Roles and Programming of Arabidopsis ARGONAUTE Proteins during Turnip Mosaic Virus Infection

In eukaryotes, ARGONAUTE proteins (AGOs) associate with microRNAs (miRNAs), short interfering RNAs (siRNAs), and other classes of small RNAs to regulate target RNA or target loci. Viral infection in plants induces a potent and highly specific antiviral RNA silencing response characterized by the formation of virus-derived siRNAs. Arabidopsis thaliana has ten AGO genes of which AGO1, AGO2, and AGO7 have been shown to play roles in antiviral defense. A genetic analysis was used to identify and characterize the roles of AGO proteins in antiviral defense against Turnip mosaic virus (TuMV) in Arabidopsis. AGO1, AGO2 and AGO10 promoted anti-TuMV defense in a modular way in various organs, with AGO2 providing a prominent antiviral role in leaves. AGO5, AGO7 and AGO10 had minor effects in leaves. AGO1 and AGO10 had overlapping antiviral functions in inflorescence tissues after systemic movement of the virus, although the roles of AGO1 and AGO10 accounted for only a minor amount of the overall antiviral activity. By combining AGO protein immunoprecipitation with high-throughput sequencing of associated small RNAs, AGO2, AGO10, and to a lesser extent AGO1 were shown to associate with siRNAs derived from silencing suppressor (HC-Pro)-deficient TuMV-AS9, but not with siRNAs derived from wild-type TuMV. Co-immunoprecipitation and small RNA sequencing revealed that viral siRNAs broadly associated with wild-type HC-Pro during TuMV infection. These results support the hypothesis that suppression of antiviral silencing during TuMV infection, at least in part, occurs through sequestration of virus-derived siRNAs away from antiviral AGO proteins by HC-Pro. These findings indicate that distinct AGO proteins function as antiviral modules, and provide a molecular explanation for the silencing suppressor activity of HC-Pro.

RNA-based silencing is triggered by dsRNA that is processed by DCLs into 21-to 24-nt short interfering RNAs (siRNAs), which subsequently associated with AGO proteins to form the RNA-induced silencing complex (RISC) [7,8]. Inhibition of target RNA can occur by endonucleolytic cleavage ("slicing"), translational repression, or delivery of chromatin-modifying complexes to a locus [9][10][11]12]. In some cases, amplification of the silencing response occurs by triggering dsRNA synthesis and secondary siRNA accumulation [13].
Viruses are inducers of RNA silencing; infected plants accumulate large amounts of siRNAs derived from viral RNAs [1]. Most plant viruses encode one or more silencing suppressor proteins that interfere with antiviral RNA silencing [13,14]. One mechanism of silencing suppression by viral suppressors is through sequestration of siRNA duplexes [1], preventing assembly of the RISC effector complex. Other viral silencing suppressors promote AGO degradation [15][16][17][18][19], prevent slicing or degradation of target RNAs by associating with AGOs [20,21], or use other mechanisms (for a recent review see Nakahara and Masuta 2014 [22]). In effect, viral suppressors mask the effects of antiviral silencing, making genetic analysis of antiviral silencing factors in host plants dependent on the use of suppressor-deficient viruses [3,4,6,23].
A. thaliana has ten AGO genes [24], of which AGO1, AGO2 and AGO7 have been implicated in antiviral defense against various viruses by genetic and biochemical criteria [6,[25][26][27][28][29][30][31]. Antiviral roles for AGO3 and AGO5 have also been suggested based on virus-derived siRNA association and/or in vitro analyses [8,32]. One model for AGO antiviral activity states that AGO proteins bind virus-derived siRNAs and directly repress viral RNA through slicing, translational repression, or other mechanisms [2,8,33]. Given that AGO-dependent regulation of gene expression affects numerous biological processes, including DNA repair [34], AGO proteins might also affect virus replication indirectly through regulation of genes with roles in defense. For example, AGO2-miR393 Ã complexes regulate the expression of MEMBRIN 12 (MEMB12), which is required for resistance to Pseudomonas syringae in A. thaliana [35]. Moreover, some AGO proteins are known to modulate the activity of other AGO proteins [36,37], which could affect AGOs with roles in antiviral defense.
Potyviral HC-Pro is a suppressor of RNA silencing. As shown using potyviruses like Turnip mosaic virus (TuMV) [23,38], the counter-defensive function of HC-Pro is necessary for establishment of infection or systemic spread. HC-Pro has been proposed to function through sequestration of virus-derived siRNAs [39][40][41][42][43][44]. HC-Pro may also function through physical interaction with factors like the transcription factor RAV2 [45], translation initiation factors eIF(iso)4E and eIF4E [46], calmodulin-related protein (CaM) [47], auxiliary proteins like Heat Shock Protein 90 (HSP90) [48], and/or through effects on downstream defense or silencing factors [49,50]. Here, the role of several A. thaliana AGOs in antiviral defense against TuMV was analyzed in various organs of systemically infected plants. The impact of HC-Pro on the loading of antiviral AGOs with virus-derived siRNAs was also studied.
Parental TuMV-GFP was detected in inoculated leaves and noninoculated inflorescences of all single ago mutants analyzed (Table 1 and Fig. 1B). Local infection of single ago1 mutants was significantly lower than that of wild-type Col-0 (Fig. 1B), but this was likely due to the difficulty of inoculating the smaller leaves of hypomorphic mutants containing ago1 alleles.

AGO1 and AGO10 have modest antiviral effects in inflorescences
To determine if the major effect of AGO2 was additive with the minor effects of AGO5, AGO7 and AGO10, and to examine if AGO1 possessed redundant or masked activities, double and triple ago mutant plants were inoculated with TuMV-GFP or TuMV-AS9-GFP, and virus accumulation was measured in inoculated and noninoculated organs as described above. To reduce the effect of differences in leaf size, we planted mutant lines with the ago1-27 allele one week earlier than the other mutant lines inoculated at the same time. Parental TuMV-GFP infected locally ( Fig. 2A panels I and II) and moved systemically into the inflorescence of all double and triple ago mutants analyzed (Tables 2 and 3), with no significant differences in infection efficiency.
In double mutants harboring the ago2-1 allele and one of ago5-2, zip-1, or ago10-5 alleles, no significant differences in number of infection foci were detected at 7 dpi in rosette leaves inoculated with TuMV-AS9-GFP ( Fig. 2A panel I and Table 2). Similarly, no significant differences were detected in TuMV-AS9-GFP coat protein accumulation in cauline leaves at 15 dpi ( Fig. 2B panel I). As observed for the ago single mutants, TuMV-AS9-GFP was not detected in inflorescences from double mutant plants containing the ago2-1 allele (Fig. 2B panel I). These results indicate that the minor activities of AGO5, AGO7 and AGO10 are not additive with the major antiviral activity of AGO2. Double and triple mutants harboring the ago1-27 allele were generated and inoculated with parental TuMV-GFP or suppressor-deficient TuMV-AS9-GFP. Col-0 plants and ago1-27, ago2-1 and ago10-5 single mutant lines were included as controls. Local TuMV-AS9-GFP infection foci were observed in inoculated rosette leaves, and virus was detected in noninoculated cauline leaves, from ago1-27 ago2-1 double mutant plants, but ago1-27 had no enhancing or suppressing effects when combined with ago2-1 (panel II in Fig. 2A and 2B, Table 3). Combining ago1-27 with ago10-5, or with ago2-1 and ago10-5 in a triple mutant, had no effects on local TuMV-AS9-GFP infection foci ( Fig. 2A panel II) or accumulation in cauline leaves beyond those measured in the single ago2 or double ago2 ago10 mutants ( Fig. 2B panels I and II, and Table 3). However, combining ago1-27 with ago10-5 resulted in an increase in TuMV-AS9-GFP CP accumulation in cauline leaves relative to single ago10-5 mutants (Fig. 2B panel II). Infection efficiency of ago1 single, double or triple mutants by TuMV-GFP was similar to that of wild type plants ( Fig. 2A panels II and III), and infection efficiency of ago1-27 ago2-1 double and ago1-27 ago2-1 ago10-5 triple mutants by TuM-V-AS9-GFP was similar to that of dcl2-1 dcl3-1 dcl4-2 plants used as susceptible control ( Fig. 2A panel II). Thus, both the lack of TuMV-AS9-GFP infection in single ago1 mutants and the lack of systemic infection of inflorescence in ago1-27 ago2-1 double mutants were not due to pleiotropic effects.
Collectively, the genetic analysis of local and systemic infection using TuMV-AS9-GFP revealed two sets of AGOs that limit infection. In inoculated rosette and noninoculated cauline leaves, AGO2 plays a major antiviral role, while AGO5, AGO7 and AGO10 play minor roles that are non-additive with AGO2. In noninoculated inflorescence tissues, AGO1 and AGO10 play overlapping or redundant antiviral roles, but these functions likely account for only a fraction of the RNA-mediated antiviral activity. It is possible that other factors, including AGO proteins not analyzed here, have a role in protecting inflorescence tissue from virus infection. The scope of subsequent AGO analyses was restricted to the functions of AGO1, AGO2 and AGO10 in the presence and absence of functional HC-Pro.
Differential association of AGO2 with viral siRNAs in the presence and absence of functional HC-Pro We hypothesized that AGO proteins with anti-TuMV activity associate with TuMV-derived siRNAs. This idea was tested first with epitope-tagged AGO2 in plants inoculated with parental TuMV or HC-Pro-defective TuMV-AS9 (lacking GFP) [23]. AGO2 immunoprecipitation and small RNA sequence analyses were done using transgenic A. thaliana expressing a triple-hemagglutinin (HA) epitope-tagged, catalytically inactive form of AGO2 (HA-AGO2 DAD ). The second of three aspartic acid residues of AGO2 was substituted with alanine; this substitution eliminates antiviral activity of AGO2, but preserves both the siRNA-binding and target RNAbinding functions [31]. These experiments require the use of plants lacking AGO2-mediated antiviral functions, as infection by TuMV-AS9 would otherwise be blocked (Figs. 1 and 2) [31]. Small RNAs from the input (pre-immunoprecipitated) and HA-AGO2 DAD co-immunoprecipitated fractions from inoculated rosette leaves and noninoculated inflorescences of TuMVinfected plants were analyzed from duplicate biological samples. Only reads that matched to either the A. thaliana or TuMV genomes without mismatches were analyzed (S1 Table). For each individual sample, read counts were scaled with respect to the total number of adaptorparsed reads (reads per million) for the corresponding flow cell (eight individual samples). In mock-inoculated plants, a small number of reads from the input fractions mapped to TuMV (S1-S4 Tables, and S1 Fig). The source of these reads could be contamination, sequencing error, or portions of the A. thaliana genome. Based on the number of reads from mock-inoculated plants mapping to the TuMV genome, the false positive rate (proportion of parsed reads artifactually mapping to TuMV) was estimated to be between 9.8X10 -6 and 1.0X10 -4 , which should not have affected subsequent analyses.
In input fractions from TuMV-infected plants expressing HA-AGO2 DAD , the proportion of reads mapping to the A. thaliana genome, as opposed to TuMV, varied from 77% (averaged across replicates) to 84% for different tissues (S1A Micro-RNA read counts for input and immunoprecipitates from this and subsequent analyses are provided in S1 Dataset. MiR390 and miR393 Ã were shown previously to co-immunoprecipitate with AGO2 [35,52]. In mock-inoculated and TuMV-infected rosette leaves, the number of miR390 reads in HA-AGO2 DAD immunoprecipitates was 260 and 65 fold higher, respectively, than in the corresponding input samples. Similarly, miR393 Ã reads were enriched 125 and 60 fold in HA-AGO2 DAD immunoprecipitates from mock-inoculated and TuMV-infected rosette leaves, respectively. Therefore, enrichment of A. thaliana small RNA populations that are known to be associated with AGO2 occurred as expected.  In leaves of TuMV-AS9-infected plants, endogenous A. thaliana small RNAs were again enriched (2.7 to 4.6 fold) in HA-AGO2 DAD immunoprecipitates, with patterns expected of AGO2-associated small RNAs (S2A Fig). Virus-derived siRNAs represented 7% or 16% of mapped reads in input samples from inoculated rosette leaves or systemically infected cauline leaves, respectively (S1A Fig). However, in striking contrast to TuMV-infected samples, both 21-and 22-nt TuMV-AS9-derived siRNAs were highly enriched relative to TuMV-derived siRNAs in HA-AGO2 DAD immunoprecipitates from both inoculated rosette leaves and systemically infected cauline leaves (Fig. 3A panels I and II, Fig. 3B and 3C panels III and IV, and S3 Fig). Among co-immunoprecipitated siRNAs, those containing a 5'A were overrepresented (Fig. 3A panel III). Association of AGO2 with siRNAs derived from TuMV-AS9, but not from TuMV, was verified by small RNA northern blot assays (S9 Fig). These results indicate that programming of AGO2 with TuMV-derived siRNAs is inhibited in the presence of active HC-Pro.

Differential association of AGO1 and AGO10 with viral siRNAs in the presence and absence of functional HC-Pro
A similar experimental design was used to test the association of tagged AGO1 and AGO10 with TuMV and TuMV-AS9-derived siRNAs. To enable infection by suppressor-deficient TuMV-AS9, transgenic A. thaliana plants expressing catalytically defective HA-AGO1 DAH [31] or HA-AGO10 DAH were produced in the TuMV-AS9-permissive ago2-1 background. Phenotypic defects associated to catalytic mutant HA-AGO1 DAH were more severe in an ago1-25 mutant that in a wild-type (AGO1) background [31]. Effects of catalytically defective HA-AGO10 DAH on plant phenotype were not known, so transgenic A. thaliana plants expressing catalytically active HA-AGO10 DDH in a wild-type Col-0 background were also generated. Transgenic lines were inoculated with TuMV or TuMV-AS9 and samples from inoculated rosette leaves and systemically infected cauline leaves or inflorescences were collected from biological replicates. Small RNAs from input samples and immunoprecipitated fractions were sequenced, and reads were mapped and counts were scaled as described above. Tagged versions of AGO1 and AGO10 associated with small RNAs with a 5'U, as expected (S2B and S2C Fig  panel II) [36,[52][53][54], and the proportion of A. thaliana and TuMV-derived siRNAs (S1B and S1C Fig In mock-inoculated samples, endogenous A. thaliana 21-nt small RNAs were enriched 5 to 15 fold, and 5 to 7 fold, in HA-AGO1 DAH and HA-AGO10 DAH immunoprecipitates, respectively. In TuMV-and TuMV-AS9-infected samples, A. thaliana 21-nt small RNAs were enriched 5 and 15 fold, respectively, in HA-AGO1 DAH immunoprecipitates (S2B Fig panel I). In TuMV-infected samples, A. thaliana 21-nt small RNAs were enriched 1.5 and 2.5 fold in HA-AGO10 DDH immunoprecipitates from inflorescences and rosette leaves, respectively (S2C Fig panel I). In TuMV-AS9-infected samples, A. thaliana 21-nt small RNAs were enriched 7 fold in HA-AGO10 DAH immunoprecipitates from cauline leaves (S2C Fig panel I). Sequences with a 5'U were enriched with both AGOs (panel II in S2B and S2C Fig), as expected [36,[52][53][54]. MiRNAs were enriched in HA-AGO1 DAH and HA-AGO10 DAH immunoprecipitates from both mock-inoculated (7 to 50 fold) and TuMV-infected (3 to 25 fold) samples, while miRNA Ã and tasiRNA populations were variable (S8B and S8C Fig). For example, miR166 reads were enriched 30 and 45 fold in HA-AGO1 DAH immunoprecipitates from inflorescences of mock-inoculated and TuMV-infected plants, respectively. MiR168 reads were likewise enriched 20 and 12 fold. MiR166 reads were enriched 900 and 60 fold in HA-AGO10 DAH immunoprecipitates from mock-inoculated and TuMV-infected plants, respectively, in agreement with previous observations [36]. In rosette and inflorescence tissues from each of the transgenic lines, TuMV infection triggered abundant 21-and 22-nt siRNAs that originated from sense and antisense strands across the entire viral genome (Figs. 4B and 5B). However, as with HA-AGO2 DAD immunoprecipitates, TuMV-derived siRNAs were depleted in both HA-AGO1 DAH (Fig. 4A-4C (Fig. 5A panels I and II, Fig. 5B and 5C panels III, and S5 Fig), and had predominantly a 5'U nucleotide (Fig. 5A panel III). Individual highly enriched sequences were distributed across the TuMV-AS9 genome (Fig. 5C  panel III and S5 Fig), suggesting that AGO10 may target all regions of TuMV-AS9 genome. TuMV-AS9-derived siRNAs were present in HA-AGO1 DAH immunoprecipitates at a higher level than in immunoprecipitates from plants infected with parental TuMV, although the overall population of TuMV-AS9-derived siRNAs was depleted relative to the input fraction ( Fig. 4A panels I and II, Fig. 4B and 4C panel III, and S4 Fig). Only a few individual sequences were enriched; these sequences had predominantly a 5'U nucleotide (Fig. 4A panel III). Because depletion of TuMV-AS9-derived siRNAs in HA-AGO1 DAH immunoprecipitates was 60 to 1,200 fold lower than in TuMV-infected samples, we reasoned that AGO1 does interact with virus-derived siRNAs, but to a lesser extent than both AGO2 and AGO10.

HC-Pro associates with siRNAs derived from the entire TuMV genome
Results described above show that AGO1, AGO2 and AGO10 associate at low levels with parental TuMV-derived siRNAs. In contrast, AGO2 and AGO10, and to a much lesser extent AGO1, associate with siRNAs derived from the suppressor-deficient TuMV-AS9 genome. Only two residues (R238A and V240A) in HC-Pro differ between TuMV and TuMV-AS9 ( Fig. 6A panel I) [23,38]. We hypothesized that i) HC-Pro associates with siRNAs-derived from the entire TuMV genome and sequesters them from AGO proteins, and ii) the AS9 mutation in HC-Pro reduces siRNA-binding activity. HC-Pro is known to have small RNA-binding activity [39,43,44,55], but the extent to which it binds siRNAs in the context of TuMV infection has not been described. To measure the extent to which HC-Pro binds small RNA using the immunoprecipitation assay, we introduced an N-terminal 6xHistidine tag (HIS 6 ) in the context of the TuMV (TuMV-HIS) and TuMV-AS9 (TuMV-HIS-AS9) genomes ( Fig. 6A panel I). The addition of HIS 6 to HC-Pro did not affect viral coat protein accumulation ( Fig. 6A panel II), but enabled specific immunoprecipitation of HC-Pro from plants infected with TuMV-HIS and TuMV-HIS-AS9 (Fig. 6B).
Small RNAs from input and immunoprecipitated fractions obtained from plants inoculated with TuMV-HIS and TuMV-HIS-AS9 were sequenced. Because TuMV-HIS-AS9 accumulated more slowly than TuMV-HIS, TuMV-HIS samples were collected earlier than TuMV-HIS-AS9 samples (10 and 15 dpi, respectively), and twice as much input and immunoprecipitate materials for TuMV-HIS-AS9 samples were analyzed. The longer infection time and doubling of materials for TuMV-HIS-AS9 resulted in similar protein levels for HIS-HC-Pro and HIS-HC-Pro-AS9 input and immunoprecipitate fractions (Fig. 6B).
In contrast with results obtained for HA-AGO1 DAH , HA-AGO2 DAD and HA-AGO10 DDH from TuMV-infected plants (compare panel I in Fig. 3C-5C to Fig. 6E), TuMV-derived siRNAs were highly enriched in HIS-HC-Pro immunoprecipitates from cauline leaves and inflorescence (Fig. 6C panels I and II, and Fig. 6D and 6E panels I and II). No 5' nt preference was evident ( Fig. 6C panels III and IV). HIS-HC-Pro associated preferentially with 21-nt over 22-nt siRNAs in samples from both cauline leaves and inflorescences (Fig. 6C, 6D-E panels I and II, and S7 Fig). In contrast, TuMV-HIS-AS9-derived siRNAs from across the genome were depleted in the HIS-HC-Pro-AS9 immunoprecipitates from systemically infected cauline leaves; only a few individual sequences were enriched (Fig. 6C panels I and II, 6D and 6E panel III, and  S7 Fig). These results indicate that wild-type HC-Pro associates with TuMV-derived siRNAs, and that the AS9 mutation disrupts this association. We concluded that HC-Pro interferes with antiviral silencing, at least in part, by sequestering TuMV-derived siRNAs and preventing their association with antiviral AGO proteins. Suppression activity of HC-Pro is not tissue specific and affects AGO1, AGO2, AGO10 and possibly other AGO proteins.

Discussion
Genetic and co-immunoprecipitation analyses were combined to reveal that i) several AGOs function as anti-TuMV defense modules in A. thaliana, ii) viral siRNAs generally fail to load into AGO proteins with antiviral functions during wild-type TuMV infection, and iii) HC-Pro sequesters viral siRNA away from AGOs with antiviral functions.

Functions of AGO-small RNA complexes in anti-TuMV defense
AGO proteins target endogenous transcripts to regulate plant development and innate immunity [2,56], which may indirectly affect susceptibility to viruses. It is likely, however, that at least some AGO proteins with an antiviral role are programmed with virus-derived siRNA to directly target viral RNA [8,10,57,58]. The genetic analysis described here revealed several AGO proteins that participate in modular fashion during anti-TuMV defense (Fig. 7). AGO2 has the most influential role in protecting inoculated rosette and cauline leaves (Fig. 1), while AGO1 and AGO10 have genetically redundant roles in protecting inflorescence tissues. A larger proportion of ago1 ago2 ago10 triple mutants than ago1 ago10 double mutants were systemically infected (Table 3), perhaps suggesting that AGO2 also contributes to restricting virus spread to inflorescences.
The antiviral effects of different AGO proteins in different tissues may depend on a number of factors, including expression patterns, AGO-interacting partners, small RNA binding preferences, or subcellular localization. Microarray data suggest that AGO10 and AGO1 are expressed more strongly than AGO2 in flowers and meristems [59]. However, AGO1 and AGO10 transcript levels are also higher than AGO2 transcript levels in rosette leaves. Therefore, expression levels alone do not explain the effectiveness of individual AGOs in different organs. It is conceivable that modular, tissue-specific functionality is controlled by AGO-interacting or AGOpromoting factors that are tissue-specific. In ago1 ago10 double mutants, systemic infection of inflorescences could be partially restricted because AGO2 limits virus accumulation in leaves, acts directly in inflorescences, or functions in both of these tissues.
Direct down-regulation of viral RNA requires that AGOs bind virus-derived siRNAs (or endogenous small RNAs complementary to a given viral genome) and then viral RNA, followed by slicing of the viral RNA, repression of translation, and/or recruitment of factors for silencing amplification. Results described here show that AGO2, AGO10 and at much lower levels AGO1 associate with TuMV-AS9-derived siRNA in the absence of HC-Pro (Fig. 3C panels III and IV, and Figs. 4C and 5C panel III). AGO2-mediated slicing of viral RNAs could be a significant anti-viral mechanism, as catalytically defective forms of AGO2 lack anti-TuMV activity [31]. Evidence of direct targeting of TuMV RNA by AGO1 and AGO10 is lacking. In other studies, AGO1 was reported to bind small RNAs derived from Turnip yellow mosaic virus and CMV strains Fny and NT9 [20], but not CMV strain I17F or Crucifer-infecting tobamovirus [60]. The basis for differential interaction of TuMV-derived siRNAs and AGO1, AGO2 and AGO10 is not clear. It is possible that different AGOs have privileged access to viral siRNAs. In this context, AGO1 pools may have limited access to viral siRNAs during TuMV infection.
In inoculated rosette leaves of ago2 mutant and dcl2 dcl3 dcl4 triple mutant plants, TuM-V-AS9 accumulated to comparable levels ( Figs. 1 and 2). In contrast, accumulation of TuM-V-AS9 was consistently lower in cauline leaves and inflorescences of all ago mutants tested, Restriction of TuMV by ARGONAUTES including the ago1 ago2 ago10 triple mutant, compared to the respective tissues in dcl2 dcl3 dcl4 mutant plants. If it is assumed that all small RNA-mediated antiviral activity is lost in the dcl triple mutant, then it is reasonable to conclude that all antiviral silencing in inoculated rosette leaves is mediated by AGO2. The far greater effect of the dcl mutations, relative to the ago mutations, in systemic tissues, especially inflorescences, argues that the combined effects of AGO1, AGO2 and AGO10 account for only a small proportion of overall anti-TuMV silencing activity. This could indicate that other AGO proteins that were not tested here, or that were not tested in the right genetic combinations, play specific roles in systemic tissues. It could also mean that DCL proteins play a more dominant, direct antiviral role in systemic tissues, as suggested by genetic analyses with CMV [4,29], BMV [30], PVX [27], Tobacco rattle virus [61], TCV [3,6,33,62], Cauliflower mosaic virus, Cabbage leaf curl virus, and Oil rape mosaic virus [63].
Different antiviral AGO proteins may also have distinct effects on amplification of secondary, virus-derived siRNAs, which may be important for production of systemic signals [2,7,13,64]. Full anti-TuMV silencing requires both RDR1 and RDR6 [23], presumably for production of dsRNA from viral RNA. If this occurs like dsRNA formation during tasiRNA biogenesis, then RDR proteins may be recruited to viral RNA after targeting by AGO-small RNA complexes [52,[65][66][67][68]. Given the role of AGO1-small RNA complexes in triggering formation of several families of tasiRNA, AGO1 could conceivably play a trigger role for secondary viral siRNA.
The interpretation of ago1 mutant susceptibility experiments is challenging because of the pleiotropic developmental phenotypes of ago1 hypomorphic mutants and the large number of genes that are dysregulated when AGO1 is disrupted. In particular, disruption of AGO1-miR403 activity increases AGO2 mRNA and protein levels [26,69], which could result in a net increase in virus resistance, even if AGO1 directly targets viral RNA.
Other AGOs might also have indirect roles in anti-TuMV defense, perhaps by affecting expression of defense-related genes [35,56,70]. Expression of potyviral HC-Pro [45], infection with TCV [26], and infection with Pseudomonas syringae [35] result in increased AGO2 expression; AGO2 regulates expression of MEMB12 [35] and possibly other genes. AGO2 also associates with virus-activated endogenous siRNAs [56]. The significance of AGO2-dependent gene regulation for virus infection, if any, is not yet clear.

Suppression of antiviral silencing by HC-Pro
Multiple virus-encoded suppressors of RNA silencing target AGO1 [16,17,20,21,33,60], and P25 from PVX interact with AGO2, AGO3 and AGO4 [17] although the biological significance of this interaction remains to be elucidated. During TuMV infection, no evidence was obtained to indicate that AGO1, AGO2 or AGO10 were destabilized or otherwise down-regulated. Each AGO accumulated to normal levels.
TuMV-infected plants accumulate large amounts of virus-derived siRNAs that map across the entire genome (Figs. 3B, 4B, 5B, 6B, and S3-S5 Figs) [23], and co-immunoprecipitation and high-throughput sequencing showed that HC-Pro associates with viral siRNAs in leaf and inflorescence tissue (Fig. 6E panels I and II). Viral siRNAs associate with HC-Pro without a 5' nt preference (Fig. 6C panels III and IV). Importantly, HC-Pro was shown to sequester viral siRNAs away from AGO1, AGO2 and AGO10 (Figs. 3C, 4C and 5C panels I and II), leading to the obvious proposal that HC-Pro interferes with antiviral silencing by preventing AGOs from loading with virus-derived siRNAs (Fig. 7). Mutant HC-Pro-AS9 is deficient in associating with viral siRNAs (Fig. 6C-E panels III, and S6 Fig), and concomitantly loses silencing suppression activity.
The basis for sequestration of siRNAs by HC-Pro is not yet clear. HC-Pro may outcompete AGOs for siRNAs. Alternatively, HC-Pro may intercept viral siRNAs prior to AGO loading, perhaps due to subcellular localization properties. Further analyses will be necessary to resolve this issue.

DNA plasmids
Recombinant plasmids were made as follows.

Plant materials
All Arabidopsis thaliana plants used in this study (including mutant lines and transgenic lines) descended from the Columbia-0 (Col-0) accession, and were grown under long day (16 h light/ 8 h dark) at 22°C. The following single mutant lines were described before: ago1-25 and ago1-27 [25], ago2-1 [72], ago3-2 [32], ago4-2 [73], ago5-2 [32], ago6-3 [32], zip-1 [74], ago8-1 [32], and ago9-5 (SALK_126176). T-DNA insertion mutant GABI_818H06 (ago10-5) was obtained from The GABI KAT project [75]. Homozygous mutants were confirmed by PCR-based genotyping using a three-primer reaction: one on the left border, one in the flanking DNA, and one in the T-DNA insertion site [76]. Lack of AGO10 expression in homozygous plants was confirmed by RT-PCR using oligos AGO10_qF (GGTATTCAGGGAACAAGCAG) and by electrophoresis. The area from 16 to 26 nt was sliced and used for small RNA purification. 30 ng of small RNAs were used to make the libraries from total fraction. 50% of the immunoprecipitated RNA was used without fractionation to make libraries from immunoprecipitate fractions. For each treatment, small RNA libraries were made independently from two biological replicates. Bar-coded PCR amplification primers were used for multiplexing purposes. Eight individual samples were multiplexed and run in a single flow cell.

Bioinformatic analysis of small RNA libraries
Bioinformatic analysis of endogenous and TuMV-derived siRNAs was as described [23,31,80]. After removing 5' and 3' adaptors, sequences were aligned to the A. thaliana genome and to the TuMV genome. Only sequences with a perfect match were used for downstream analysis. For each sample, reads were normalized per 1,000,000 total reads (RPM), including all size classes. Enrichment with respect to the immunoprecipitate was calculated as the ratio of reads in the immunoprecipitate to reads in the input, and expressed on a log 2 scale.

Accession numbers
Sequence data from this article can be found in Gene Expression Omnibus (GEO, http://www. ncbi.nlm.nih.gov/geo) accession number GSE64911.