The 21-Nucleotide, but Not 22-Nucleotide, Viral Secondary Small Interfering RNAs Direct Potent Antiviral Defense by Two Cooperative Argonautes in Arabidopsis thaliana

Arabidopsis thaliana defense against distinct positive-strand RNA viruses requires production of virus-derived secondary small interfering RNAs (siRNAs) by multiple RNA-dependent RNA polymerases. However, little is known about the biogenesis pathway and effector mechanism of viral secondary siRNAs. Here, we describe a mutant of Cucumber mosaic virus (CMV- D 2b) that is silenced predominantly by the RNA-DEPENDENT RNA POLYMERASE6 (RDR6)-dependent viral secondary siRNA pathway. We show that production of the viral secondary siRNAs targeting CMV- D 2b requires SUP- PRESSOR OF GENE SILENCING3 and DICER-LIKE4 ( DCL4 ) in addition to RDR6 . Examination of 25 single, double, and triple mutants impaired in nine ARGONAUTE ( AGO ) genes combined with coimmunoprecipitation and deep sequencing identiﬁes an essential function for AGO1 and AGO2 in defense against CMV- D 2b, which act downstream the biogenesis of viral secondary siRNAs in a nonredundant and cooperative manner. Our ﬁndings also illustrate that dicing of the viral RNA precursors of primary and secondary siRNA is insufﬁcient to confer virus resistance. Notably, although DCL2 is able to produce abundant viral secondary siRNAs in the absence of DCL4, the resultant 22-nucleotide viral siRNAs alone do not guide efﬁcient silencing of CMV- D 2b. Possible mechanisms for the observed qualitative difference in RNA silencing between 21- and 22-nucleotide secondary siRNAs are discussed.

Recent studies have demonstrated production of viral secondary siRNAs by multiple host RDR pathways in Arabidopsis infected by distinct positive-strand RNA viruses (Diaz-Pendon et al., 2007;Donaire et al., 2008;Garcia-Ruiz et al., 2010;Wang et al., 2010). RDR recruitment and synthesis of secondary siRNAs in Caenorhabditis elegans occur after cleavages of RNA transcripts targeted by primary siRNAs loaded in a specific AGO. Arabidopsis may encode six RDRs, of which RDR2 and RDR6 participate in the biogenesis of repeat-associated siRNAs and transgene siRNAs, respectively (Mourrain et al., 2000;Boutet et al., 2003;Xie et al., 2003). Plant ta-siRNAs resemble secondary siRNAs because production of ta-siRNAs requires AGO1/AGO7-mediated cleavages of transcripts targeted by microRNAs (miRNAs), followed by SGS3/RDR6-dependent production of dsRNAs and DCL4-dependent processing (Voinnet, 2008). However, unlike C. elegans secondary siRNAs that are Dicer independent, plant RDR-dependent siRNAs are Dicer products and are thus structurally indistinguishable from primary siRNAs (Voinnet, 2008).
Infection of Arabidopsis rdr mutants by virus mutants deficient in silencing suppression has recently revealed that resistance to distinct positive-strand RNA viruses depends on the production of viral secondary siRNAs by RDR1, RDR2, or RDR6 (Donaire et al., 2008;Garcia-Ruiz et al., 2010;Wang et al., 2010). These studies support early findings that plant lines compromised for the function of either RDR1 or RDR6 exhibit increased susceptibility to some of the RNA viruses examined (Mourrain et al., 2000;Xie et al., 2001;Yang et al., 2004;Qu et al., 2005;Schwach et al., 2005;Donaire et al., 2008). In this study, we investigated the biogenesis pathway of the viral secondary siRNAs following identification of a cucumber mosaic virus (CMV) mutant that was targeted for silencing predominantly by the RDR6-dependent viral secondary siRNA pathway. We also determined if the mutant virus was silenced by the destruction of the virus RNAs during the production of viral secondary siRNAs or by the activity of the resultant viral secondary siRNAs. Examination of 25 single, double, and triple ago mutants involving nine of the 10 Arabidopsis AGO genes revealed the role of AGO1 and AGO2 in the defense against the mutant virus. Notably, our studies revealed a qualitative difference in silencing virus RNAs between 21-and 22-nucleotide siRNAs produced by DCL4 and DCL2, respectively, which might be mechanistically related to the distinct activities of 21-and 22-nucleotide miRNAs in triggering the biogenesis of ta-siRNAs reported recently Cuperus et al., 2010).

A Mutant CMV Uncouples RDR1-and RDR6-Dependent Virus Resistance
We have shown previously that disease resistance to a mutant of CMV deficient in silencing suppression is mediated by viral secondary siRNAs produced by both the RDR1 and RDR6 pathway (Wang et al., 2010). Here, we constructed a different mutant of CMV in which the start codon and two closely spaced AUG codons of the 2b open reading frame were mutated by point mutations so that the 2a sequence coding for the viral RDR in the overlapping reading frame was not altered ( Figure 1A). This mutant was designated CMV-D2b and the mutant used in the previous study (Wang et al., 2010) renamed as CMV-2aTD2b since the 295-nucleotide deletion introduced in CMV RNA2 also led to early termination of the 2a protein ( Figure 1A). As expected, expression of the 2b protein was detected in plants infected with CMV but not in those infected with either CMV-D2b or CMV-2aTD2b (see Supplemental Figure 1A online).
Inoculation of wild-type Arabidopsis plants with CMV-D2b induced no phenotypic change compared with mock inoculation ( Figure 1B, top right) as found in previous studies (Ding et al., 1995;Soards et al., 2002;Lewsey et al., 2009). However, we found that CMV-D2b caused severe disease symptoms in all of the Arabidopsis mutants that contained a loss-of-function mutation in RDR6, including rdr6, rdr1 rdr6, rdr2 rdr6, and rdr1 rdr2 rdr6 mutants ( Figure 1B; see Supplemental Figure 1B online). By contrast, rdr1, rdr2, and rdr1 rdr2 mutants that were wild type for RDR6 were as resistant as wild-type plants to CMV-D2b ( Figure  1B; see Supplemental Figure 1B online). Thus, CMV-D2b induced severe disease symptoms following genetic inactivation of the single host gene RDR6 and therefore was distinct from CMV-2aTD2b, which became highly virulent only in mutant plants where neither RDR6 nor RDR1 was functional (Wang et al., 2010) (Figure 1B, compare middle and right columns). Moreover, we found that CMV-D2b, but not CMV-2aTD2b, was highly virulent in plants carrying a loss-of-function allele in SGS3, which is required for RDR6-dependent biogenesis of transgene/ ta-siRNAs (Voinnet, 2008). RNA gel blot hybridizations detected greatly increased levels of CMV-D2b genomic and subgenomic RNAs in rdr6, rdr1 rdr6, and sgs3 plants that exhibited severe disease symptoms compared with wild-type and rdr1 plants that displayed no visible symptoms ( Figures 1B and 1C). These results show that the RDR6/SGS3 pathway mediates the Arabidopsis resistance to CMV-D2b in an RDR1-independent manner.
Characterization of the Genetic Pathway for Viral Secondary siRNA Biogenesis RNA gel blot hybridizations were used to compare the accumulation of viral siRNAs in wild-type and mutant plants inoculated with CMV-D2b. CMV-D2b replicated to lower levels and was expected to produce less dsRNA replicative intermediates for dicing into viral primary siRNAs in wild-type plants than in rdr6 plants. However, we found that viral siRNAs in fact accumulated to higher levels in wild-type plants than in rdr6 plants ( Figure 2A). This indicates that most of the viral siRNAs detected in CMV-D2b-challenged wild-type plants are viral secondary siRNAs amplified by an RDR6-dependent pathway. Amplification of viral siRNAs was also detected in rdr1 plants but not in rdr1 rdr6 and sgs3 plants (Figure 2A). These findings show that the RDR6/ SGS3 pathway produces viral secondary siRNAs to confer resistance to CMV-D2b in an RDR1-independent manner. We noted that CMV-D2b accumulated to higher levels in the dcl2 dcl4 mutant than in rdr6 and sgs3 plants (Figure 2A; see below). This indicates that production of RDR-independent viral primary siRNAs in rdr6 and sgs3 plants was associated with silencing of CMV-D2b but was insufficient to confer resistance to CMV-D2b, which was similar to that reported previously (Wang et al., 2010).
To verify RDR6-dependent production of viral secondary siRNAs targeting CMV-D2b, total small RNAs were extracted from wild-type and rdr6 plants 3 weeks after CMV-D2b inoculation and sequenced by the Illumina Analyzer II as described previously (Wang et al., 2010). Small RNAs sequenced from both wild-type and rdr6 plants that were 100% identical or comple-mentary in sequence to the three genomic RNAs of CMV-D2b were predominantly 21 nucleotides in length (see Supplemental Figure 2 online). Approximately 10% of CMV-D2b-derived small RNAs were 22 nucleotides in length. These results are consistent with previous observations that DCL4 has a dominant role over DCL2 in the production of siRNAs targeting positive-strand RNA viruses (Ding, 2010;Llave, 2010).
Mapping of the perfectly matched siRNAs to the three viral genomic RNAs revealed shared features in the production of viral siRNAs between CMV-D2b-challenged wild-type plants ( Figure  2B) and CMV-2aTD2b-challenged rdr1 mutant plants (Wang et al., 2010) because amplification of viral siRNAs only relies on the RDR6 pathway in both cases (GSM498721-6). Compared with wild-type plants, the profile of viral siRNAs in rdr1 plants is characterized with a reduced density of RDR1-dependent siRNAs targeting the 59-terminal region of the viral RNAs 1 and 2 and an increased production of RDR6-dependent siRNAs targeting the 4th region of the viral RNA3 as specified in Figure 2C ( Wang et al., 2010). Low density of the 59-terminal viral siRNAs and high density of siRNAs targeting the same region of RNA3 were both detected in CMV-D2b-challenged wild-type plants ( Figure 2B (C) Accumulation of CMV-D2b and CMV-2aTD2b genomic RNAs (RNAs 1-3) and the subgenomic RNA (RNA4) in wild-type and mutant plants 3 weeks after inoculation detected by a probe specific for the 39-untranslated region conserved in all of the four CMV RNAs. 25S rRNA was used as the loading control. Values at the bottom of this panel refer to the averages of the relative hybridization signal intensity of RNA1, 2, and 3 in three independent experiments with the accumulation levels of CMV-2aTD2b in wild-type plants (left panel) or sgs3 plants (right panel) set as 1. Note that the hybridization signal intensity readings were from measuring both a short exposure (for rdr6, rdr1/6, sgs3, dcl4, and dcl2 dcl4 samples) and a long exposure (for wildtype, rdr1, rdr6, rdr1/6, sgs3, and dcl4 samples). The image shown was from a longer exposure to detect the accumulation of mutant viruses in wild-type and rdr1 plants.
the density of siRNAs targeting the RNA3 region was markedly reduced in rdr6 plants ( Figure 2B, bottom; see Supplemental Figure 3 online) compared with wild-type plants, indicating that amplification of siRNAs targeting this region of RNA3 is RDR6 dependent. Moreover, we noted an overall reduction in the density of siRNAs targeting both RNA1 and RNA2 in rdr6 plants compared with wild-type plants ( Figure 2B). These results from deep sequencing support the genetic analysis and show that siRNAs targeting CMV-D2b in Arabidopsis plants are amplified by the RDR6-dependent pathway in an RDR1-independent manner.
Gel blot hybridization approaches have been used previously to map hot spots of siRNAs targeting distinct genomic regions of CMV and other viruses (Szittya et al., 2010;Wang et al., 2010). As described previously (Wang et al., 2010), RNA3 of CMV was first divided evenly into five regions for DNA synthesis by PCR. Equal amount of DNA fragments corresponding to each region was fractionated and hybridized to the 32 P-labeled total small RNAs harvested and gel purified from wild-type and rdr6 plants 3 weeks after CMV-D2b inoculation ( Figure 2C). We found that hybridization with the small RNA probe from CMV-D2b-challenged wild-type plants produced much stronger signal to the 4th region of RNA3 than the neighboring regions. However, probing the same panel of DNA fragments with the small RNA probe from CMV-D2b-infected rdr6 plants failed to detect a stronger signal to the 4th region of RNA3 compared with the neighboring regions ( Figure 2C). This region corresponds to the 59 half of RNA4, which is the most abundantly expressed viral RNA to act as mRNA for the viral coat protein encoded by RNA3. Thus, both deep sequencing and the gel blot hybridization approach demonstrate RDR6-dependent production of the hot spots of siRNAs targeting the 4th region of CMV-D2b RNA3 in the challenged Arabidopsis plants.
We then used the diagnostic hot spots of siRNAs targeting the 4th region of CMV-D2b RNA3 as a molecular marker to follow RDR6-dependent production of viral secondary siRNAs in rdr1, rdr1 rdr6, and sgs3 plants ( Figure 2C). We found that the strong signal hybridizing to the 4th region of RNA3 was detected in rdr1 plants but not in rdr1 rdr6 and sgs3 plants ( Figure 2C). Therefore, these findings together show that challenge by CMV-D2b induces production of viral secondary siRNAs in a pathway dependent on RDR6 and SGS3 but independent of RDR1.

Potent Antiviral Silencing Requires Production of Viral Secondary siRNAs by DCL4 but Not by DCL2
Regardless of RDR6-dependent amplification, viral siRNAs were predominantly 21 nucleotides in length in all of the CMV-D2binoculated plants where DCL4 was functional ( Figure 2A). These included wild-type plants and rdr1, rdr6, rdr1 rdr6, sgs3, and dcl2 mutant plants. By contrast, DCL2-dependent 22-nucleotide viral siRNAs accumulated to readily detectable levels only in the dcl4 mutant where DCL4-mediated biogenesis of 21-nucleotide siRNAs was eliminated ( Figure 2A). Notably, the diagnostic hot spots of siRNAs targeting the 4th region of CMV-D2b RNA3 were detected in both dcl2 and dcl4 single mutant plants but not in the dcl2 dcl4 mutant in which both DCL1 and DCL3 are functional ( Figure 2C). In addition, Arabidopsis plants infected with CMV-D2b produced high density of 21-and 22-nucleotide viral siRNAs targeting this region of RNA3 in an RDR6-dependent manner (see Supplemental Figure 3 online). These results show that either DCL4 or DCL2 alone can recognize and process the (A) Accumulation of CMV-D2b RNAs and siRNAs in wild-type (wt) and mutant plants 3 weeks after inoculation. 25S rRNA and U6 RNA were used as the high and low molecular RNA loading controls, respectively. nt, nucleotide. (B) Mapping of perfect match 18-to 25-nucleotide virus-derived small RNAs (reads per million of total reads) sequenced by Illumina from wildtype and rdr6 mutant plants challenged by CMV-D2b. Positive-and negative-strand reads are shown in gray and black, respectively. Note that the scale for RNA3 is 5 times that of RNAs 1 and 2.
(C) Mapping of siRNA hot spots targeting RNA3 by gel blot hybridization. Equal amount of DNA fragments specific to RNA3 (five fragments, 450 bp each) or b-glucuronidase (GUS; as negative control) was fractionated and hybridized to the labeled total small RNAs (18-to 30-nucleotide) harvested and gel purified from wild-type and mutant plants challenged by CMV-D2b. dsRNA products of the RDR6/SGS3 pathway, leading to abundant accumulation of viral secondary siRNAs of 21 nucleotides in length in the dcl2 mutant and of 22 nucleotides in length in the dcl4 mutant, respectively ( Figure 2A). By contrast, neither DCL1 nor DCL3 has access to the dsRNA precursor of viral secondary siRNAs, although DCL3 is able to produce viral primary siRNAs.
We found that, despite the absence of 22-nucleotide viral siRNAs, the dcl2 mutant was as resistant to CMV-D2b as wildtype plants and potently inhibited the replication of CMV-D2b as efficiently as wild-type plants (Figure 2A). By contrast, the dcl4 mutant was as susceptible to CMV-D2b as rdr6 and sgs3 mutants and supported high levels of CMV-D2b replication ( Figure 2A). It should be pointed out that the accumulation of CMV-D2b was higher in dcl2-dcl4 plants than in dcl4, rdr6, and sgs3 plants (Figure 2A, see below), indicating that silencing of CMV-D2b occurs in dcl4, rdr6, and sgs3 mutants, likely through primary viral siRNAs. Together, these results show that production of 21-nucleotide viral siRNAs by DCL4 is sufficient to confer resistance to CMV-D2b in the absence of the 22-nucleotide viral siRNAs produced by DCL2, whereas 22-nucleotide viral siRNAs produced by DCL2 do not guide efficient silencing of CMV-D2b in the absence of the 21-nucleotide viral siRNAs produced by DCL4. Therefore, secondary siRNA-dependent silencing of CMV-D2b relies on DCL4-dependent 21-nucleotide siRNAs but not DCL2-dependent 22-nucleotide siRNAs, revealing a difference between 21-and 22-nucleotide viral siRNAs in antiviral defense.

AGO1 and AGO2 Have Independent Function in Defense against CMV-D2b
Viral secondary siRNAs may guide specific silencing of viral RNAs in an AGO-containing effector complex as shown in previous studies that did not distinguish between viral primary and secondary siRNAs (Omarov et al., 2007;Pantaleo et al., 2007;Azevedo et al., 2010). However, the step of dicing to produce viral secondary siRNAs may be sufficient to confer virus resistance because it would destroy the viral RNAs targeted for dsRNA synthesis by the RDR6 pathway. To investigate these possibilities, we systematically probed the role of AGOs in defense against CMV-D2b, which was highly susceptible to silencing by RDR6-dependent viral secondary siRNAs. Arabidopsis encodes 10 AGOs that are divided into three clades (Vaucheret, 2008). One AGO clade contains AGO1, AGO5, and AGO10 (Vaucheret, 2008). We found that CMV-D2b induced disease symptoms in hypomorphic ago1 mutants but not in ago5 or ago10 mutants (Figure 3). Both RNA gel blot hybridizations and real-time PCR detection of the viral positive and negative strand RNA2 showed that CMV-D2b accumulated to ;9-fold higher levels in ago1 mutant plants compared with wild-type plants ( Figures 4B and 5B; P values < 0.005). The accumulation levels of CMV-D2b were similar among ago1, rdr6, and sgs3 mutant plants but were ;10-fold lower than those in dcl2-dcl4 plants (Figures 2A, 4B, and 5B), suggesting that AGO1 is required for antiviral silencing by viral secondary siRNAs. We observed no major differences in disease symptoms and accumulation levels of CMV-D2b in the ago1 ago5, ago1 ago10, and ago1 ago5 ago10 combinations compared with ago1 ( Figure 4B). These results indicate that AGO1 plays a key role in Arabidopsis resistance to CMV-D2b, whereas AGO5 and AGO10 do not have an independent or additive role in wild-type plants, even though a previous study detected loading of CMV siRNAs in plants overexpressing AGO5 (Takeda et al., 2008).
The second AGO clade contains AGO2, AGO3, and AGO7, among which AGO2 and AGO3 have high sequence similarity to each other and their respective genes are found directly in tandem in the genome (Vaucheret, 2008). We found that CMV-D2b induced mild dwarfing in the ago2 mutant ( Figure 3) and accumulated to ;3-fold higher levels in this mutant compared with wild-type, ago3, and ago7 plants (Figures 4C and 5B; P values < 0.005). The ago2 ago7 double mutant exhibited a response to CMV-D2b similar to that of the ago2 single mutant, and the ago3 ago7 double mutant was as resistant to CMV-D2b as wild-type, ago3, and ago7 plants. These findings reveal an antiviral function for AGO2 that is independent of AGO1, whereas AGO3 and AGO7 do not have an independent or additive role.
The last AGO clade contains three full-length proteins, AGO4, AGO6, and AGO9, and a truncated protein, AGO8, which lacks the catalytic DDH motif of the PIWI domain because of incorrect splicing (Takeda et al., 2008). We found that none of the ago4, ago6, ago9, ago4 ago6, ago4 ago9, ago6 ago9, and ago4 ago6 ago9 mutants was detectably compromised in resistance to CMV-D2b ( Figure 4A). This result suggests that AGO4, AGO6, and AGO9 do not play a critical role in the silencing of CMV-D2b by the viral secondary siRNAs.

AGO1 and AGO2 Act Cooperatively Downstream of Viral Secondary siRNA Biogenesis
Six additional double and triple ago mutants were constructed between members of the AGO1 and AGO2 clades and examined for susceptibility to CMV-D2b. We found that pyramiding ago3 and ago7 mutations either alone or simultaneously over the ago1 allele did not further increase Arabidopsis susceptibility to CMV-D2b (Figures 3 and 4D). By contrast, CMV-D2b caused more severe disease symptoms ( Figure 3) and accumulated at higher levels in the ago1 ago2 double mutant compared with either ago1 or ago2 single mutant ( Figures 5A and 5B), indicating cooperative activities of AGO1 and AGO2. Adding either the ago10 mutation or the ago7 mutation over the ago1 ago2 double mutant did not further increase Arabidopsis susceptibility to CMV-D2b ( Figures  3 and 5A), indicating that AGO7 and AGO10 do not have additive roles.
In contrast with the reduced accumulation of viral siRNAs in rdr6 and sgs3 plants (Figure 2A), we found that viral siRNAs were more abundant in ago1 and ago2 single mutants as well as in ago1 ago2, ago1 ago2 ago7, and ago1 ago2 ago10 mutants compared with wild-type plants following CMV-D2b challenges ( Figure 5A). We further determined if the abundant viral siRNAs accumulated in mutant plants carrying ago1 and/or ago2 alleles included RDR6-dependent viral secondary siRNAs using the gel blot hybridization approach described above ( Figure 5C). The characteristic strong hybridization signal corresponding to the 4th region of CMV-D2b RNA3 was detected in ago1, ago2, and ago1 ago2 mutant plants ( Figure 5C). These findings show that AGO2 and possibly AGO1 are dispensable for the biogenesis of viral secondary siRNAs. We therefore conclude that the defect of ago1, ago2, and ago1 ago2 mutant plants in resistance to CMV-D2b is not caused by loss of production of viral secondary siRNAs and that AGO1 and AGO2 act cooperatively downstream the biogenesis of viral secondary siRNAs. Our results also showed that active dicing of the viral siRNA precursors, which occurred in the ago1 ago2 mutant highly susceptible to CMV-D2b, was insufficient to confer virus resistance.
By comparison, CMV-D2b accumulated to significantly higher levels in the mutant plants containing both the ago1 and ago2 alleles than in rdr6 and sgs3 mutants (Figures 5A and 5B). As described above, rdr6 and sgs3 mutants were defective in the biogenesis of RDR6-dependent viral secondary siRNAs, and silencing of CMV-D2b by viral primary siRNAs was insufficient to confer resistance to CMV-D2b. Therefore, our results indicate that AGO1 and AGO2 are required for the antiviral activities of viral primary and secondary siRNAs. We noted that the accumulation of CMV-D2b in ago1 ago2 plants was ;2-fold lower than those in dcl2 dcl4 plants (Figures 5A and 5B; P values < 0.01), probably because the ago1 allele used in this study is a hypomorphic allele. Indeed, only hypomorphic ago1 mutants that retain partial AGO1 activity develop enough to be challenged by viruses (Morel et al., 2002).

Strong Positive-Strand Bias of Viral siRNAs Loaded in AGO1 but Not in AGO2
To characterize the viral siRNA populations loaded in AGO1 and AGO2, we constructed and sequenced small RNA libraries from CMV-D2b-infected plants expressing tagged AGO1 or AGO2 both before and after immunoprecipitation. Both AGO1 and AGO2 complexes contained viral siRNAs in addition to the known classes of endogenous small RNAs ( Figure 6) as reported previously (Baumberger and Baulcombe, 2005;Zhang et al., 2006;Mi et al., 2008;Montgomery et al., 2008;Takeda et al., 2008). Comparisons between the small RNA populations obtained before and after immunoprecipitation revealed several notable differences ( Figure 6). The virus-derived siRNAs were highly enriched following coimmunoprecipitation with AGO2 in two biological repeats ( Figure 6A). However, we observed enrichment of viral siRNAs after AGO1 coimmunoprecipitation only in one of the two biological repeats due to variations of the endogenous small RNA contents. We found that sorting of viral siRNAs into AGO1 and AGO2 depends on 59-terminal U and A, respectively, as found previously for endogenous small RNAs (Mi et al., 2008;Montgomery et al., 2008;Takeda et al., 2008). Approximately equal ratios of the total viral siRNAs produced in the CMV-D2b-infected plants were mapped to the positive and negative strands of the viral genomic RNAs, respectively (Figures 6B). Viral siRNAs loaded in AGO2 also contained approximately equal ratios of the plus and minus strands. By contrast, viral siRNAs loaded in AGO1 were predominantly from positive strands ( Figures 6B). In addition, we noted that the viral siRNAs loaded in AGO2 contained a lower ratio of 22-and 21-nucleotide species than those loaded in AGO1. Nevertheless, we detected the characteristic strong hybridization signal corresponding to the 4th region of CMV-D2b RNA3 using as the probe the total small RNAs coimmunoprecipitated with either AGO1 or AGO1 ( Figure  6C; see Supplemental Figure 4 online), indicating that both AGO1 and AGO2 have the ability to bind viral secondary siRNAs.

Silencing of Distinct Mutant Viruses Depends on Amplification of Viral siRNAs by Single or Double RDR-Dependent Pathways
As described previously (Wang et al., 2010), CMV-2aTD2b challenges induce amplification of viral siRNAs by both RDR1-and RDR6-dependent pathways with diagnostic hot spots of siRNAs targeting the 59-terminal regions of RNA1/RNA2 and a region near the 39 terminus of RNA3, respectively. Importantly, production of viral secondary siRNAs from either genetic pathway in rdr1 and rdr6 single mutants is sufficient to confer resistance so that CMV-2aTD2b causes diseases and replicates to high levels only in plants carrying both rdr1 and rdr6 mutant alleles (Wang et al., 2010). Here, we show that Arabidopsis resistance to CMV-D2b is associated with amplification of viral siRNAs predominantly by the RDR6-dependent pathway (Figure 2). Notably, CMV-D2b became highly virulent in rdr6 and sgs3 mutants (Figures 1 and 2) and accumulated to ;9-fold higher levels than in wild-type plants ( Figure 5B; P values < 0.025), demonstrating that effective defense against CMV-D2b requires production of viral secondary siRNAs. However, eliminating both primary and secondary siRNAs in the dcl2 dcl4 mutant further increased accumulation of CMV-D2b compared with that in rdr6 and sgs3 mutants defective only in the production of secondary siRNAs ( Figure 5B). Therefore, production of RDR-independent viral primary siRNAs in rdr6 and sgs3 mutants is associated with silencing of CMV-D2b but is insufficient to confer virus resistance in the absence of viral secondary siRNAs (Figure 1), which is consistent with previous studies Wang et al., 2010). Differential induction and/or suppression of RDR1-and RDR6dependent amplification of viral siRNAs by distinct viruses may explain why RDR6-deficient Arabidopsis and Nicotiana benthamiana plants are more susceptible than wild-type plants to only some of the RNA viruses examined (Dalmay et al., 2000;Mourrain et al., 2000;Qu et al., 2005;Schwach et al., 2005;Vaistij and Jones, 2009). To account for the specific induction of RDR1and RDR6-dependent amplification of viral siRNAs, we hypothesize that RDR6 and RDR1 are recruited to viral positive-strand RNAs acting as mRNA for translation and as template for RNA replication, respectively. In this model, recognition and silencing of viral mRNAs is analogous to the silencing of sense transgenes by the RDR6/SGS3 pathway (Voinnet, 2008). This is consistent with the observed preferential targeting of the more abundant RNAs of CMV, RNA3/RNA4, in plants challenged by either CMV-D2b or CMV-2aTD2b. On the other hand, production of the 59terminal siRNAs targeting a viral positive-strand RNA genome in Drosophila melanogaster is triggered by the synthesis of progeny positive-strand viral RNAs from the negative-strand RNA template (Aliyari et al., 2008). Similarly produced 59-terminal viral primary siRNAs in plants may facilitate recruiting RDR1 to target the nascent products synthesized by the truncated viral RdRp of CMV-2aTD2b ( Figure 1A). However, viral replication complexes assembled from the full-length viral RdRp encoded by CMV-D2b may prevent RDR1 access to the nascent replication products, leading to the absence of RDR1-dependent production of viral secondary siRNAs. The accumulation of CMV-2aTD2b was slightly lower than that of CMV-D2b in wild-type, rdr1, and rdr1 rdr6 plants ( Figure 1C), indicating that the truncated viral RdRp encoded by CMV-2aTD2b may indeed be partially defective to support replication. (A) to (C) All available single, double, and triple mutants within the AGO4 clade (A), the AGO1 clade (B), or the AGO2 clade (C) were challenged by CMV-D2b, and the accumulation of virus RNAs was analyzed 3 weeks after inoculation by RNA gel blot hybridization. wt, wild type. (D) No additive effect was detected between the ago1 allele and the ago3 and ago7 alleles from the AGO2 clade.

Effector Mechanisms of Antiviral Silencing in Plants
Among ago1, ago2, ago3, ago4, ago5, ago6, ago7, ago9, and ago10 single mutants, only ago1 and ago2 mutants exhibited enhanced disease susceptibility to CMV-D2b (Figures 3 to 5), demonstrating independent function of AGO1 and AGO2 in defense against CMV-D2b. However, viral siRNAs including viral secondary siRNAs accumulated to high levels in ago1, ago2, and ago1 ago2 mutants infected with CMV-D2b ( Figures 5A and 5C). These findings indicate that dicing of the viral precursor RNAs of siRNAs in absence of AGO1, AGO2, or both is insufficient to confer resistance and that AGO1 and AGO2 are essential for the antiviral activities of viral siRNAs downstream siRNA biogenesis. Coimmunoprecipitation with AGO1-and AGO2-specific antibodies followed by Illumina sequencing and gel blot hybridizations ( Figure 6) revealed that in wild-type plants, both AGOs bind in vivo viral siRNAs, including viral secondary siRNAs. Further analysis showed that viral siRNAs with 59 terminal uridine and adenosine are recruited preferentially by AGO1 and AGO2, respectively (Figure 6), as found previously for endogenous small RNAs (Mi et al., 2008;Montgomery et al., 2008;Takeda et al., 2008). Interestingly, whereas approximately equal ratios of the positive-and negative-strand viral siRNAs are detected in the AGO2 complex and in the total viral siRNAs, viral siRNAs loaded in AGO1 exhibited strong bias for the positive strands ( Figure 6). It is known that ago1 mutants are hypersensitive to virus infection and AGO1 binds to viral siRNAs in the infected cells (Morel et al., 2002;Zhang et al., 2006;Qu et al., 2008). A previous study detected loading of viral siRNAs in plants overexpressing AGO2 but did not demonstrate an antiviral function for AGO2 in wildtype plants (Takeda et al., 2008). During revision of our manuscript, Harvey et al. (2011) reported similar enhanced disease phenotypes of the single ago2 mutant plants to wild-type viruses and AGO2 loading of viral siRNAs.
Further examination of 16 double and triple mutants generated both within and between the three clades of AGOs revealed additive effects of ago1 and ago2 in enhancing the symptom severity and virus accumulation (Figures 3 to 5) but not between ago3, ago4, ago5, ago6, ago7, ago9, and ago10 or in combination with ago1 and ago2. In particular, ago1 ago5 mutants behaved like ago1 toward CMV-D2b (Figure 4), and ago1 ago7, ago2 ago7, and ago1 ago2 ago7 mutants behaved like ago1, ago2 single, or double mutants (Figures 4 and 5). These results suggest that AGO5 and AGO7 have no detectable antiviral role against CMV-D2b in wild-type plants even through previous studies detected the binding of CMV siRNAs to AGO5 overexpressed in plants and an enhanced accumulation of turnip crinkle virus in both ago1 and ago7 single mutants (Qu et al., 2008;Takeda et al., 2008). Notably, CMV-D2b accumulated to significantly higher levels in the ago1 ago2 mutant than in rdr6 and sgs3 mutants (Figure 5), suggesting that AGO1 and AGO2 mediate antiviral activities by both viral primary and secondary siRNAs.
RDR-dependent RNA silencing in C. elegans and Arabidopsis sometimes requires sequential activities of AGOs to bind primary and secondary siRNAs, respectively (Yigit et al., 2006;Montgomery et al., 2008). For example, AGO7 associates with miR390 to target TAS3 single-stranded RNA that subsequently produce ta-siRNAs that associate with AGO1. As a result, TAS3 ta-siRNAs are not detectable in Arabidopsis ago1 or ago7 null alleles (Adenot et al., 2006;Montgomery et al., 2008). However, abundant CMV siRNAs were detected in both the single and double ago1/ago2 mutants, suggesting that AGO1 and AGO2 may not act sequentially in the biogenesis of viral secondary siRNAs. It should be pointed out that a hypomorphic ago1 allele was used in this work because ago1 null mutants exhibit pleiotropic developmental defects that make virus inoculation difficult (Morel et al., 2002). This hypomorphic ago1 mutant retains partial AGO1 activity (Baumberger and Baulcombe, 2005), which could target CMV RNAs and is also expected to overaccumulate AGO2 because of reduced regulation of AGO2 mRNA by miR403. Thus, an essential role for AGO1 in the biogenesis of viral secondary siRNAs cannot be ruled out. Alternatively, other AGOs, such as AGO5 (Takeda et al., 2008), may act redundantly in the biogenesis of viral secondary siRNAs without being able to target virus RNAs for silencing.
Future studies are necessary to understand why AGO1 and AGO2 act in a nonredundant and cooperative manner in antiviral defense. Nevertheless, it could be related to one or more of the following different properties of AGO1 and AGO2. First, AGO1 and AGO2 associate to different sets of viral siRNAs with distinct 5-terminal nucleotides ( Figure 6B). This distinction is important at least for the activities of miRNAs because replacing the 59 terminal uridine of a miRNA by an adenosine redirects the modified miRNA from AGO1 to AGO2 and impairs the regulation of its target (Mi et al., 2008). Whether AGO2/siRNA complexes can execute mRNA cleavage or translational repression like AGO1/ siRNA complexes remains to be determined. Second, AGO1 and AGO2 appear to exhibit different affinities to positive-to negative-strand viral siRNAs ( Figure 6B). Thus, AGO1 and AGO2 may have different antiviral activities because negativestrand siRNAs are more active than positive-strand siRNAs in antiviral silencing against positive-strand RNA viruses (Pantaleo et al., 2007;Schubert et al., 2007). Third, different ratios of 22-to 21-nucleotide viral siRNAs associate with AGO1 and AGO2, which could influence the overall activity of AGO1/siRNA and AGO2/siRNA complexes (see below).

22-Nucleotide Viral Secondary siRNAs Produced by DCL2 Do Not Mediate Efficient Antiviral Defense
Several lines of evidence indicate that Arabidopsis resistance to CMV-D2b is conferred by the viral secondary siRNAs made by DCL4. The 21-nucleotide viral siRNAs were the most dominant species of viral siRNAs detected in the wild-type, rdr1, and dcl2 plants, which were resistant to CMV-D2b and supported production of viral secondary siRNAs targeting CMV-D2b ( Figure 2). By comparison, the accumulation of 21-nucleotide viral siRNAs was markedly reduced in the highly susceptible plants, such as rdr6, rdr1 rdr6, and sgs3 mutants, none of which supported production of viral secondary siRNAs targeting CMV-D2b ( Figure  2). Importantly, loss of 21-nucleotide viral siRNAs in the dcl4 mutant abolished resistance to CMV-D2b, whereas preventing production of 22-nucleotide viral siRNAs in the dcl2 mutant did not decrease resistance to CMV-D2b ( Figure 2).
RNA gel blot hybridizations showed that 22-nucleotide viral siRNAs in fact accumulated to higher levels in the dcl4 mutant than 21-nucleotide viral siRNAs in the dcl2 mutant ( Figure 2A). Importantly, the dcl4 mutant supported production of the RDR6dependent hot spots of siRNAs targeting the 4th region of RNA3, indicating that when DCL4 was not functional, DCL2 was able to recognize and process the RDR6-dependent dsRNA precursor into secondary siRNAs. As a result, viral secondary siRNAs were undetectable only in the dcl2 dcl4 mutant ( Figure 2C). Therefore, abundant production of viral primary and secondary 22-nucleotide siRNAs by DCL2 in the dcl4 mutant failed to confer effective antiviral defense in absence of the DCL4-processed 21-nucleotide siRNAs. Nevertheless, CMV-D2b reproducibly accumulated to significantly higher levels in the dcl2-dcl4 mutant than in dcl4, sgs3, and rdr6 mutants (Figures 2 and 5; P values < 0.001). This suggests that DCL2-processed 22-nucleotide viral siRNAs direct silencing of CMV-D2b in the dcl4 mutant with similar efficiency to the silencing of CMV-D2b by the RDR6-independent viral primary siRNAs found in rdr6 and sgs3 mutants.
A previous study found that a VSR-deficient mutant of turnip crinkle virus is partially rescued in the dcl4 mutant but not in the dcl2 mutant and that increasing the inoculum strength could bypass the antiviral effects of DCL2 but not those of DCL4 (Deleris et al., 2006). Although the study did not differentiate viral primary and secondary siRNAs, the results are consistent with the qualitative difference in antiviral defense observed in this work between 21-and 22-nucleotide viral secondary siRNAs processed by DCL4 and DCL2, respectively. The same qualitative difference is also known for siRNAs that mediate non-cellautonomous transgene silencing (Dunoyer et al., 2005). Indeed, when dsRNAs are specifically expressed in the phloem, spreading of RNA silencing in the adjacent cells depends on DCL4-processed 21-nucleotide siRNAs, whereas 22-nucleotide siRNAs processed by DCL2 in the dcl4 mutant are ineffective. By contrast, both DCL4-and DCL2-processed secondary siRNAs direct effective RNA silencing when inducing transgenes are expressed constitutively (Fusaro et al., 2006;Moissiard et al., 2007;. Therefore, the efficiency of siRNAs may depend on their site of production and/or their route for spreading. Alternatively, the qualitative difference between 21-and 22-nucleotide secondary siRNAs in RNA silencing could be mechanistically related to the distinct properties of 21-and 22-nucleotide miRNAs loaded in AGO1. Indeed, although both species of miRNAs guide AGO1 to cleave the target transcripts, only 22-nucleotide miRNAs trigger RDR6-dependent synthesis of the dsRNA precursor of ta-siRNAs Cuperus et al., 2010). It is possible that efficient and multiple rounds of RNA cleavage are directed only by AGO1/AGO2 complexes loaded with 21-nucleotide siRNAs/miRNAs (Hutvá gner and Zamore, 2002). By contrast, AGO1/AGO2 complexes loaded with 22-nucleotide siRNAs/miRNAs either direct inefficient cleavages or fail to efficiently release the cleaved products of the target transcripts, leading to the recruitment of RDR6 and SGS3.

Viruses and Infection Assays
The mutant virus CMV-2aTD2b was described as CMVf-D2b previously (Wang et al., 2010). CMV-D2b was obtained by introducing three missense point mutations that abolished three ATG codons of 2b open reading frame at nucleotides 2420 (the start codon of 2b), 2441, and 2471. Purified virions propagated in Nicotiana clevelandii were used as the inocula at the concentration of 20 mg/mL.

RNA Gel Blotting and Immunoblotting Analyses
The systemically infected leaves of 20 to 30 plants were pooled for RNA extraction three weeks after inoculation. As described previously, high (5 mg) and low (15 mg) molecular RNAs were used to detect viral RNAs and siRNAs, respectively (Wang et al., 2010). The probe for viral RNA detection was the randomly labeled cDNA corresponding to the 39 terminal 240 nucleotides of RNA2. The small RNAs were detected by the labeled DNA oligonucleotides corresponding to the genomic RNA3 (nucleotides 241 to 280, 741 to 780, 1041 to 1080, 1341 to 1380, 1600 to 1640, 1681 to 1710, and 1731 to 1770). The gel blot hybridization assay to map the hot spots of viral siRNAs was performed as described previously (Wang et al., 2010). Briefly, an equal amount of each of the DNA fragments specific to RNA3 (divided into five fragments of 450 bp each) was fractionated and hybridized to the [ g-32 P]ATP-labeled total small RNAs harvested and gel purified from wild-type Col-0 and mutant Arabidopsis plants infected with CMV-D2b. The blot signals were detected by phosphor imager and multiple film exposures. The systemically infected leaves were used for immunoblot detection of the 2b protein using the polyclonal anti-2b of the Fny strain of CMV as described previously (Zhang et al., 2006). Cloning, sequencing, and analysis of viral siRNAs were performed as described (Wang et al., 2010).

Real-Time PCR
Real-time quantitative PCR was used to measure the accumulation of the positive and negative strands of the viral RNA2 in infected leaves using Arabidopsis Actin2 mRNA as the control for normalization. Briefly, 1 mg of total RNA extracted from the upper systemically infected leaves was used for reverse transcription using Super Script III reverse transcriptase (Invitrogen) and subsequent real-time quantitative PCR reactions using iQ SYBR Green Supermix (Bio-Rad) according to the manufacturer's instructions. Three independent repeats were performed for each experiment. The data analysis of relative expression was conducted using the 2 2DDCt method (Livak and Schmittgen, 2001).

Construction, Sequencing, and Analyses of Small RNA Libraries
Construction, Illumina sequencing, and analyses of small RNA libraries from the CMV infected plants were performed as described (Wang et al., 2010). To analyze the small RNA populations loaded in AGO1 and AGO2, libraries were constructed from total small RNAs purified from the leaves of FLAG-AGO1/ago1-36 and HA-AGO2/AGO2 plants systemically infected with CMV-D2b both before and after coimmunoprecipitation with FLAG-and HA-specific antibodies as described (Baumberger and Baulcombe, 2005). After sequencing, adapters' sequences were removed and the downstream analyses were performed with in-house pipelines. Perfect matches to the Arabidopsis and CMV genomes were extracted with Bowtie (Langmead et al., 2009) and considered as the total reads in each library. The Arabidopsis-specific reads were further classified into miRNAs, ta-siRNAs, transposable elements and repeats, coding regions and untranslated regions, and organellar (mitochondria and chloroplast) small RNAs. The fraction of small RNAs that did not match any of the above-mentioned categories was assigned to the "others" category, such that the sum of all categories equals the total number of reads in each library. The following databases were used: TAIR10 blast_datasets (ftp://ftp.Arabidopsis.org/home/tair/Sequences/), Repbase (http://www.girinst.org), ASBP (http://asrp.cgrb.oregonstate. edu), and miRBase (http://microrna.sanger.ac.uk/sequences).

Supplemental Data
The following materials are available in the online version of this article.