|Home | About | Journals | Submit | Contact Us | Français|
Argonaute (AGO) proteins recruit small RNAs to form the core of RNAi effector complexes. Arabidopsis encodes ten AGO proteins and a large network of small RNAs. How these small RNAs are sorted into specific AGO complexes remains largely unknown. We have cataloged small RNAs resident in four AGO complexes. We found that AGO2 and AGO4 preferentially recruit small RNAs with a 5′ terminal adenosine, whereas AGO1 harbors microRNAs (miRNAs) that favor a 5′ terminal uridine. AGO5 predominantly binds small RNAs that initiate with cytosine. Changing the 5′ terminal nucleotide of an miRNA predictably redirected it into a different AGO complex and alters its biological activity. These results reveal a role for small RNA sequences in assorting among AGO complexes. This suggests that specialization of AGO complexes might involve remodeling the 5′ end-binding pocket to accept certain small RNA sequences, perhaps explaining the evolutionary drive for miRNAs to initiate with uridine.
RNA interference (RNAi) is a conserved mechanism that regulates gene expression at either the transcriptional or posttranscriptional level in eukaryotes. The RNAi pathway is triggered by exogenous or endogenous, perfect or near-perfect double-stranded RNAs, which are processed by Dicer enzymes into ~21–25 nt small RNAs (Hannon, 2002). Arabidopsis contains a complex system of small RNAs including miRNAs and several classes of endogenous siRNAs (Vaucheret, 2006). miRNAs are derived from primary transcripts that adopt characteristic stem-loop structures. The stem-loop structures are processed by one of the four Dicers, DCL1, into a duplex formed by miRNA and miRNA star (miRNA*, miRNA’s partner strand that arises from the opposite arm in the precursor). The miRNA strand is more efficiently incorporated into effector complex, where it directs the cleavage of its target mRNA (Jones-Rhoades et al., 2006).
Based upon their origins and functions, Arabidopsis endogenous siRNAs can be divided into three classes: trans-acting siRNAs (tasiRNAs), natural antisense transcript-derived siRNAs (natsiRNAs), and repeat-associated siRNAs (rasiRNAs). tasiRNA biogenesis is initiated by miRNA-directed cleavage of noncoding transcripts. Such cleavage products then serve as substrates for RNA-dependent RNA polymerase 6 (RDR6) to generate dsRNAs, which are in turn processed by DCL4 into ~21 nt tasiRNAs (Allen et al., 2005; Axtell et al., 2006; Peragine et al., 2004; Vazquez et al., 2004; Xie et al., 2005; Yoshikawa et al., 2005). A subset of tasiRNAs can regulate the expression of some members of the ARF gene family to control the vegetative phase transition during Arabidopsis development (Adenot et al., 2006; Fahlgren et al., 2006; Garcia et al., 2006; Hunter et al., 2006). natsiRNAs are derived from natural antisense transcripts. dsRNAs formed by convergent transcription of two genes are processed by DCL2 into 24 nt natsiRNAs, which direct the further production of 21 nt natsiRNAs in a DCL1-dependent fashion (Borsani et al., 2005). natsiRNAs are involved in cellular responses to salt stress and pathogen attack (Borsani et al., 2005; Katiyar-Agarwal et al., 2006). There is also an abundant class of ~24 nt rasiRNAs derived from transposons and repetitive elements. The biogenesis of rasiRNAs depends on the activities of DCL3, RDR2, and a plant-specific DNA-dependent RNA polymerase, PolIVa. rasiRNAs are involved in the methylation and silencing of many transposons and repeats (Henderson and Jacobsen, 2007; Zaratiegui et al., 2007).
Small RNAs associate with Argonaute (AGO) proteins to form effector complexes, collectively termed RNA-induced silencing complexes (RISCs). RISCs are guided by small RNAs to their targets (RNA or chromatin) based on sequence complementarity, resulting in target mRNA cleavage, translational repression, or chromatin modification (Hannon, 2002). AGO proteins have three characteristic domains, PAZ, Mid, and PIWI (Song and Joshua-Tor, 2006). Arabidopsis encodes 10 AGOs that can be phylogenetically divided into four clades (Morel et al., 2002; Zheng et al., 2007). As the founding member of Argonaute family (Bohmert et al., 1998, AGO1 predominates in the miRNA pathway and is also involved in posttranscriptional gene silencing (PTGS) of transgenes (Baumberger and Baulcombe, 2005; Morel et al., 2002; Qi et al., 2005; Vaucheret et al., 2004). Within the same clade with AGO1, AGO10 (PINHEAD/ZWILLE) plays a critical role in maintaining the undifferentiated state of stem cells in the developing shoot meristem (Lynn et al., 1999; Moussian et al., 1998). AGO4 and AGO6 fall into the same clade, playing redundant roles in rasiRNA accumulation and in controlling DNA methylation and transcriptional gene silencing at specific genomic loci including transposons and repeats (Qi et al., 2006; Zheng et al., 2007; Zilberman et al., 2003). AGO7 (ZIPPY) forms a distinct clade and has been implicated in regulation of developmental timing by tasiRNAs (Adenot et al., 2006; Fahlgren et al., 2006; Garcia et al., 2006; Hunter et al., 2006). The biological functions of the other AGO proteins remain to be elucidated.
The complexity of small RNA populations and the diversity of the AGO protein family in Arabidopsis raise important questions as to mechanisms by which small RNAs are specifically sorted into distinct AGOs. We cataloged small RNAs associated with four AGOs that belong to three clades by deep sequencing. We found that each AGO recruits a specific subset of small RNAs. The most obvious distinguishing feature of the RNAs that join each AGO is their 5′ terminal nucleotide. We demonstrate that distribution of small RNAs among different AGOs can be manipulated both in vitro and in vivo by alteration of their first nucleotides. Our findings reveal a mechanism through which Arabidopsis small RNAs are sorted into distinct AGO complexes, which has important implications for the functions of Arabidopsis miRNAs and siRNAs.
To examine whether Arabidopsis AGO complexes contain distinct subsets of small RNAs, we purified AGO1, AGO2, AGO4, and AGO5, representing three AGO clades (Zheng et al., 2007). Purification of AGO1 was carried out as previously described (Qi et al., 2005). The AGO4 complex was immunopurified from a transgenic Arabidopsis line expressing myc-tagged AGO4 under its native promoter (Qi et al., 2006). AGO2 and AGO5 complexes were immunopurified using specific peptide antibodies. As shown on silver-stained gels (Figures 1 and S1), AGO proteins of expected sizes were recovered from wild-type but not from mutant plants, and their identities were further verified by mass spectrometry (Figure S2).
RNAs were extracted from the purified AGO complexes, resolved on denaturing polyacrymide gels, and visualized by SYBR-gold staining. Consistent with our previous findings (Qi et al., 2006), AGO1- and AGO4-associated small RNAs were predominantly 21 and 24 nt in length, respectively (Figure 1). AGO2 was associated predominantly with 21 nt small RNAs, and AGO5 bound all three known size classes (21, 22, and 24 nt) of small RNAs (Figure 1).
To catalog the small RNAs present in each AGO complex, gel-purified small RNAs were sequentially ligated to RNA adaptors at their 5′ and 3′ ends. The ligation products were used to generate cDNAs for Illumina 1G sequencing. After removal of adaptor sequences from the 35 bp reads, the trimmed sequences were compared to Arabidopsis nuclear and organellar genomes by BLAST. In total, 1683581, 136662, 711517, and 309010 genome-matched small RNA reads were obtained for AGO1, AGO2, AGO4, and AGO5 complexes, respectively (Table S1). These reads represent 56696 (AGO1), 66248 (AGO2), 322086 (AGO4), and 154418 (AGO5) unique small RNA sequences (Table S1). Consistent with SYBR-gold staining results, the majority of small RNAs associated with AGO1 and AGO2 were 21 nt in length, those associated with AGO4 were mostly 24 nt, and those associated with AGO5 were 21 nt (23%), 22 nt (28%), or 24 nt (30%) in length (Figure 2A). A relatively large fraction of the small RNAs were sequenced only once (57%, 84%, 73%, and 73% of the AGO1, AGO2, AGO4, and AGO5 libraries, respectively), indicating that a highly complex small RNA population associated with each Arabidopsis AGO complex.
Small RNAs were mapped to the Arabidopsis genome and categorized based on their genomic locations and functions (Figure 2B; Table S1). While all categories of small RNAs bound to some degree to each AGO, individual AGO complexes demonstrated a distinct preference for small RNAs of specific types. For comparison, we performed the same analysis with a previously published small RNA data set generated from Arabidopsis total RNA (referred to as Total hereafter) (Rajagopalan et al., 2006).
Consistent with the established role of AGO1 in the miRNA pathway, 82% of the AGO1-bound small RNAs were annotated miRNAs. Our AGO1 data set covered 67 of the known 86 miRNA families (Table S2). miR413-420 and 426 were not identified among this data set, supporting the notion that these small RNAs are not bona fide miRNAs (Jones-Rhoades et al., 2006). In AGO1, we identified a substantial number of small RNAs derived from various locations in the antisense strands of the annotated precursors of miR401, 405, 407, 783, 854, and 855 (data not shown), which suggests that these small RNAs might be produced from perfect dsRNAs rather than from stem-loop structures. Intriguingly, miR390 is underrepresented in the AGO1 complex, suggesting that miR390 may be preferentially associated with other AGOs. Indeed, we found a large number of miR390 reads in AGO2 (Table S2). In contrast to the prevalence of miRNA*s in the Total data set (~7% of the miRNA reads), significantly fewer miRNA* reads (~0.1% of the miRNA reads) were present in the AGO1 complex.
Importantly, 13% of small RNAs in AGO1 complex are derived from annotated repeats (Figure 2B). Compared to AGO4-associated rasiRNAs, AGO1-associated rasiRNAs appear to be more enriched for inverted repeats (Table S1). Thus, in addition to its role in the miRNA pathway, AGO1 might also participate in repeat silencing.
In accord with our previous findings (Qi et al., 2006), 57% of the small RNAs in AGO4 matched repetitive sequences. Of the remainder, 32% of small RNAs were from unannotated intergenic regions and 9% originated from protein coding genes. Compared to the distribution of each small RNA category in the Total data set, the AGO2 reads were enriched for tasiRNAs, whereas AGO5 preferentially bound to small RNAs derived from unannotated intergenic sequences (Figure 2B; Table S1). Genome-wide analyses on AGO2- and AGO5-associated small RNAs suggest that they are essentially evenly distributed along the five Arabidopsis chromosomes, with a weak enrichment in the pericentromeric regions (Figure S3).
To probe the relationship between small RNAs present in different AGO complexes, we compared the small RNA sequences among the four AGO data sets. We found that only a limited number of small RNAs were shared between the data sets (Figure 2C), indicating that distinct AGO complexes recruit specific subsets of small RNAs. Interestingly, although AGO1 and AGO5 belong to the same clade, only a small fraction of small RNAs (<4%) were present in both complexes.
In Drosophila and C. elegans, sorting of miRNAs and siRNAs into distinct AGO complexes is determined by the structural characteristics of their precursors (Forstemann et al., 2007; Steiner et al., 2007; Tomari et al., 2007). The presence of both miRNAs and siRNAs in each Arabidopsis AGO complex suggests that the association of small RNAs with a particular AGO protein may not be simply specified by either their biogenesis pathways or the structures of their precursors. Instead, we detected distinctive sequence preference from each AGO.
Bioinformatic analysis revealed a strong bias for sequences beginning with a 5′ terminal uridine (U; 86%) among the AGO1-associated small RNAs, whereas AGO5 read set was highly enriched for small RNA sequences initiating with a 5′ cytosine (C; 83%). Ninety-three percent of the small RNA reads associated with AGO2 have a 5′ terminal adenosine (A; 93%). AGO4 displays a similar preference for small RNAs with a 5′ terminal A (79%), albeit less pronounced than that seen for AGO2. Overall, 5′ nucleotide biases are absent from the Total small RNA population (37% terminal U, 23% terminal A, 18% terminal C, and 21% terminal G) (Figure 3A). No obvious preference for a particular nucleotide could be detected at other positions within small RNAs associated with each AGO complex (Figure 3B). Such 5′ terminal nucleotide-specific recruitment does not seem to depend on the sizes of the small RNAs and the biogenesis pathways that produce them (Figures 3C and S4). Similar results were obtained from analyses that were performed with data sets containing unique small RNA sequences (Figure S5). Thus, each Arabidopsis AGO preferentially recruits a group of small RNAs containing a particular 5′ terminal nucleotide.
Following these initial findings, we examined in greater detail whether the recruitment of miRNAs/miRNA*s and tasiRNAs by AGO complexes was specified by their 5′ terminal nucleotides. Most Arabidopsis miRNAs have a 5′ terminal U, and these were predominantly associated with AGO1. In contrast, few miRNA*s have a 5′ terminal U, which explains the low frequency of miRNA*s sequenced in AGO1 library compared to the Total small RNA population. As expected, the miRNA*s with a 5′ terminal A were predominantly associated with AGO2, whereas those with a 5′ terminal C were overwhelmingly found in AGO5 (Table S2). For example, miR391 and miR393b have a 5′ terminal U and were bound to AGO1, whereas their star strands have a 5′ terminal A and were found to be associated with AGO2 (Figure 4A). Two miRNAs, miR163.1 and miR163.2, are processed sequentially from their common precursor (Kurihara and Watanabe, 2004) and contain a 5′ terminal U and A, respectively. Correspondingly, we found miR163.1 was predominantly associated with AGO1, while miR163.2 was preferentially bound to AGO2. It is noteworthy that miR390, a miRNA with a 5′ terminal A, was underrepresented in the AGO1 read set. Instead, miR390 was abundant in the AGO2 library while miR390*, with a 5′ terminal C, was found primarily in the AGO5 library (Figure 4A; Table S2). Similar results were obtained in an analysis of tasiRNAs (Figures 4B and S6).
To rule out the possibility that the observed 5′ nucleotide preferences resulted from cloning or sequencing bias, we performed northern blots with small RNAs isolated from purified AGO complexes. Consistent with the sequencing results, miR391 and 393 were abundant in the AGO1 complex, but much less represented in AGO2 and AGO5 immunoprecipitates. In contrast, their star sequences were present in AGO2, but not detectable in AGO1 and AGO5. The signal for miR390 was strongest in AGO2, weaker in AGO1, and none in AGO5, whereas its star accumulated in AGO5, rather than in AGO1 or AGO2. While miR163.1 was only detected in AGO1, miR163.2 was more abundant in AGO2 than in AGO1 and AGO5 (Figure 4C).
Considered together, these data indicate that the loading of Arabidopisis small RNAs into AGO complexes is specified by the 5′ terminal nucleotide.
The spectrum of small RNAs associated with each AGO protein led us to propose that Arabidopsis AGOs have distinct binding affinities for small RNAs with different 5′ terminal nucleotides. To test this hypothesis, we designed four 21 nt RNA oligos that only differ at their 5′ ends (U, A, C, and G, respectively). For convenience, they are referred to as U21, A21, C21, and G21. The RNA oligos were 5′ end labeled and incubated with immunopurified AGO1, AGO2, and AGO5 proteins, respectively. The reaction mixtures were then irradiated with UV and resolved by SDS-PAGE. As shown in Figure 5A, AGO1 exhibited a much higher binding affinity for U21 than for A21, C21, or G21. In contrast, AGO2 and AGO5 showed the highest binding affinity for A21 and C21, respectively. To confirm the observed binding specificities, we performed competition assays using unlabeled U21 and A21. Increasing amounts (0.02–20 pmols) of unlabeled U21 or A21 were added into the binding reaction mixture containing AGO1 and labeled U21 (0.04 pmol). As shown in Figures 5B and 5C, the signal indicating crosslinked U21 decreased proportionally to the amount of unlabeled U21 added. The addition of unlabeled A21 had less of an effect on the binding of labeled U21 to AGO1, and ~10-fold higher amounts (20 pmols) were needed to impact the signal from bound RNAs. Conversely, in the binding reaction between labeled A21 and AGO2, unlabeled A21 served as a much stronger competitor than U21 (Figures 5B and 5C). These results demonstrate that the Arabidopsis AGO proteins indeed have differential binding affinities for small RNAs having different 5′ terminal nucleotides.
The UV-crosslinking experiments indicated a specific recognition of the 5′ terminal nucleotide of a small RNA by an AGO protein complex. We predicted that substituting the 5′ terminal nucleotide of a small RNA might redirect the small RNA from one AGO complex to another that specifically recognizes the new 5′ terminal nucleotide. To test this hypothesis, we performed mutagenesis on several miRNA precursors (pre-miRNAs).
We substituted the 5′ terminal U’s of miR391 and miR393b with A’s and the 5′ terminal A’s of the corresponding star strands with U’s. Additional mutations were also introduced in order to maintain the stem-loop structures of the pre-miRNAs (Figure 6A). The wild-type and mutant pre-miRNA constructs were agroinfiltrated into Nicotiana benthamiana leaves together with constructs expressing TAP-tagged AGO1 and AGO2, respectively. Primer extension experiments showed that the miRNAs processed from both wild-type and mutant precursors had the correct 5′ ends (Figure 6B). Northern blots indicated that the expression levels of wild-type and mutant miRNAs were comparable (Figure 6C). Next, we immunoaffinity-purified AGO1 and AGO2 complexes and assayed their association with particular small RNAs by northern blots. Consistent with our predictions, wild-type miR391 and 393b were detected in AGO1 but not AGO2 complexes, whereas their star strands were detected in AGO2 but not AGO1 complex (Figure 6C). Strikingly, the mutant miRNAs with a 5′ terminal U to A substitution were only detected in AGO2 but not in AGO1. The mutated star strands were redirected and became associated with AGO1, as predicted by their 5′ terminal U’s (Figure 6C). Similar results were also obtained for an artificial miRNA, amiR-trichome (Schwab et al., 2005), miR163.1, 163.2, 166g, and 390a (Figure 6C).
Our data indicate that different Arabidopsis AGOs have distinct binding affinities for small RNAs with different 5′ terminal nucleotides both in vitro and in vivo. As a first step to understand the biochemical basis for such specific recognition, we swapped the egions encoding Mid and PIWI domains between AGO1 and AGO2, resulting in two chimeric AGO proteins: AGO1PAZ:AGO2Mid+PIWI and AGO2PAZ:AGO1Mid+PIWI. The constructs expressing the chimeric AGOs were each agroinfiltrated into N. benthamiana together with wild-type (having a 5′ terminal A) and mutant (having a 5′ terminal U) pre-miR390a constructs. As shown in Figure 6C, AGO1 recruited mutant but not wild-type miR390a, and AGO2 bound wild-type but not mutant miR390a. Intriguingly, AGO1PAZ:AGO2Mid+PIWI recruited wild-type miR390a, whereas AGO2PAZ:AGO1Mid+PIWI preferentially bound mutant miR390a (Figure 6D), indicating that the recognition of the 5′ terminal nucleotide by AGO proteins is conferred by the Mid and PIWI domains. This is consistent with the structural analyses of A. fulgidus PIWI/RNA complexes, showing that the 5′ end of a small RNA is anchored in a basic pocket in the Mid domain (Ma et al., 2005; Parker et al., 2005).
In light of our findings, it was critical to assess the biological consequences of redirecting a particular small RNA into a different AGO. We tested whether loading of an artificial miRNA (amiRNA) into an alternative AGO complex would lead to the loss of its activity using a transgenic system. amiR-trichome is an artificial miRNA that has been designed to target two genes (CPC and TRY). Overexpression of this small RNA recapitulates the phenotypes of try and cpc single or double mutants (Schwab et al., 2005). We overexpressed amiR-trichome (having a 5′ terminal U) and amiR-trichomeM (having a 5′ terminal A) in Arabidopsis using pre-miR168 as the backbone. The empty backbone only was tested independently as a control. Primer extension and Northern blot experiments demonstrated that both amiR-trichome and amiR-trichomeM were correctly produced at similar levels in the transgenic plants (Figures 7A and 7B). Northern blots with RNAs prepared from immunopurified AGO1 and AGO2 complexes indicated that amiR-trichome was incorporated into AGO1 but not AGO2, whereas amiR-trichomeM was loaded into AGO2 but not AGO1 (Figure 7B). Confirming published results (Schwab et al., 2005), 43 of the 44 primary transformants of 35S:amiR-trichome displayed different degrees of clustered trichomes on leaves and flowers (Figures 7F–7H). In contrast, none of the 108 35S:amiR-trichomeM primary transformants exhibited clustered trichome phenotypes (Figures 7I–7K). Consistent with the phenotypes, the expression level of CPC in the 35S:amiR-trichome plants was reduced to ~50% of the control transformants, whereas its expression in the 35S:amiR-trichomeM remained essentially unchanged (Figure 7L). Thus, the 5′ terminal U to A substitution of amiR-trichome prevented the miRNA from reducing the expression of its target, resulting in the lack of a developmental phenotype upon overexpression in the transgenic plants. Overall, these data demonstrate the importance of selective small RNA loading for biological activity.
Our understanding of the mechanisms by which small RNAs are loaded into RISCs is derived mainly from biochemical studies in Drosophila. There, siRNA loading is facilitated by the RISC-loading complex (RLC) that contains Dicer-2 and its partner R2D2 (Liu et al., 2003; Pham et al., 2004; Tomari et al., 2004a). These proteins form a heterodimer, which senses the thermodynamics of a siRNA duplex and loads the strand with the less-stably paired 5′ end into RISC (Tomari et al., 2004b). A similar mechanism has also been proposed for loading Drosophila miRNAs into Ago1-RISC, but the proteins facilitating miRNA loading remain to be identified (Okamura et al., 2004).
Recent studies in Drosophila and C. elegans have addressed the mechanisms that assort small RNAs into specific Argonaute complexes. In neither case is small RNA loading coupled to precise biogenesis mechanisms, such as the Dicer that generates the mature RNA. Instead, structural features of precursor duplexes determine their ultimate Ago partner (Forstemann et al., 2007; Steiner et al., 2007; Tomari et al., 2007). Cloning and sequencing data indicate that most Arabidopsis miRNAs follow asymmetry guidelines (Jones-Rhoades et al., 2006), which implies an RLC analogous to Dicer-2/R2D2. However, the coexistence of miRNAs and siRNAs in each Arabidopsis AGO complex suggests that, in contrast to the loading of animal small RNAs, the structure of an Arabidopsis small RNA duplex may not play a role in sorting the small RNA. Instead, our data support a small RNA sorting mechanism in which the 5′ terminal nucleotide determines its loading into a particular AGO complex.
Structural analysis of A. fulgidus PIWI in complex with an siRNA revealed that the 5′ end of the RNA is anchored in a basic pocket in the Mid domain (Ma et al., 2005; Parker et al., 2005). Our data raise the possibility that the analogous pocket in each Arabidopsis AGO protein has evolved to recognize a specific nucleotide. Such specific recognition could be conferred exclusively by the AGO protein itself, as supported by our in vitro binding studies (Figure 5) and domain-swap experiments (Figure 6D), or may be additionally aided by an unidentified factor. We cannot rule out the possibility that there might also exist one factor recognizing a particular 5′ terminal nucleotide and pooling small RNAs to be accepted by an AGO protein recognizing the same nucleotide.
A sorting mechanism directed by the 5′ terminal nucleotide explains the predominant association between Arabidopsis miRNAs and AGO1. The need to act in concert with AGO1 has likely driven Arabidopsis miRNAs toward a strong 5′ U bias. In contrast, excluding miRNA*s from AGO1 complexes simply required evolution of a different 5′ nucleotide on this mature strand (Table S2). This type of discrimination is likely to be particularly important for the loading of miRNAs that do not follow thermodynamic asymmetry rules. In the miR391/miR391* and miR393b/miR393b* duplexes, the 5′ ends of the miRNA strands are more stable than those of the miRNA* strands (Table S2). These miRNAs, which initiate with a U, are nevertheless selectively loaded into AGO1. Their miRNA*s, which contain a 5′ terminal A, are instead incorporated into AGO2 complexes (Figures 4A, 4C, and and6C6C).
DCL1 cleavage is not always precise, and occasional processing errors could give rise to miRNA variants with 5′ heterogeneity (Rajagopalan et al., 2006). Moreover, some miRNA precursors, especially newly evolved ones, produce particularly heterogeneous small RNAs (Rajagopalan et al., 2006). For example, the miR163 precursor produces two miRNAs, miR163.1 and miR163.2, which have 5′ terminal U and A, respectively (Kurihara and Watanabe, 2004). Only miR163.1 is efficiently incorporated into AGO1 (Figures 4A, 4C, and and6C).6C). Therefore, the specific recruitment of small RNAs bearing a 5′ terminal U by the AGO1 complex can help to compensate for inaccurate Dicer processing, which could otherwise lead to off-target effects.
A terminal nucleotide-directed loading mechanism is clearly not the only determinant of the destination of a small RNA. Our sequencing data indicate that both AGO2 and AGO4 complexes prefer small RNAs having 5′ terminal A. Thus, it is surprising that AGO2 and AGO4 bind a limited number of small RNAs in common (<8%) (Figure 2C). Moreover, the types of small RNAs that join these two complexes differ significantly (Figure 2B). In part, this is reflected also in the binding of different size classes of small RNAs by each AGO protein (Figure 2A), suggesting a preferential coupling of each AGO with one or more Dicer-like protein. We are also unable to explain the loading of miR172 (with a 5′ terminal A) in AGO1 (Table S2).
We envision that several mechanisms act in concert to sort small RNAs into specific AGO complexes. First, the localization of AGO proteins to specific subcellular compartments likely determines their access to different classes of small RNAs. AGO4 is localized in nucleus (Li et al., 2006; Pontes et al., 2006). This could explain why 24 nt rasiRNAs that are made and function in the nucleus are mainly associated with AGO4. Second, the tissue-specific and developmental expression pattern of an AGO protein may enable it to preferentially recruit small RNAs that share that expression pattern. Third, distinct RLCs may interact with different groups of small RNAs to facilitate their loading into distinct AGO complexes.
Given that there are ten AGO proteins in Arabidopsis, redundancy must exist between the AGO proteins in their recognition of the four possible 5′ terminal nucleotides. miR390-mediated cleavage of TAS3 is required to initiate the production of tasiRNAs, which regulate the vegetative phase transition in Arabidopsis (Allen et al., 2005; Axtell et al., 2006; Peragine et al., 2004; Vazquez et al., 2004; Xie et al., 2005). We found that miR390 (bearing a 5′ terminal A) was underrepresented in AGO1 and instead accumulated in AGO2 (Figure 4). However, there was no detectable change in the accumulation of miR390 and tasiRNAs in an ago2 null plant (Salk_037548), and the vegetative phase transition is normal in ago2 mutant plants (Figure S7). These data suggest that another AGO protein must play a predominant role in recruiting miR390. Genetic studies have shown that ago7 mutants have defects in the regulation of vegetative phase change (Adenot et al., 2006; Fahlgren et al., 2006; Garcia et al., 2006; Hunter et al., 2006), which points to AGO7 as the best candidate.
Arabidopsis contains one of the most complex small RNA regulatory networks yet discovered. While 5′ end recognition mechanism may be restricted to plants, it may also operate in other organisms with highly elaborated Argonaute families. C. elegans has 27 Argonaute proteins that must distinguish between primary and secondary siRNAs bearing monophosphate and triphosphate termini, respectively (Pak and Fire, 2007; Sijen et al., 2007; Yigit et al., 2006). Presumably, other Argonautes specifically associate with 21U RNAs that initiate with 5′ U (Ruby et al., 2006) and siRNAs that begin predominantly with a G (Ambros et al., 2003; Ruby et al., 2006). In several animals, Piwi-interacting RNAs (piRNAs) show a strong preference for a 5′ U (Aravin et al., 2006; Girard et al., 2006; Grivna et al., 2006; Lau et al., 2006). It has been presumed thus far that this bias reflects the biogenesis mechanism for this RNA class; however, our findings raise the possibility that the observed bias is instead generated by selection of RNAs with an appropriate 5′ end by Piwi clade Argonautes. Overall, many tantalizing hints suggest that the mechanisms that we have uncovered in plants are general properties of Argonaute proteins that have been exploited more broadly to help compartmentalize small RNA regulatory pathways.
The rabbit polyclonal antiserum against AGO1 was described previously (Qi et al., 2005). Peptides AGO2N (NRGQGRGEQQC) and AGO5N (RRSDQRQDQSSC) were conjugated to mcKLH using the Imject Maleimide Activated mcKLH Kit (Pierce) and used to raise rabbit polyclonal antibodies against AGO2 and AGO5, respectively. The antisera were affinity purified and used for western blot (1:1000 dilution) and immunoprecipitation (1:100 dilution). Arabidopsis ecotype (Col-0) was used for the purification of AGO1, AGO2, and AGO5 complexes. A commercial monoclonal myc antibody (Roche) was used for the purification of AGO4 complex from a transgenic Arabidopsis line expressing myc-tagged AGO4 under its native promoter (Qi et al., 2006).
Immunopurification of each AGO complex was performed as previously described (Qi et al., 2005). To examine the quality of the purifications, small fractions of the immunoprecipitates were examined by SDS-PAGE followed by silver staining. The bands of expected sizes were confirmed as AGO proteins by mass spectrometry.
RNAs were extracted from the purified AGO complexes by Trizol reagent (Invitrogen), resolved on a 15% denaturing PAGE gel, and visualized by SYBR-gold (Invitrogen) staining. Gel slices within the range of 18–28 nt were excised, and the RNAs were eluted and purified for cloning.
Cloning of small RNAs was carried out essentially as described (Qi et al., 2006), except that sequences of the 5′ RNA adaptor and primers for reverse transcription PCR were changed for Illumina 1G sequencing (Margulies et al., 2005). A detailed protocol is available upon request.
The adaptor sequences in Illumina 1G sequencing reads were removed by using “vectorstrip” in the EMBOSS package. The small RNA reads with length of 19–27 nt were mapped to the Arabidopsis nuclear, chloroplast, and mitochondrial genomes (http://www.arabidopsis.org). The small RNAs with perfect genomic matches were used for further analysis. Annotation of small RNAs was performed using the following databases: TAIR7 annotations for coding sequences and noncoding RNAs (rRNAs, tRNAs, snoRNAs, snRNAs), and sequences from the intergenic regions (ftp://ftp.arabidopsis.org/Sequences/blast_datasets/TAIR7_blastsets), Repbase (http://www.girinst.org) for transposons and repeats, ASRP for tasiRNA annotations (http://asrp.cgrb.oregonstate.edu), and miRBase for miRNA annotations (http://microrna.sanger.ac.uk/sequences). Annotations for the cis- or trans-natural antisense genes were extracted from published databases (Margulies et al., 2005; Wang et al., 2006). The relative frequency of each nucleotide at each position of the small RNAs was calculated and graphically represented by using WebLogo (Crooks et al., 2004). The determination of miRNA/miRNA* duplex asymmetry was done by using the method described (Rajagopalan et al., 2006).
Northern blot analysis with enriched small RNAs or RNAs prepared from purified AGO complexes was performed as described (Qi et al., 2005). LNA probes complementary to miRNAs or miRNA*s were 32P-end labeled by T4 polynucleotide kinase and used for the northern blots.
Primer extension reactions were performed essentially as described (Steiner et al., 2007). 5′ end-labeled 18 nt DNA oligonucleotides that are complementary to the positions 4–21 of miR391, miR393b and amiR-trichome, were used as primers.
UV-crosslinking assays were performed essentially as described (Rivas et al., 2005). RNA oligos bearing different 5′ terminal nucleotides were used: U21(UUCACAUUGCCCAAGUCUCdTdT), A21(AUCACAUUGCCCAAGUCUCdTdT), C21(CUCACAUUGCCCAAGUCUCdTdT), and G21(GUCACAUUGCCCAAGUCUCdTdT). In the competition experiments, increasing amounts (0.02–20 pmols) of unlabeled U21 or A21 were added in the reactions.
Detailed procedures for DNA construction and plant transformation are described in Supplemental Experimental Procedures.
The expression level of CPC gene was examined by real-time quantitative RT-PCR. Total RNAs were prepared from 35:amiR-trichome, 35:amiR-trichomeM, and control tranformants, and converted to cDNAs using Superscript III reverse transcriptase (Invitrogen) primed by oligo(dT). The cDNAs were then used as templates for real-time PCR with gene-specific primers CPC-F (TTCCGAAGAGGTGAGTAGTA) and CPC-R (AGTCTCTTCGTCTGTTGGCA). Real-time PCR was performed using SYBR Premix EX Taq (TaKaRa) on Mastercycler ep realplex (Eppendorf). The actin gene was detected in parallel and used as the internal control.
We thank M. Carmell, B. Ding, X. Wang, and J. Liu for critical reading of the manuscript. We thank the NIBS Antibody Facility for generating the antisera used in this study. G.J.H. is an investigator of the Howard Hughes Medical Institute. Y.Q. is supported by the Chinese Ministry of Science and Technology.
The small RNAs from AGO1, AGO2, AGO4, and AGO5 complexes were deposited in the NCBI Gene Expression Omnibus (http://www.ncbi.nlm.nih.gov/geo) under accession number GSE10036.
Supplemental Data include seven figures, two tables, Supplemental Experimental Procedures, and Supplemental References and can be found with this article online at http://www.cell.com/cgi/content/full/133/1/116/DC1/.