Heterochromatic siRNAs in Arabidopsis are typically 23 to 24 nt long and match both strands of corresponding genetic loci [5], [16], [18], [57]. A known 5S rDNA-related species, siR1003, was first characterized by Xie et al. (2004) and matches specific repeats on Chromosomes 3 and 5 [18]. In the present study, we identified 5S rDNA siRNAs in existing high-throughput sequencing data and mapped these to the ∼500 bp unit repeat.

Analysis of the origin of siRNAs is complicated by natural variation among the ∼1000 5S rDNA repeats. To resolve this problem, we generated a gapped alignment of 283 individual 5S rDNA repeats derived from an A. thaliana YAC library [58]. Next, 5′-end positions of individually matched small RNAs were aligned with a composite 530-bp repeat. Figure 1A summarizes our analysis of small RNA datasets from leaves obtained by Rajagopalan et al. (2006) and Kasschau et al. (2007) [57], [59]. Interestingly, 60% of the 3662 matched sequencing reads represented the intergenic spacer (IGS). A cluster of small RNAs ( Figure 1A , **) in the IGS matched both strands (44% and 56%, respectively) and included siR1003 ( Figure 1B ). This IGS cluster was also apparent in the maps we generated for other Arabidopsis tissues ( Figure S1 , seedlings and inflorescences). The remaining sequences that matched 5S rDNA, 40%, correspond to the 120-bp 5S rRNA genes: these reads were disproportionately 20 to 22-nt long and corresponded predominantly to the 5S rRNA-encoding strand (92% of 1481 genic reads). Analysis of inflorescence data yielded a spike of reads with 5′-ends corresponding to the 5S rRNA 5′-terminus ( Figure S1 ). Together with the abundant reads with 3′-ends aligned to the 5S rRNA 3′-terminus ( Figure 1A and Figure S1 ), this strongly suggests that most genic small RNAs are products of 5S rRNA degradation.

Numerous IGS siRNAs occur at positions consistent with the possibility that they derive from aberrant run-on transcripts of 5S rRNA genes that extend into the IGS. To test this hypothesis, we used RT-PCR to measure levels of two long transcripts known to be derived from 5S rDNA, namely: 5S LT1, which is 210 nt in length and extends 90 nt into the 3′ IGS, and 5S LT2, which has the same 5′ and 3′ ends as 5S LT1, but is 70 nt shorter due to an internal deletion ([52], [53], Figure 1A ).

Mutants homozygous for null alleles of the two largest Pol IV subunits, nrpd1 and nrpd2, showed higher levels of 5S LT1 than wild type plants; these elevated levels were comparable to 5S LT1 derepression in ddm1 ( Figure 1C ). Other mutants disrupting heterochromatic siRNA biogenesis, namely rdr2 and dcl3, also showed higher 5S LT1 levels. In contrast, a mutant deficient for trans-acting siRNA biogenesis [60], rdr6, did not show 5S LT1 derepression ( Figure S5 ). These data suggest that Pol IV/RDR2-dependent siRNA production, but not RDR6-dependent pathways, contribute to silencing of aberrant transcripts that include 5S rRNA gene sequences (120 bp) but continue at least 90 bp downstream into the IGS. Pol V is functionally and structurally distinct from Pol IV and its largest subunit, NRPE1, is unique [8]–[10], [19]. The nrpe1 mutant showed modest derepression of 5S LT1. Therefore, both plant-specific RNA polymerases are potentially required for suppressing long 5S rDNA transcripts. We did not detect reproducible changes in 5S LT2, the truncated transcript [52].

Triple (dcl2/3/4) and quadruple (dcl1/2/3/4) dicer mutants showed slightly more 5S LT1 accumulation than did dcl3 alone ( Figure 1C ). Thus, alternate DCLs might mediate 5S LT1 silencing, compensating for DCL3 deficiency. To test this hypothesis, we used RNA blot hybridization to analyze genetic requirements for IGS siRNA biogenesis ( Figure 1D ). We confirmed that siRNAs hybridizing to an siR1003 probe accumulate in wild type but not in nrpd1, nrpd2 or rdr2 mutants [6], [7], [18]. Comparison to Figure 1C suggests that the absence of siR1003 in these mutants correlates with derepression of 5S LT1 transcripts. Accumulation of 24-nt siR1003-related siRNAs was also decreased in dcl3 mutants, but was accompanied by the appearance of alternate siRNAs in the range 21–23 nt ( Figure 1D ), presumably due to the action of other dicers. Overall, the siR1003 signal in the 21 to 24-nt range in each dcl mutant roughly correlated with the extent of 5S LT1 silencing ( Figure 1C ). A longer probe spanning a region of the IGS that includes siR1003 ( Figure 1A , labeled 3′flank), detected siRNA accumulation patterns identical to those detected using the siR1003 probe ( Figure 1D ).

To infer processing functions for individual DCLs, we analyzed siR1003 accumulation in double mutant combinations of three DCLs (DCL2, DCL3 and DCL4). We detected ∼22-nt siRNAs in dcl3, but this size-class was absent in the double mutant dcl2/3. Moreover, ∼21-nt species detected in dcl2/3 were absent in the triple mutant dcl2/3/4 ( Figure 2A ). These hierarchical DCL dependencies, previously recognized for other repeat-derived and viral siRNAs [61]–[64], suggest that 21-nt and 22-nt IGS siRNAs are products of DCL4 and DCL2, respectively. Essentially the same results were obtained using probes for the reverse complement of siR1003 (data not shown). Together, our data suggest that dsRNA derived from aberrant 5S rDNA transcripts can be processed by multiple DCLs when DCL3 is mutated, generating IGS siRNAs that contribute to silencing of 5S rDNA.

We confirmed earlier reports that IGS transcript (5S LT1) and siRNA levels are elevated in ddm1 [7], [52]. Noting this phenomenon, we sought to localize IGS-derived RNA in interphase nuclei using RNA fluorescent in situ hybridization (RNA-FISH). In 83% of wild type nuclei (n = 136) the siR1003 probe detected RNA restricted to a distinct, sub-compartment of the nucleolus, a prominent interphase structure devoid of DAPI staining ( Figure 2B ). In contrast, the 5S RNA-FISH signal in 94% of ddm1 nuclei (n = 158) was typically dispersed throughout the nucleolus and nucleoplasm (see also Table S1 ). These results suggest that IGS-derived siRNAs or their precursors overaccumulate and/or mislocalize in ddm1 nuclei.

Based on the RNA-FISH result and siRNA overaccumulation observed in both ddm1 and met1 mutants [7], [54], we hypothesized that DDM1 and MET1 repress formation of dsRNA precursors that are processed by DCLs to generate IGS siRNAs. If this hypothesis is correct, a double mutant ddm1 dcl3 should overexpress alternate siRNA size-classes (i.e., those noted in Figure 2A ). We generated ddm1 dcl3 by genetic cross and, indeed, detected elevated siRNA levels in ddm1 dcl3 relative to dcl3 ( Figure 2C ).

DDM1 and MET1 are required to maintain a silent chromatin state associated with CG methylation [40], [41], [43], which is not immediately re-established upon replacement of functional DDM1 or MET1 [40], [65], [66]. To test whether this silent chromatin state suppresses siRNA precursors, we crossed homozygous nrpd1 (−/−) and ddm1 (−/−) lines to wild type Col-0, generating F1 plants that were heterozygous for nrpd1 (+/−) or ddm1 (+/−). Re-introduction of a functional NRPD1 allele in this manner restored siR1003 production to wild-type levels, as expected based on previous complementation and out-cross experiments [5], [19]. In contrast, reintroduction of a functional DDM1 allele failed to reduce siR1003 overaccumulation ( Figure 2D ). Persistent siRNA overaccumulation and CG hypomethylation in ddm1 heterozygotes ( Figure S3 ) suggests that an epigenetic state correlated with CG methylation, rather than expression of DDM1 per se, represses IGS siRNA precursor production.

Next, we tested whether elevated IGS siRNA levels are associated with enhanced RNA-directed DNA methylation (RdDM). Using Southern blot analysis we assayed cytosine methylation in asymmetric sequence contexts (CHH), which largely results from RdDM [35], [67]. After digestion with Alu I, which occurs once per 5S gene repeat, 5S rDNA arrays appear on Southern blots as a ladder of bands, each separated by 500-bp intervals. Longer than unit-length (500 bp) fragments reflect Alu I site methylation, which is more prominent in ddm1 and met1 than in wild type or siRNA biogenesis mutants (nrpd1, rdr2 and dcl3) ( Figure 2E ). Methylation is reduced in nrpd1 and rdr2 mutants, indicating that siRNA biogenesis is important for Alu I site methylation. However, dcl3 mutants resemble wild-type plants suggesting that other dicers compensate for the loss of 24-nt siRNAs in dcl3 mutants (see Figure 2A ). Collectively, the observed increase in DNA methylation in met1 and ddm1 mutants is consistent with overproduction of siRNAs leading to increased de novo methylation of asymmetric sites. An analogous finding has been reported for siRNA overaccumulation and 5S rDNA hypermethylation in met1 [54].

To further test whether hypermethylation in ddm1 is siRNA-dependent, we generated the double mutants nrpd1 ddm1 and rdr2 ddm1, in addition to dcl3 ddm1 mentioned above. The three double mutants were subjected to Southern blot analysis, alongside wild type and ddm1 controls. Long Alu I fragments typical of ddm1 mutants were absent in nrpd1 ddm1 and rdr2 ddm1 ( Figure 3A ), indicating that Alu I site hypermethylation in ddm1 is indeed siRNA-dependent (i.e., RdDM). dcl3 ddm1 showed 5S rDNA methylation similar to ddm1, again consistent with production of abundant siRNAs of variable lengths in this double mutant ( Figure 2C ). Analysis of Hae III methylation, used to test asymmetric 5S rDNA methylation in previous studies [19], [54], yielded identical results ( Figure 3A ). An additional assay using 5S LT1-specific primers, in which PCR amplification indicated DNA methylation status ( Figure 3A ), also supports the hypothesis that the hypermethylation is siRNA-dependent.

Analysis of 5S aberrant transcripts revealed that 5S LT1 accumulates to higher levels in nrpd1 ddm1 and rdr2 ddm1 double mutants than in ddm1 alone; this indicates that both DDM1 maintenance of silent chromatin and siRNA-dependent silencing contribute to suppression of aberrant transcripts ( Figure 3B ). Because siRNA and DDM1 deficiencies are additive in effect, they apparently act in separate pathways.

The previously observed localization of 5S rDNA outside chromocenters in ddm1 nuclei implies that 5S rDNA arrays are partially decondensed in ddm1 [51]. Using DNA-FISH we analyzed nuclei of the ddm1 mutant in genetic combination with nrpd1, rdr2 or dcl3. In the majority of wild type nuclei, 5S rDNA colocalized predominantly with heterochromatic domains known as chromocenters, which stain intensely with DAPI ( Figure 4A ). In contrast, decondensation of 5S rDNA was evident in ddm1 nuclei, resulting in smaller and more numerous signals. Analysis of 5S rDNA signals colocalized with DAPI-stained chromocenters confirmed that ddm1 nuclei contained more 5S rDNA outside chromocenters (52%, n = 158) than did wild-type nuclei (23%, n = 134). Importantly, nrpd1 ddm1 and rdr2 ddm1 double mutants displayed FISH signal outside chromocenters more frequently (68%, n = 186 and 73%, n = 215, respectively) than did ddm1, indicating an additive effect of these mutations on 5S rDNA decondensation ( Figure 4B ). These observed differences were all statistically significant (P<0.05, see Table S1 ). By contrast, 5S rDNA decondensation status in dcl3 ddm1 (58% outside, n = 190) was not distinguishable from ddm1.

Interestingly, mutants showing pronounced 5S rDNA decondensation also exhibited growth deficiencies; this was particularly evident in the double mutants nrpd1 ddm1 and rdr2 ddm1 ( Figure 5A ). The single mutants nrpd1, rdr2, dcl3 and ddm1 showed no appreciable effect on fresh weight ( Figure 5B ). In contrast, ddm1 nrpd1 and ddm1 rdr2 double mutant individuals were smaller and weighed, on average, 50% that of ddm1 or wild-type plants ( Figure 5 ). The overall stature and fresh weight of ddm1 dcl3 plants were similar to those of ddm1 plants.

nrpd1a-3 (nrpd1), nrpd1b-11 (nrpe1), nrpd2a-1 (nrpd2) are SALK T-DNA insertion lines described in Onodera et al. (2005) and Pontes et al. (2006) [7], [19]. rdr2-1 and rdr6-15 are described in Xie et al. (2004) and Allen et al. (2005), respectively [18], [74]. dcl2, dcl3 and dcl4 double and triple mutant combinations are described in Blevins et al. (2006) [62], with the constituent alleles being dcl2-5 of Akbergenov et al. (2006) [75], dcl3-1 of Xie et al. (2004) [18] and dcl4-2 of Xie et al. (2005) [76]. dcl3-4 is a GABI T-DNA insertion described by Daxinger et al. (2008) [77]. ddm1-2 used for crosses is described in Vongs et al. (1993) [39], and met1-3 in Saze et al. (2003) [41].

Quadruple mutant dcl1/2/3/4 material was obtained by crossing dcl1-9 (caf, [78]) to the triple homozygous mutant dcl2 (−/−) dcl3 (−/−) dcl4 (−/−). F2 progeny of the genotype dcl1-9(+/−) dcl2 (−/−) dcl3 (−/−) dcl4 (−/−) were then selected and self-fertilized. About 25% of resulting F3 progeny were quadruple homozygous d1/2/3/4. dcl1-9 used in the above crosses had been introgressed into Col-0 from Ler/Ws backgrounds by 5× outcrossing. Double mutants with ddm1-2 were generated as follows: nrpd1, rdr2 and dcl3 were individually crossed to ddm1. F2 plants homozygous for both ddm1 and each RNA silencing mutation were then self-fertilized for two generations and then analyzed with ddm1 and wild-type lines propagated in parallel.

Sequences of 283 individual 5S rDNA repeats ( File S1 )–from Yeast Artificial Chromosomes (YACs) 4E4, 6A1, 7E7, 7G3, 9A5, 9D3 and 11A3 of the CIC project [58], and representing arrays from chromosomes 3, 4 and 5 [32]–were retrieved from TIGR Plant Repeat Databases [79]. Small RNA datasets of Rajagopalan et al. (2006) and Kasschau et al. (2007) were retrieved from GEO (http://www.ncbi.nlm.nih.gov/geo) via accession numbers GSM118372 and GSM154370. Sequencing reads from each tissue were combined and filtered to obtain 20–25 nt small RNAs. BLAST (NCBI v2.2.16) was used to query reads to individual 5S rDNA sequences. A Perl script (incorporating Bioperl module Bio::SimpleAlign) converted the 5′-end positions of small RNA hits to individual repeats into coordinates along a multiple alignment of all repeats generated using Clustal W [80] ( File S2 ). Tallies of small RNA hits were weighted by number of sequencing reads and collected in 10 nt bins, using the Microsoft Excel 2003 Chart functionality for map generation. Small RNA distribution analyses and maps for leaves, inflorescences and seedlings are presented in Table S2 , Figure 1A and Figure S2 .

RNA was isolated using RNeasy Mini Columns (Qiagen). In brief, 100 mg of mixed-stage inflorescences were ground in liquid nitrogen and processed following the manufacturer's protocol. On-column DNase I digestion was performed (Qiagen), in addition to a second digestion after elution (DNase I Amplification Grade, Invitrogen). Following the protocol of Wierzbicki et al. (2008) [23], 400 ng RNA were subjected to gene-specific RT-PCR. To detect 5S LT1 and 5S LT2, primer 5SUNIV2 (R) was used for reverse transcription (SuperScript III, Invitrogen; 50°C for 30 min.). The RT enzyme was then inactivated (70°C for 15 min.), primer RTPCR5S1 (F) added and PCR amplification performed (Platinum Taq, Invitrogen; 52°C annealing, 30 s elongation, 35 cycles). See Vaillant et al. (2006) for 5S rDNA primer sequences [52]. ACT2 controls were amplified in separate reactions using primers described in Blevins et al. (2006) [62] and different cycling parameters (57°C annealing, 40 s elongation, 25 cycles). Reactions were subjected to agarose gel electrophoresis and stained with ethidium bromide.

Total RNA was isolated from 600 mg inflorescence tissue by grinding in liquid nitrogen followed by TRIzol extraction (Invitrogen). 100 µg total RNA sample was size-fractionated on RNeasy Mini Columns (Qiagen) as in Blevins et al. (2006) [62], then 8 µg low molecular weight RNA was separated on an 18% polyacrylamide gel. Probes for specific RNA species were DNA oligonucleotides end-labeled with γ[³²P]-ATP: the siR1003 probe described in Xie et al. (2004), U6 snRNA probes in Blevins et al. (2006) [62], and miR-160 probe 5′-TGG CAT ACA GGG AGC CAG GCÀ-3′. DNA oligo probe hybridization was performed as in Blevins et al. (2006). The 5S rRNA 3′-flank probe was generated by T7 polymerase transcription of a PCR fragment in the presence of α[³²P]-CTP. Primers for this template were: T7_5SUNIV2, 5′-taa tac gac tca cta tag gga gaC GAA AAG GTA TCA CAT GCC-3′ and 3′flank_F, 5′-GCA TCC CTC TTT TAT GTT TAA CCT-3′. RNA probe hybridization was performed as in Onodera et al. (2005) [7].

Genomic DNA was isolated from Arabidopsis inflorescences using Nucleon PhytoPure DNA extraction kit (Amersham). 3 µg aliquots of DNA were digested overnight with 30 U of Alu I, Hpa II or Hae III in the corresponding commercial buffer (New England Biolabs). Following size-separation by 0.8% agarose gel electrophoresis, gel-bound DNA was depurinated, denatured and transferred to Hybond-N+ nylon membrane (Amersham). Hybridization was performed overnight at 55°C in Church Buffer (0.25 M Sodium Phosphate, pH 7.2; 1 mM EDTA; 6.6% SDS; 1% BSA). The 5S rDNA probe was a 5S LT1 PCR fragment internally labeled with α[³²P]-dCTP by random priming. The membrane was washed twice 5 min. with 2× SSC, 0.1% SDS and twice 20 min. with 0.1× SSC, 0.1% SDS; signal was detected by phosphoimaging. PCR analysis of cytosine methylation used 100 ng aliquots of the above digests compared to undigested DNA controls. PCR was performed for 25 cycles using Platinum Taq (Invitrogen) and primers RTPCR5S1 and 5SUNIV2 described in Vaillant et al. (2006) [52].

The 5S rDNA gene probe was labeled with biotin-16-dUTP by PCR as previously described [7]. Meristem nuclei were processed for DNA fluorescence in situ hybridization (FISH) as previously described [81], using 50% formamide and 2× SSC for the hybridization steps. Post-hybridization washes were performed in 50% formamide and 0.1× SSC at 42°C. Biotin-labeled probes were detected using goat anti-biotin conjugated with avidin (1∶200, Vector Laboratories) followed by streptavidin-Alexa 546 (1∶200, Molecular Probes). Nuclear DNA was counterstained with DAPI in Prolong antifade medium (Invitrogen). Preparations were inspected with a Nikon Eclipse E800i epifluorescence microscope equipped with a Photometrics Coolsnap ES Mono digital camera. Images were acquired using Phylum software, and pseudocolored and merged in Adobe Photoshop 7.