Figure 6). Since previous studies have indicated that PAZ, MID and PIWI domains function together in small RNA binding [11,22,23], we constructed a pAGO1:AGO1 ZLL PAZ-PIWI chimera where the AGO1 genomic region containing PAZ, MID and PIWI domains was replaced by the corresponding ZLL genomic sequence ( Figure S8). pAGO1:AGO1 ZLL PAZ-PIWI resulted in similar effects as pAGO1:AGO1 ZLL MID-PIWI (Table 1, Figures 6 and 7, and Figure S7). This suggested that the failure of the ZLL MID-PIWI domains to restore the majority of ago1 defects was not due to an incompatibility with the AGO1 PAZ domain or the disruption of the region connecting the PAZ and PIWI domains. Notably, although the pAGO1:AGO1 ZLL PAZ-PIWI did not rescue CSD2 silencing, it restored miR398 accumulation in ago1-27 ( Figure 6C).
In summary, these results indicate that the ZLL PAZ domain has the capacity to fulfill AGO1 functions in development, the miRNA pathway, and PTGS whereas the ZLL N-terminal and MID-PIWI domains are largely incompatible with AGO1 activity.

Discussion
As part of the small RNA-directed RNA silencing machinery, the closely related ZLL and AGO1 proteins fulfill important roles during Arabidopsis development. Previous studies of mutant phenotypes indicate the presence of both, redundant, specific, and even opposite functions of ZLL and AGO1. Here, we investigate the diversity of ZLL and AGO1 functions and show that ZLL acts as a negative regulator of AGO1, and that the activities of the two proteins are determined by both functionally equivalent and distinct domains.

Redundant functions of ZLL and AGO1
We find that double mutant combinations of strong zll and ago1 alleles are embryo lethal with strong patterning defects, revealed by abnormal expression of marker genes for the shoot and root meristem stem cell niche. This indicates that ZLL and AGO1 have a significant set of redundant functions required during early embryo development, in line with previous reports [14]. Although we have been unable to directly determine the small RNAs bound to ZLL due to the instability of the ZLL protein, we present several lines of indirect evidence suggesting that ZLL and AGO1 have partially redundant functions in small RNA-mediated silencing, and that ZLL domains are capable of binding a subset of small RNAs bound by AGO1: (1) Our protein domain swapping experiments indicate that the PAZ domain, which has been shown to bind small RNAs in several AGO proteins [11], is interchange-   able between ZLL and AGO1, providing fully active proteins, (2) miR398 accumulation is restored to wildtype levels in an ago1 hypomorph by expression of pAGO1:ZLL, pAGO1:AGO1 ZLL PAZ , pAGO1:AGO1 ZLL MID-PIWI and pAGO1:AGO1 ZLL PAZ-PIWI chimeras, and (3) both AGO1 and ZLL negatively regulate AGO1.

Opposing effects of ago1 and zll mutations
In addition to redundant functions of AGO1 and ZLL, our results using hypomorphic ago1 alleles to circumvent embryo lethality demonstrate opposing effects of ago1 and zll mutations. First, loss of ZLL function re-establishes both PTGS of the L1 transgene and miRNA-directed repression of a subset of target mRNAs deregulated in ago1-27, including miR398-and miR164 directed repression of their CSD2 and CUC2 targets, respectively. Furthermore, we observe partial suppression of hypomorphic ago1 leaf serration defects by zll mutations, which could be due to the partial re-establishment of miR164-directed CUC2 regulation in ago1 zll double mutants (Figures 1 and 3 and Figure S1) [24]. These opposite effects of ago1 and zll mutations are consistent with recent findings showing that mRNAs of leaf polarity-related HD-ZIP transcription factors and the corresponding miR165/166 are affected oppositely in zll and in ago1 single mutants (S. Bosca and T.L. unpublished) [9,10,25]. A plausible explanation for the restoration of developmental and RNA silencing defects caused by reduced AGO1 activity is provided by our finding that loss of ZLL activity results in upregulation of AGO1 protein levels in ago1-27. This negative regulation of AGO1 by ZLL suggests that homeostasis of AGO activity involves cross-regulation between different AGO proteins, which in the case of ZLL affects AGO1 protein but not mRNA levels, consistent with the recent implication of ZLL in translational repression [7]. Importantly, since ZLL expression itself is not a target of small RNA-mediated repression whereas AGO1 is [9,26], ZLL has the potential to provide an input into RNA silencing activity that is independent of negative feedback dynamics and thus might serve to mediate, for example, developmental tuning of RNA silencing.
However, silencing of all miRNA targets deregulated in ago1-27 is not restored by the absence of ZLL function. One possible explanation is that upregulation of AGO1 protein levels in ago1 zll double mutants does not restore AGO1 activity completely to wildtype levels, which might be required for efficient silencing of a subset of target genes. Alternatively, since the miRNA pathway is cell autonomous [27,28], the re-establishment of silencing of miRNA targets is expected to be limited to tissues where AGO1 and ZLL are co-expressed but will not take place in tissues where only AGO1 is expressed. This explanation is consistent with the pZLL:AGO1 analysis, where limiting AGO1 expression to the ZLL domain in ago1 mutants restored systemic L1 PTGS but did not fully restore miR398 accumulation and CSD2 regulation. Future experiments comparing AGO1, ZLL and miRNA tissue-specific expression will help to discriminate between these two possibilities.

Determinants of specific AGO1 and ZLL activities
Even though the sequences of ZLL and AGO1 proteins are closely related, the corresponding single mutants display different developmental defects. The pleiotropic ago1 mutants are defective in leaf morphology, general growth, and fertility, whereas zll mutants in the Ler accession display specific developmental defects in shoot apical meristem, flower, and silique development with allele specific penetrance. In contrast to the interchangeable PAZ domain, the non-conserved N-terminal domains, for which a function has yet to be assigned, cannot be exchanged between AGO1 and ZLL without loss of activity. Similarly, exchange of the MID and PIWI domains, which in AGO1 have been shown to provide selectivity for small RNAs possessing a 59 U [22] and to function as a slicer domain that cleaves mRNA, respectively [3,4], also cannot provide fully active proteins. This indicates that these domains contribute to functional differences. It is possible that the inability of the ZLL MID-PIWI fragment to replace the AGO1 domains reflects different preferences for 59 nucleotide selectivity. Since the consensus amino acid residues essential for mRNA cleavage in several AGO1 proteins [29] are present in the ZLL PIWI domain, it is conceivable that both AGO1 and ZLL have the capacity to silence via mRNA cleavage and translational inhibition, but that each protein has a different preference for one of the two mechanisms, in line with recent findings [7].
Future dissection of AGO1 and ZLL properties will help to reveal how the interplay between AGO1 and ZLL proteins influences silencing specificity and efficiency in development.

RNA analysis and GUS activity quantification
For RNA gel blot analyses, frozen tissue was homogenized in a buffer containing 0.1 M NaCl, 2% SDS, 50 mM Tris-Hcl (pH 9), 10 mM EDTA (pH 8) and 20 mM beta mercaptoethanol and RNAs were extracted two times with phenol. RNA gel blot analyses and quantification of GUS activity were performed as described [31]. Hybridization signals were quantified using a Fuji phosphor imager and normalized to a U6 oligonucleotide probe   for miRNA gel blot analyses. GUS mRNA and GUS activity analyses were performed on the aerial parts of 7-day-, 15-day-and 21-day-old seedlings grown on Bouturage media (Duchefa) in 16 hours light, 8 hours dark at 22uC. For the CSD2 and miR398 analyses, seeds were germinated on media [32] without sucrose in both the presence and absence of 0.5 mM CuSO 4 , and plants were grown in 16 hours light, 8 hours dark at 22uC for 12 days at which time the aerial portion of the seedlings were harvested and homogenized in liquid nitrogen. For the CUC2, miR164, AGO1 and miR168 analyses, plants were grown for 10 days on media [32] in the presence of 0.5 mM CuSO 4 . For cDNA synthesis, RNAs were extracted with the Plant RNeasy kit (Qiagen), treated with DNAseI (Invitrogen) and l mg of DNA-free RNA was reverse transcribed with oligo-dT (Invitrogen). Quantitative real time (QRT)-PCR, was performed on a MasterCycler ep realplex (Eppendorf) with the RealMAster SYBR ROX mix (5PRIME) according to the manufacturer's protocol. Each reaction was performed on 5 ml of 1:60 dilution of the cDNA and synthesized in a 20 ml total reaction. Specific oligonucleotide pairs were: EF1a: 59-CTGGAGGTTTTGAG GCTGGTAT -39, 59-CCAAGGGTGAAAGCAAGAAGA -39;  CSD2: 59-CAGAAGATGAGTGCC GTCATGCGG -39, 59-CCGAGGTCATCCTTAAGCTCGTG -39; CUC2: 59-GCA CCAACACAACCGTCACAG -39, 59-GAATGAGTTAACGTCTAAGCCCAAGG-39 and AGO1: 59-AAGGAGGTCGAGGAGGGTATG -39, 59-CAAATTGCTGAGCCAGAACAG -39. The reactions were incubated at 95uC for 2 minutes to activate the hot-start recombinant Taq DNA polymerase, followed by 45 cycles of 15 seconds (s) at 95uC, 15 s at 60uC and 20 s at 68uC to ensure primer extension and to measure the fluorescence signal. The specificity of the PCR amplification procedures was checked with a heat dissociation protocol (from 60uC to 95uC) after the final cycle of PCR. The efficiencies of the primer sets were evaluated by performing QRT-PCR on several dilutions of a mix of the different strands. The results obtained on the different genotypes were standardized to the expression level of EF1a. For microarray analyses, RNAs were extracted using the RNeasy Plant Mini Kit (Qiagen), labelled according to the manufacturer's instructions using the Quick-Amp One-Color Labelling Kit (Agilent Technologies) and hybridized to Agilent custom microarrays. Three replicates were performed for each genotype.

Microscopy and image analysis
For fluorescence studies, embryos where dissected from ovules using fine tip syringes in 10% glycerol, mounted on slides and analyzed using an AxioImager microscope (Zeiss) with YFP or CFP filter sets. Images were taken using Axiovision 4.4 software (Zeiss) and figures were generated using Photoshop 7.0 (Adobe). For confocal pictures, a Leica TCS SP2 AOBS spectral confocal microscope was used. Embryos were stained with DAPI (1 mg/ml) for 5 minutes and mounted in 50% glycerol in 16PBS.

Construction of fluorescent protein genes and chimeric genes
All AGO1 and ZLL sequences for both the fluorescent protein fusion and chimeric constructs are derived from the Col accession. AGO1 and ZLL chimeric constructs were made by exchanging five genomic domains; the 59 sequence upstream of the ATG, the Nterminal, PAZ and the MID-PIWI domains and the 39 region downstream of the stop codon. For cloning, restriction sites were introduced within introns at the appropriate positions (Table S4 and Figure S8). During the course of this work, we re-sequenced the ZLL Ler gene and several new ZLL cDNA clones and found that the original report of six amino acid differences between the ZLL Col and ZLL Ler proteins [15] was in error. The ZLL Ler amino acid sequence is identical to that of ZLL in Col, as previously published [14].