Arabidopsis RNA-dependent RNA polymerases and dicer-like proteins in antiviral defense and small interfering RNA biogenesis during Turnip Mosaic Virus infection.

Plants respond to virus infections by activation of RNA-based silencing, which limits infection at both the single-cell and system levels. Viruses encode RNA silencing suppressor proteins that interfere with this response. Wild-type Arabidopsis thaliana is immune to silencing suppressor (HC-Pro)-deficient Turnip mosaic virus, but immunity was lost in the absence of DICER-LIKE proteins DCL4 and DCL2. Systematic analysis of susceptibility and small RNA formation in Arabidopsis mutants lacking combinations of RNA-dependent RNA polymerase (RDR) and DCL proteins revealed that the vast majority of virus-derived small interfering RNAs (siRNAs) were dependent on DCL4 and RDR1, although full antiviral defense also required DCL2 and RDR6. Among the DCLs, DCL4 was sufficient for antiviral silencing in inoculated leaves, but DCL2 and DCL4 were both involved in silencing in systemic tissues (inflorescences). Basal levels of antiviral RNA silencing and siRNA biogenesis were detected in mutants lacking RDR1, RDR2, and RDR6, indicating an alternate route to form double-stranded RNA that does not depend on the three previously characterized RDR proteins.


INTRODUCTION
In plants and some animal lineages, such as insects, RNA silencing is a potent defense mechanism against viruses and has remarkable specificity and adaptability (Ding and Voinnet, 2007). To counter this defense mechanism, viruses encode suppressor proteins that interfere with RNA silencing. Antiviral silencing can be conceptualized into initiation, amplification, and systemic spread phases (Voinnet, 2005). Initiation consists of the recognition of the trigger RNA and formation of primary small interfering RNAs (siRNAs), while amplification is characterized by the synthesis of double-stranded RNA (dsRNA) by one or more RNA-dependent RNA polymerases and the formation of secondary siRNA. Systemic spread involves cell-to-cell and phloemdependent transport of a silencing signal (Ding and Voinnet, 2007).
Antiviral RNA silencing depends on some of the core factors that participate in the biogenesis and activity of endogenous siRNAs (Ding and Voinnet, 2007). DCL4 catalyzes formation of 21-nucleotide siRNAs from several RNA and DNA viruses (Blevins et al., 2006;Deleris et al., 2006;Fusaro et al., 2006;Diaz-Pendon et al., 2007). In the absence of DCL4, 22-and 24-nucleotide-long virus-derived siRNAs are produced by DCL2 and DCL3, respectively (Blevins et al., 2006;Deleris et al., 2006;Fusaro et al., 2006;Diaz-Pendon et al., 2007). DCL1 may play an indirect role as a negative regulator of DCL4 (Qu et al., 2008) and as a facilitator in the biogenesis of geminivirus-and caulimovirusderived siRNAs (Blevins et al., 2006;Moissiard and Voinnet, 2006). siRNA biogenesis or antiviral silencing have also been shown to be dependent on one or more of RDR1, RDR2, and RDR6 (Mourrain et al., 2000;Qu et al., 2005Qu et al., , 2008Schwach et al., 2005;Diaz-Pendon et al., 2007;Donaire et al., 2008;Qi et al., 2009;Wang et al., 2010). Systemic RNA silencing in Arabidopsis requires RDR1 and RDR6 for amplification of Cucumber mosaic virus (CMV)-derived sRNAs (Wang et al., 2010). RDR1 may couple with other defense responses because its expression is induced by salicylic acid (Ji and Ding, 2001;Xie et al., 2001;Yu et al., 2003). It is noted, however, that most studies to date do not clearly link virus-derived siRNA accumulation patterns and bona fide antiviral defense where the virus is actually suppressed or limited. When wild-type viruses are used, the activity of a virus-encoded silencing suppressor can mask the activity of silencing factors. Thus, wild-type plants often exhibit virus susceptibility phenotypes similar to those of mutants that lack RNA silencing factors (Dalmay et al., 2000;Mourrain et al., 2000;Yang et al., 2004;Deleris et al., 2006;Wang et al., 2010).
In this report, a silencing suppressor (HC-Pro)-deficient Turnip mosaic virus (TuMV) mutant encoding a green fluorescent protein tag (TuMV-AS9-GFP) was produced and propagated with the aid of a heterologous silencing suppressor provided in trans. The sensitivity of TuMV-AS9-GFP and parental TuMV-GFP viruses to antiviral silencing, and the siRNA-generating patterns, were compared in a series of single and combination dcl and rdr Arabidopsis mutants. The results revealed a distinct set of requirements for antiviral silencing and siRNA generation in inoculated and noninoculated tissues.

Mutational Inactivation of TuMV HC-Pro
Ala-scanning (AS) mutations were introduced in the central domain of Tobacco etch virus (TEV) HC-Pro (Kasschau et al., 1997). This resulted in several mutants, including TEV-AS9, which replicated to low levels in tobacco (Nicotiana tabacum) protoplasts and were unable to systemically infect plants (Kasschau et al., 1997). These phenotypes correlated with loss of RNA silencing suppression activity (Kasschau and Carrington, 2001). The AS9 mutation was introduced into the equivalent position in TuMV HC-Pro, and effects on silencing suppression activity were tested using a miRNA-directed silencing assay in which miR171 guides targeting of SCL6-IV transcript (Llave et al., 2002a;Kasschau et al., 2003). Both wild-type and AS9 mutant versions were tested in parallel with several other wild-type and mutant forms of silencing suppressors, including p19 from Tomato bushy stunt virus and p21 from Beet yellows virus (Chapman et al., 2004). Leaves were agroinfiltrated with constructs to coexpress SCL6-IV mRNA, miR171, and a silencing suppressor. The wild type, but not the mutant, forms of all the suppressors inhibited miR171-guided cleavage of SCL6-IV mRNA (Figure 1), showing that the AS9 mutation debilitated TuMV HC-Pro silencing suppressor activity.

Generation and Propagation of Suppressor-Deficient TuMV-AS9-GFP
We reasoned that infection of Arabidopsis plants by a suppressor-deficient TuMV would be arrested at points where antiviral RNA silencing responses are active. This approach would allow analysis of local and systemic infection of Arabidopsis plants with inactivating mutations in genes encoding RNA silencing factors. The AS9 mutation was introduced into TuMV-GFP, generating TuMV-AS9-GFP. This mutant accumulated to low levels in Nicotiana benthamiana, limiting the potential to propagate the virus to generate reliable inoculum. However, coexpression of constructs expressing the TuMV-AS9-GFP genome with p19 was predicted to rescue the suppressor-defective virus. Infection efficiency of TuMV-AS9-GFP was compared with parental TuMV-GFP in N. benthamiana. In fact, p19 rescued TuMV-AS9-GFP, as revealed by fluorescence and TuMV coat protein (CP) accumulation, in a dose-dependent manner ( Figure 2A) and enhanced local infection efficiency of TuMV-GFP ( Figure 2B). In the absence of p19, the number of TuMV-AS9-GFP infection (multicellular fluorescent) foci was only 36% of that measured for TuMV-GFP. In the presence of p19, infection efficiency of TuMV-AS9-GFP was similar to that of TuMV-GFP ( Figure 2B). These results suggest that TuMV-GFP HC-Pro was not able to block completely the silencing response in N. benthamiana. Alternatively, or additionally, some viral RNA transcribed from the cDNA cassettes carried by Agrobacterium tumefaciens may have been targeted by the silencing machinery before RNA replication was established. Parental and mutant forms of HC-Pro (HC), or two other suppressors (p19 and p21), were coinfiltrated with 35S:miR171-precursor and 35S:SCL6-IV. Leaves were collected 48 h after infiltration. SCL6-IV mRNA (a) and its miR171-guided 39 cleavage product (b) were analyzed in three independent samples by RNA gel blot assays using randomly primed 32 P-radiolabeled probes. For each treatment, the average (a)/(b) ratio for three replicates is indicated. Note that the parental, but not mutant, forms of the suppressors inhibited target cleavage.
Systemic infection of N. benthamiana by TuMV-GFP was detected at 6 d after inoculation (DAI). By contrast, no systemic infection of N. benthamiana by TuMV-AS9-GFP was detected either in the presence or absence of p19 at 15 DAI. At 21 DAI, mild systemic infection of N. benthamiana by TuMV-AS9-GFP was detected, although GFP intensity was substantially lower than that observed for TuMV-GFP at 6 DAI.
TuMV-AS9-GFP Is Limited by siRNA-Dependent Antiviral Defense That Is Normally Suppressed by HC-Pro TuMV-AS9-GFP failed to infect wild-type Arabidopsis (Columbia-0 [Col-0]) (Table 1, Figure 3A) despite using enriched, undiluted inoculum (at least 100-fold more concentrated than needed to give 100% systemic infection efficiency in dcl2-1 dcl3-1 dcl4-2 plants) (see Supplemental Figure 1 online). If immunity of wildtype plants was due entirely or predominantly to effective antiviral silencing in the absence of suppression function, as opposed to the loss of another HC-Pro function, then the mutant should gain infection competence in the siRNA-defective triple mutant with dcl2-1 dcl3-1 dcl4-2 alleles. For TuMV-GFP, an equal number of local infection foci, and similar amounts of CP, were measured in inoculated leaves and inflorescences of Col-0 and dcl2-1 dcl3-1 dcl4-2 mutant plants ( Figures 3A and 3B). No local infection foci or CP were detected in inoculated leaves or inflorescence, and symptoms did not develop in Col-0 inoculated with TuMV-AS9-GFP ( Figures 3A to 3C). By contrast, both local and systemic infection occurred in dcl2-1 dcl3-1 dcl4-2 mutant plants inoculated with TuMV-AS9-GFP (Figures 3A and 3B). Rescue of the defective infection phenotype in dcl2-1 dcl3-1 dcl4-2 mutant plants strongly supports the idea that RNA silencing limits the suppressor-defective virus and that HC-Pro promotes systemic infection by suppressing an siRNA-dependent activity. Interestingly, in dcl2-1 dcl3-1 dcl4-2 triple mutant plants, symptoms caused by TuMV-AS9-GFP were similar to those caused by TuMV-GFP ( Figure 3C).
To rigorously confirm that infection of the triple dcl mutant by TuMV-AS9-GFP was not due to viral reversion mutations (Garcia-Ruiz and Ahlquist, 2006), a bioassay was done using virus recovered from inflorescences of systemically infected plants at 15 DAI. Virus was concentrated and used to inoculate leaves of Col-0 and dcl2-1 dcl3-2 dcl4-2 triple mutant plants.  formed with pCB-TuMV-AS9-GFP or pCB-TuMV-GFP were provided at 10-fold dilutions of OD 600 = 5 3 10 À1 . Empty vector was used to normalize the cell density to OD 600 = 1.0. Images were taken at 5 DAI under UV light. Infection foci were plotted for samples receiving cultures at OD = 5 3 10 À4 . The histogram shows the average and SE for 16 leaves per treatment.
(C) Time course of accumulation of TuMV-AS9-GFP CP in N. benthamiana leaves infiltrated with Agrobacterium cultures containing pP19-HA and pCB-TuMV-AS9-GFP (OD 600 = 5 3 10 À1 ). Relative accumulation was plotted using 6 DAI measurements equal to 1.0. The histogram shows the average and SE for three replicates (individual leaves) per treatment.
was detected in Col-0 plants ( Figure 3D), confirming that the original infections of triple mutant plants by TuMV-AS9-GFP were not due to reversion mutations.

Genome-Wide Profiling of TuMV-Derived siRNAs
Genetic rescue of TuMV-AS9-GFP infection in dcl2-1 dcl3-1 dcl4-2 mutant plants ( Figure 3) shows that TuMV is targeted by an RNA silencing mechanism that requires one or more of the siRNA-generating DCLs. To obtain a detailed view of the siRNAs formed during TuMV infection, deep sequencing analysis of small RNA from Col-0, dcl2-1 dcl3-1 dcl4-2 triple mutant, and dcl1-7 single mutant plants was done. The dcl1-7 mutant, which is a partial loss-of-function mutant, was included because of the possibility that DCL1 might act on foldback structures within the TuMV genome and because of potential effects of DCL1 on the siRNA-generating enzymes (Blevins et al., 2006;Moissiard and Voinnet, 2006;Qu et al., 2008). Wild-type TuMV was used instead of TuMV-GFP to prevent GFP sequences from affecting the formation or accumulation of virus-derived siRNAs. As described for TuMV-GFP ( Figure 3B), TuMV accumulated to similar levels in inflorescences of Col-0, dcl1-7 single, and dcl2-1 dcl3-1 dcl4-2 triple mutant plants both at 7 and 10 DAI.
Small RNA libraries (three independent biological replicates/ treatment) were made from whole plants at both 7 and 10 DAI using sequencing-by-synthesis methods (Fahlgren et al., 2009). The frequency of reads obtained for the internal small RNA standards (Fahlgren et al., 2009) indicated consistent amplicon preparation across libraries. Results described below are from averages of biological replicates.
In infected plants, the number of reads with a perfect match to TuMV varied from 12,000 to 108,000 ( Figure 4A; see Supplemental Table 1 online). In Col-0 and dcl1-7 plants at 7 and 10 DAI, 98% of TuMV-derived small RNAs were 21 to 24 nucleotides long, with the vast majority 21 and 22 nucleotides ( Figure 4A). In agreement with a recent study (Donaire et al., 2009), small RNAs originated from both strands, although with a slight bias toward antisense reads ( Figure 4A). By contrast, the abundance of virusderived siRNAs in all size classes was significantly reduced in dcl2-1 dcl3-2 dcl4-2 triple mutant plants ( Figure 4A).
Genome-wide maps of small RNA reads from 10 DAI data sets revealed a high density of TuMV-derived siRNAs along both strands in Col-0 plants. Based on the numbers of reads mapping to the TuMV genome from mock-inoculated plants due to sequencing errors, the false discovery rate in our sequencing was estimated at 0.0079 per nucleotide and at 0.000053 per read. The distribution trends were similar between Col-0 and dcl1-7 ( Figure 4B), indicating that decreased levels of DCL1 have little effect on the pattern of TuMV-derived small RNA accumulation.
By contrast, ;26-fold fewer reads were detected in dcl2-1 dcl3-1 dcl4-2 triple mutant plants, with the decrease distributed proportionally across the TuMV genome ( Figure 4B). As the majority of these reads were 21 and 22 nucleotides in length, it is possible that DCL1 accounted for at least some of the reads in the triple mutant. In all plant genotypes tested, a large number of reads were mapped to the antisense strand of the 59 untranslated region (UTR), specifically to the first 40 nucleotides ( Figure 4C; see Supplemental Table 1 online). In dcl2-1 dcl3-1 dcl4-2 triple mutants, these small RNAs were still prevalent, accounting for 74% of the total TuMV-derived reads ( Figure 4C). Similar results were obtained for 21-nucleotide TuMV-derived siRNAs at 7 DAI (see Supplemental Figure 2 online) and for 22-nucleotide siRNAs from both 7 and 10 DAI (see Supplemental Figures 3 and 4 online). However, blot assays showed that the 21-nucleotide siRNA from the 59 UTR region were sensitive to loss of DCL4 (see Supplemental Figure 5 online). In the absence of DCL4, this region yielded siRNAs that were 22 nucleotides in length and dependent on DCL2. The basis for the unusual abundance of antisense 59 UTR-derived siRNAs in each of the sequencing data sets was unclear.
The ratio of antisense-to-sense 21-nucleotide TuMV-derived siRNAs, and the occurrence of each of the four nucleotides at the siRNA 59 position, was analyzed. In Col-0 and dcl1-7, the antisense strand bias was modest (see Supplemental Figure  6A online). The antisense/sense strand bias in the dcl2-1 dcl3-1 dcl4-2 triple mutant was far greater. However, the antisense strand bias was due almost entirely to the relatively high proportion of reads from the antisense 59 UTR region (see Supplemental Figure 6A online). Reads of 21 nucleotides were also grouped by their first nucleotide (A, U, G, or C) and their relative abundance compared with values expected from the nucleotide composition of the sense and antisense strands. The proportion of reads containing a 59 A, C, or G were close to the expected values. TuMV-derived siRNAs with a 59 U, however, were overrepresented in each of Col-0, dcl1-7, and the dcl2-1 dcl3-1 dcl4-2 triple mutant plants, with accumulation levels that were 1.3-to 2.6-fold higher than expected from the TuMV genome nucleotide composition (see Supplemental Figure 6B online). These results suggest that there is preferential accumulation of virus-derived siRNAs with a 59 terminal U, possibly through association with AGO1, which is known to have a 59 U preference (Mi et al., 2008;Montgomery et al., 2008a).

Participation of Arabidopsis DCL4 and DCL2 in Antiviral Defense against TuMV
To systematically analyze the requirement of DCL1, DCL2, DCL3, and DCL4 in antiviral RNA silencing and siRNA biogenesis, single and combination mutants were inoculated with parental TuMV-GFP and the suppressor-deficient TuMV-AS9-GFP. TuMV CP (as a reflection of virus accumulation) and virusderived siRNAs were measured in inoculated rosette leaves, noninoculated cauline leaves, and inflorescence tissues. Virus spread was also monitored using GFP fluorescence in inoculated and noninoculated tissue. The experiment with replicates was repeated twice with similar results.
TuMV-GFP infected, both locally and systemically, Col-0 and all dcl mutants tested (Table 2, Figure 5A), but no significant differences (P > 0.05) were detected in numbers of TuMV-GFP infection foci in inoculated leaves ( Figure 5A) or the general pattern of accumulation in systemic tissues (Table 2, Figures 5B  and 5C). By contrast, TuMV-AS9-GFP infection foci in inoculated leaves were visible only in inoculated rosette leaves of mutants containing the dcl4-2 allele (Table 2, Figure 5A). This was reflected in similar CP levels in inoculated leaves of dcl4-2 single, dcl2-1 dcl4-2 double, dcl3-1 dcl4-2 double, and dcl2-1 dcl3-1 dcl4-2 triple mutants ( Figure 5D; see Supplemental Figure 7 online). Each mutant containing the dcl4-2 allele also supported systemic infection by TuMV-AS9-GFP to cauline leaves, although combining the dcl4-2 and dcl2-1 mutations enhanced virus accumulation by approximately sixfold (Figures 5B and 5D). Interestingly, only mutants containing both dcl4-2 and dcl2-1 alleles supported TuMV-AS9-GFP spread to inflorescence tissues ( Figure 5D). These data suggest differences between requirements of antiviral defense at the local (DCL4-dependent) and systemic (DCL4-or DCL2-dependent) levels, supporting a  Consistent with previous findings Fusaro et al., 2006) and with our sequencing results ( Figure 4A), TuMV-GFPderived siRNAs in Col-0 plants were predominantly 21 nucleotides in all tissues tested ( Figures 6A to 6C). As was most evident in cauline leaves and inflorescences, the loss of DCL4 resulted in a size shift of the siRNA population to 22 nucleotides ( Figures  6B to 6D). Loss of both DCL2 and DCL4 led to accumulation of 24-nucleotide, DCL3-dependent siRNAs ( Figures 6B and 6C). All size classes of siRNA were lost in the dcl2-1 dcl3-1 dcl4-2 triple mutant ( Figures 6A to 6C). Thus, although there was a clear dependence on DCL4 for the abundant 21-nucleotide TuMV siRNA, and DCL2 and DCL3 could recognize TuMV-derived dsRNA to make 22-and 24-nucleotide siRNA, respectively, there was no correlation between any measurable siRNA accumulation patterns and antiviral defense against the suppressor-competent TuMV-GFP.
TuMV-AS9-GFP-derived siRNAs in inoculated leaves were below detection levels in all plants ( Figure 6A), although they were detected at low levels in cauline or inflorescence tissues of most mutant plants in which systemic spread was detected ( Figures 6A to 6C and 6E). As systemic infection by TuMV-AS9-GFP required loss of DCL4 and (in inflorescences) DCL2, the sizes of virus-derived siRNAs reflected the availability of DCLs. In cauline leaves of dcl4-2 single or dcl3-1 dcl4-2 double mutants, TuMV-AS9-GFP-derived siRNAs were 22 nucleotides in length and dependent on DCL2; in inflorescence tissues of dcl2-1 dcl4-2 double mutant plants, the DCL3-dependent 24-nucleotide size class of siRNA was detected ( Figures 6B, 6C, and 6E). All TuMV-AS9-GFP siRNAs were lost in the triple mutant.
TuMV-GFP systemically infected Col-0 and all rdr mutants tested. In all single and combination rdr mutants, TuMV-GFP infection induced symptoms similar to the observed for Col-0 ( Figures 3C and 7C), and no significant differences (P > 0.05) were detected in numbers of infection foci in inoculated leaves or virus accumulation in cauline leaves or inflorescence tissues (Table 3   However, there were significant (P < 0.05) differences in TuMV-GFP-derived siRNA accumulation among the mutants. TuMV-GFP siRNAs in rdr2-1 and rdr6-15 single, and in rdr2-1 rdr6-15 double, mutants accumulated to similar levels as those in Col-0 ( Figures 8A to 8D). By contrast, accumulation of siRNAs derived from the cylindrical inclusion (CI) or from the 59 UTR (see Supplemental Figure 5B online) of TuMV was significantly reduced in each single, double, and triple mutants harboring the rdr1-1 allele ( Figures 8A to 8D), indicating that RDR1 is a major contributor to biogenesis of TuMV-derived siRNAs. However, in (C) TuMV-GFP accumulation (CP, average + SE) in leaves (7 and 15 DAI) and inflorescence (15 DAI), expressed relative to accumulation of TuMV-AS9-GFP CP in dcl2-1 dcl3-2 dcl4-2.
(D) TuMV-AS9-GFP accumulation (CP, average + SE) in leaves (7 and 15 DAI) and inflorescence (15 DAI), relative to dcl2-1 dcl3-2 dcl4-2. Within each tissue, bars with the same letter are not statistically different (Tukey's test with a = 0.01). each tissue, a basal level of TuMV-GFP-derived siRNAs was detected in the rdr1-1 rdr2-1 rdr6-15 triple mutant ( Figures 8A to  8D). These levels were higher than those detected in the siRNAdefective dcl2-1 dcl3-2 dcl4-2 triple mutant, suggesting an alternative source of dsRNA during infection. In all rdr mutants, siRNAs were 21 nucleotides in length, which was consistent with dsRNA processing primarily by DCL4 ( Figure 3A). Surprisingly, all single, double, and triple rdr mutants inoculated with suppressor-defective TuMV-AS9-GFP yielded a similar number of infection foci, whereas no foci were detected in Col-0 plants ( Figure 7A, Table 3). Infection foci were less bright and CP accumulated to lower levels in inoculated leaves of each rdr single or combination mutant compared with what was detected in dcl2-1 dcl3-2 dcl4-2 plants ( Figures 7A and 7E). While TuMV-AS9-GFP was restricted to initial infection foci in rdr2-1 plants, the virus was able to move to cauline leaves and inflorescence tissue in each single or combination mutant containing either rdr1-1 or rdr6-15 alleles (Table 3, Figures 7B and  7E). Thus, RDR1 and RDR6 act additively or cooperatively to limit systemic infection of the suppressor-defective virus. However, in the rdr triple mutant, TuMV-AS9-GFP failed to induce symptoms ( Figure 7C) and accumulated to lower levels than in the dcl triple mutant ( Figure 7E; see Supplemental Figure 8 online), supporting the idea of a basal level of antiviral silencing that is RDR1, RDR2, and RDR6 independent. In fact, a low level of TuMV-AS9-GFP-derived siRNAs was detected in the both the rdr1-1 rdr6-15 double and rdr1-1 rdr2-1 rdr6-15 triple mutants ( Figures 8B and 8C).

DISCUSSION
Extensive use was made of TuMV-GFP and suppressor-deficient TuMV-AS9-GFP to expose antiviral silencing activities in Arabidopsis. Wild-type plants were immune to TuMV-AS9-GFP, but immunity was effectively suppressed by loss of DCL2 and DCL4, indicating that TuMV normally masks the effects of an siRNAdependent antiviral response. As noted previously (Deleris et al., 2006;Diaz-Pendon et al., 2007;Qu et al., 2008), suppressormediated masking has hampered interpretation of genetic analyses to understand antiviral silencing. This reinforces the requirement to examine suppressor-deficient viruses to gauge the contributions of silencing components during antiviral RNA silencing.

siRNA Biogenesis and Antiviral Silencing
Abundance of TuMV-derived siRNA within infected plants is a rather inaccurate reflection of antiviral silencing activity. During TuMV-GFP infection, large pools of 21-nucleotide siRNA Virus-derived siRNA were detected using 32 P-radiolabeled probes made by random priming of cDNA corresponding to the CI protein coding region. Virus-derived siRNA signals were normalized to U6 RNA signals from the same blots. In dcl4-2 single and dcl3-1 dcl4-2 double mutants, siRNAs were 22 nucleotides long. In dcl2-1 dcl4-2 double mutants, siRNAs were 24 nucleotides long.  (D) TuMV-GFP accumulation (CP, average + SE) in leaves (7 and 15 DAI) and inflorescence (15 DAI), relative to CP levels in Col-0. Within each tissue type, no statistically significant differences (Tukey's test with a = 0.01) were detected between genotypes.
(E) TuMV-AS9-GFP CP accumulation in leaves (7 and 15 DAI) and inflorescence (15 DAI), relative to dcl2-1 dcl3-2 dcl4-2. Within each tissue, bars with the same letter are not statistically different (Tukey's test with a = 0.01). accumulated concomitant with amplification and spread of the virus in wild-type plants ( Figure 4B). This can be reconciled with the known activity of HC-Pro in sequestration or inactivation of small RNAs, whereby siRNAs accumulate but lack activity (Lakatos et al., 2006). More importantly, this effect points out the need to carefully analyze both siRNA-related activities and antiviral immunity.
DCL4 and RDR1 were revealed as major contributors to the abundant pool of 21-nucleotide TuMV-GFP-derived siRNAs ( Figures 6D and 8D). However, in the absence of DCL4, 22-nucleotide DCL2-dependent siRNAs were formed. Furthermore, in the absence of DCL4 and DCL2, 24-nucleotide DCL3dependent siRNAs were produced ( Figure 6D). However, the 24-nucleotide siRNA (in the dcl2-1 dcl4-2 double mutant) had no apparent effect on antiviral silencing, and the 22-nucleotide siRNA (in the dcl4-2 mutant) did not restrict movement of the suppressor-defective virus to cauline leaves. Thus, while each of the siRNA-generating DCL proteins has hierarchical access to double-stranded viral RNA, it is clear that different size classes of siRNA function differentially, or not at all (24-nucleotide siRNAs), during antiviral defense. Similar conclusions were reached previously for infection of Arabidopsis by TRV (Deleris et al., 2006).
RDR6 is an important component of anti-TuMV silencing, as RDR6 was essential (along with RDR1) for restricting systemic infection by the suppressor-defective virus ( Figure 7E). However, the rdr6-15 mutation had no detectable effect on the biogenesis of TuMV-GFP-derived siRNAs (Figure 8). Thus, either RDR6 contributes very little to the dsRNA pool from which most siRNAs arise, or RDR6 activity is inhibited by HC-Pro during infection by TuMV-GFP.

Modular Activity of RNA Silencing Factors in Antiviral Silencing
The DCL4-dependent virus-derived siRNAs were necessary and sufficient (among the DCL family) to prevent initial infection foci in TuMV-AS9-GFP-inoculated leaves (Figures 5 and 6). By contrast, DCL2-dependent siRNAs were neither necessary nor sufficient to limit infections of TuMV-AS9-GFP in either inoculated leaves or cauline leaves. Interestingly, DCL2 limited systemic infection of inflorescence tissues in the absence of DCL4 ( Figure 5D, Table 2). Why, in the absence of DCL4, is DCL2 sufficient to prevent viral infection in inflorescences but not in inoculated leaves? It is likely not due to differential expression, as both of DCL4 and DCL2 are expressed at similar levels in both leaf and inflorescence tissues (Schmid et al., 2005). These results also cannot be explained by differential access by DCL2 to dsRNA substrates because 22-nucleotide-long TuMV-GFPderived siRNAs were detected both in leaves and in inflorescence of dcl4-2 single mutant plants ( Figure 6D). A quantitative DCL2-mediated reduction of local viral accumulation resulting in limited systemic spread is also unlikely, as DCL2 alone had no effect on suppressor-deficient virus accumulation in inoculated leaves (Figures 5A and 5D). Several nonmutually exclusive scenarios might explain the differential role of DCL2 in siRNA biogenesis and antiviral defense in inoculated leaves and inflorescences. DCL2 may have unique functions or cofactors in inflorescences or yield siRNAs that function with inflorescencespecific AGO effectors. Alternatively, in inoculated leaves, DCL2 might participate in the formation of a mobile, non-cellautonomous signal that potentiates an antiviral response in inflorescences. Interestingly, DCL2 was shown to function during transitive amplification of transgene silencing in an RDR6-dependent manner (Mlotshwa et al., 2008). No detectable effect was associated to RDR2 or RDR6 on accumulation of TuMV-GFP-derived siRNAs (Figures 7 and 8). However, the appearance of infection foci in inoculated leaves of rdr1-1, rdr2-1, and rdr6-15 single mutant plants means that the three RDR proteins work coordinately to limit primary infections. Thus, RDR1, RDR2, and RDR6 are complementary, or act in coordination, to restrict infection in primarily inoculated leaves by an unknown mechanism. Both RDR1 and RDR6 were necessary, but individually insufficient, to prevent systemic infection of cauline leaves. In inflorescence tissues, RDR1 and RDR6 were redundant in restricting infection, as TuMV-AS9-GFP was largely restricted from inflorescence tissues in single rdr1-1 and rdr6-15 mutants ( Figure 7E, Table 3).
One hypothesis to explain the differential effects associated to RDR1 and RDR6 in the biogenesis of virus-derived siRNAs and in antiviral RNA silencing is that Arabidopsis RDR-dependent pathways differ in their sensitivity to HC-Pro-mediated silencing suppression. RDR1 and RDR6 are known to have distinct functions or properties. RDR6, but not RDR1, participates in the biogenesis of tasiRNA (Peragine et al., 2004;Vazquez et al., 2004) and in posttranscriptional trans-gene silencing (Dalmay et al., 2000;Mourrain et al., 2000). Furthermore, RDR1, but not RDR6, is induced by salicylic acid or pathogen infection (Xie et al., 2001). Differential sensitivity to HC-Pro might constitute an additional difference. Alternatively, triggers that activate RDR1and RDR6-dependent activities might be differentially sensitive to HC-Pro. As proposed for RDR6-dependent formation of dsRNA during tasiRNA formation, it is postulated that RDRdependent amplification during virus infection would occur after initial AGO-siRNA targeting events on TuMV genomic RNA (Montgomery et al., 2008a(Montgomery et al., , 2008b. The initial AGO-siRNA targeting events might involve cleavage and deadenylation of viral RNA as well as noncleavage association of AGO-siRNA complexes with viral RNA. The associated AGO protein might serve as a flag for direct or indirect recruitment of RDR1 and RDR6. This line of reasoning implies that RDR1 and RDR6 triggers are different.

Entry of Viral RNA into Silencing Pathways
Two lines of evidence support the existence of an antiviral dsRNA-generating mechanism that is both independent from RDR1, RDR2, and RDR6 and that is insensitive to HC-Pro. First, suppressor-deficient TuMV-AS9-GFP accumulated to lower levels in rdr1-1 rdr2-1 rdr6-15 than in dcl2-1 dcl3-2 dcl4-2 triple mutants ( Figure 7E), indicating an alternate source of dsRNA that functions as a substrate for DCL activity in the triple rdr mutant. Second, low but consistent levels of virus-derived siRNAs were detected in rdr1-1 rdr2-1 rdr6-15 triple mutant plants infected with TuMV-GFP (Figure 8), which is consistent with similar findings by Donaire et al. (2008). The presumed dsRNA in rdr1-1 rdr2-1 rdr6-15 triple mutants could arise from the activities of one or more of RDR3, RDR4, and/or RDR5, none of which have been shown to be functional. Alternatively, in rdr1-1 rdr2-1 rdr6-15 triple mutants, the siRNA may be the result of processing of dsRNA from viral RNA replication products or intermediates as described for Flock house virus infections of Drosophila melanogaster (Aliyari et al., 2008;Flynt et al., 2009).
The initial trigger events for recognition of TuMV RNA remain unclear. Analysis of Cymbidium ringspot tumbusvirus-derived siRNAs in N. benthamiana showed that sense siRNA predominated over antisense siRNAs and that segments of the genome with extensive secondary structure yielded siRNAs at higher frequency (Molnar et al., 2005). Rather than a sense strand bias, a slight antisense bias was detected in most of the TuMV siRNA libraries analyzed ( Figure 4A; see Supplemental Figure 6 online), although this could be reflective of modest stabilization differences due to the likelihood of siRNAs forming with 59 ends containing each of the four possible nucleotides. Given that the antisense strand has a higher proportion of uracil than does the sense strand, it would be expected that more siRNA would have a 59 U residues, which would promote association with, and stabilization by, AGO1 (Mi et al., 2008;Montgomery et al., 2008a). Although no miRNA precursor-like structures were found in TuMV using criteria established for canonical miRNA precursors (Meyers et al., 2008), it is still possible that initial trigger events involve DCL-mediated cleavage of positive-strand viral RNA. These putative events could even be catalyzed by DCL1, even though the small RNA profile in dcl1-7 mutant plants was similar to that from wild-type plants. The dcl1-7 allele is a partial loss-of-function allele, meaning that residual activity might yield sufficient numbers of primary processing events that initiate amplification cycles that involve RDR1 and RDR6.

DNA Plasmids and Constructs
Recombinant plasmids and constructs used are described in the Supplemental Methods online.

Virus Propagation and Inoculum Enrichment
Both TuMV-GFP and TuMV-AS9-GFP were propagated in N. benthamiana. TuMV-GFP infection was launched by inoculation with Agrobacterium containing pCB-TuMV-GFP, and systemically infected leaves were collected 13 DAI. TuMV-AS9-GFP infection was launched by coinjection of leaves with pCB-TuMV-AS9-GFP and p19-HA (Chapman et al., 2004), and injected leaves were harvested 5 DAI ( Figure 2C). Virus preparations for subsequent inoculum were prepared using a polyethylene glycol precipitation procedure essentially as described (Choi et al., 1977) (see Supplemental Figure 1 online). The virus-enriched pellets were resuspended in 50 mM potassium phosphate, pH 7.5, and stored in 40% glycerol at 2208C.

Virus Infection Assays
Four Arabidopsis rosette leaves were dusted with carborundum and rubinoculated using 3 mL of inoculum/leaf. In all experiments, inoculum levels were adjusted to levels at least fivefold higher than needed to achieve 100% systemic infection efficiency by TuMV-AS9-GFP in dcl2-1 dcl3-1-dcl4-2 mutant plants, based on inoculum titration assays (see Supplemental Figure 1 online). Control plants were dusted with carborundum and mock-inoculated using 50 mM potassium phosphate, pH 7.5. Local and systemic infection of Arabidopsis by TuMV-GFP or TuMV-AS9-GFP was scored initially by GFP fluorescence. TuMV CP was detected by immunoblotting and chemiluminescence using antibody PVAS-134 at a 1:5000 dilution (Lellis et al., 2002) and Western Lighting plus-ECL substrate (Perkin-Elmer). Blots were exposed to film for 10, 30, 60, and 180 s to ensure signal detection in linear range. Four inoculated rosette leaves/ plant were collected at 7 DAI and pooled into a single sample. From the same plants at later time points, four cauline leaves or four inflorescence clusters were pooled. In all experiments, four replicate samples were analyzed for each virus-plant genotype combination at each time point and for each tissue type. Total protein extracts were adjusted to 1.0 mg/mL prior to dilution for blot assays using 6.5 mg of total protein. The linear range of detection was between 0.5 and 12.6 mg of total protein for TuMV-AS9-GFP-infected dcl2-1 dcl3-1 dcl4-2 triple mutant plants.

RNA Extraction and Small RNA Gel Blot Analysis
Total RNA was extracted using Trizol (Invitrogen) from whole individual plants or from independent pools of four inoculated rosette leaves, four noninoculated cauline leaves, or four inflorescence clusters. Blot hybridization was done as described (Llave et al., 2002a). 32 P-radiolabeled probes corresponding to the coding region for CI protein were generated by random priming of a PCR fragment [primers CI-F d(ACTCTCAATGA-TATAGAGGATG) and CI-R d(TTGATGGTGAACTGCCTCAAG)] using pCB-TuMV as template. Blots were stripped and reprobed for endogenous RDR2/DCL3-dependent siRNA02, RDR6/DCL1/DCL4-dependent TAS1 tasiRNA 255, and U6 RNA using cDNA oligonucleotides endlabeled with [g-32 P]ATP using Optikinase (USB). Hybridization intensities were quantified as described for the transient assays and normalized to the U6 RNA signal.
Profiling TuMV-Derived siRNAs by High-Throughput Sequencing Small RNA libraries from TuMV-infected or mock-inoculated plants were generated using sequencing-by-synthesis technology (Illumina Genome Analyzer I) (Fahlgren et al., 2009) and analyzed as described (Fahlgren et al., 2009). Bar-coded 59 adaptors (Cuperus et al., 2010) were used for multiplexing purposes. For each treatment, small RNA libraries were made independently from triplicate samples using 100 mg of total RNA extracted from aerial tissue from three plants. Four sets of amplicons were prepared using different bar-coded adaptors and were mixed in equal amounts (0.83 pMol/amplicon) and sequenced simultaneously in one lane. Small RNA sequences were parsed, and Arabidopsis and TuMV reads were identified, mapped, and quantified using CASHX (Fahlgren et al., 2009). Reads were normalized per 1,000,000 total reads.

Accession Number
Sequence data from this article can be found in Gene Expression Omnibus (http://www.ncbi.nlm.nih.gov/geo) under accession number GSE20197.

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