35S:P15FHA^SKL construct was generated by amplifying P15FHA (Incarbone et al., 2017) with a reverse primer containing an SKL-encoding sequence (TCTAAACTG) before the stop codon, and cloning it through restriction (XmaI) and ligation into binary vector pCTL-35S. 35S:P19HA and 35S:P19HA^SKL were obtained similarly, except that the sequence encoding the HA tag was added to the reverse primer upstream of the stop codon (P19HA) or the SKL-encoding codons (P19HA^SKL), and SalI/PstI restriction sites were used. 35S:AGO2 and 35S:AGO2^SKL were obtained by amplifying AGO2 (AT1G31280) genomic sequence from A. thaliana Col0, with primers containing attB sites for Gateway cloning (with SKL-encoding codons upstream of the stop in the case of AGO2^SKL), inserted into pDONR221 through BP recombination, then into pH2GW7 binary vector through LR recombination. SUC-SUL (Himber et al., 2003; Dunoyer et al., 2005) and ago2-1 (SALK_003380) plants were transformed through floral dip as previously described (Bechtold and Pelletier, 1998). Transformed lines were selected by growing them in vitro on MS medium containing hygromycin. The 35S:P15FHA/SUC:SUL line was previously described (Incarbone et al., 2017). All experiments were performed on 6–7-week old rosettes grown in growth chambers, with 12h/12h day/night cycles.

HA-epitope IP was performed as previously described (Incarbone et al., 2017). 0.2 g of frozen rosette leaves were ground in liquid nitrogen in 1 ml lysis buffer (50 mM Tris-HCl pH 8, 150 mM NaCl, 1% Triton X-100, Roche Complete protease inhibitor cocktail) and incubated 15 min on a rotating wheel. After two clarifications (12000 g for 5 min at 4°C), an aliquot was set aside as input fraction, and 50 μl of anti-HA micro-beads (MACS system, Miltenyi Biotec, ref. 130-091-122) were added to the remaining lysate and incubated 30 min at 4°C on wheel. Next, the lysate containing the beads was deposited on a magnetic M column (Miltenyi Biotech) and allowed to flow through. An aliquot of the flow-through was kept for further analysis, and beads were washed by adding 2 × 500 μl of lysis buffer to the columns and allowed to pass through, then 100 μl of 20 mM Tris-HCl pH 7.5, after which excess liquid was removed. Beads were recovered in 1 ml of Tri-reagent (Sigma).

Peroxisome isolations were performed as previously described (Reumann and Singhal, 2014), with minor modifications. Two samples were treated in each experiment, and all procedures were performed at 4°C. Note that plants must not be frozen prior to peroxisome isolation. Before isolation, plants were kept in the dark for 16–20 h. After sampling tissue for total RNA/protein analysis and IP, 20 g of whole A. thaliana rosettes per sample were harvested and left on ice 2 h. Plant tissue was minced with a knife, then ground in 60 ml grinding buffer (170 mM Tricine pH 7.5, 1 M sucrose, 2 mM EDTA, 1% BSA, 10 mM KCl, 1 mM MgCl₂, plus 0.5% PVP-40, 5 mM DTT and Roche Complete protease inhibitor cocktail added before use). The resulting pulp was filtered through Miracloth, the liquid divided into three tubes and centrifuged 1 min at 6700 g (Beckman rotor JA25.5). The clarified supernatant was pooled and deposited on 4 freshly prepared and chilled Percoll/sucrose gradients. These contain, from top to bottom, 3 ml of 15% Percoll (15% Percoll, 750 mM sucrose, 20 mM tricine, 1 mM EDTA, 0.2% BSA), 9 ml of 38% Percoll (38% Percoll, 750 mM sucrose, 20 mM tricine, 1 mM EDTA, 0.2% BSA), 2 ml of 2:1 mix 38% Percoll:36% sucrose, 2 ml of 1:2 mix 38% Percoll:36% sucrose, and 3 ml of 36% sucrose (36% sucrose w/w, 20 mM tricine, 1 mM EDTA). Gradients were centrifuged 12 min at 13200 g, then without stop 20 min at 27000 g (Beckman rotor JA25.5), with medium brake. Top layers were discarded, while the bottom 2–3 ml were kept, pooled and diluted up to a volume of 60 ml in 36% sucrose solution, divided in 3 tubes and centrifuged 30 min at 38700 g (Beckman rotor JA25.5). Next, 1 ml of the organellar phase on the bottom of each tube was directly harvested with a cut-tip pipette, transferred to a potter and gently homogenized. Then, the samples were deposited on a sucrose 41.2 to 60% discontinuous gradient (from top: 0.8 ml 41.2%, 1.6 ml 43.7%, 1.6 ml 46%, 2.4 ml 48.5%, 0.5 ml 50.5%, 1.6 ml 55.2%, 0.8 ml 60% sucrose w/w, 20 mM tricine, 1 mM EDTA). Gradients were ultra-centrifuged 40 min at 110800 g (Beckman rotor SW41), with maximum acceleration and brake. 1.5 ml of visible white peroxisome fraction within the 50.5% sucrose phase were harvested and frozen at –80°C.

RNA extraction was performed using Tri-reagent (Sigma), according to manufacturer’s instructions. In the case of immunoprecipitated and peroxisomal RNA, 1.5 μl glycogen was added during isopropanol precipitation, which was allowed O/N at 4°C. Small RNAs were resolved through PAGE (low- or high-resolution gels, according to the experiment) and blotted as previously described (Dunoyer et al., 2007). RNA was chemically crosslinked on nylon membranes by incubating 1 h 30 min at 60°C on Whatmann paper imbibed with EDC solution, composed of 0.125 M 1-Methylimidazole (Sigma–Aldrich, ref. M50834) and 3% N-(3-Dimethylaminopropyl)-N′-ethylcarbodiimide hydrochloride powder (Sigma–Aldrich, ref. E7750) and 1% HCl 1 M. Detection of RNA species was achieved by hybridizing membranes in Sigma PerfectHyb^TM Plus buffer with PCR products labeled with α-³²P-CTP through Klenow reaction (SUL, IR71) or with oligonucleotides labeled with γ³²P-ATP through PNK reaction (U6, Rep2, ta-siRNA255, TAS3 5’D7(+), miR159, miR160, miR169, miR173, miR403, miR408; Probing order is available upon request). Hybridization was carried out O/N at 42°C, followed by 3x10 min washes in 2X SSC, 2% SDS at 50°C. Radioactive signal was revealed with autoradiographic films (Fujifilm). Stripping of the membrane before reprobing is done by submerging the membrane in 500 ml of boiling stripping buffer (0.1% SDS) for 3 × 10 min. The efficiency of the stripping was assessed by checking the membrane with a Geiger counter radiation detector.

Total protein was obtained from frozen tissues through phenol extraction followed by methanol/ammonium acetate precipitation, as previously described (Hurkman and Tanaka, 1986). Peroxisomal and immunoprecipitated protein were obtained from the phenolic phase resulting from Tri-reagent RNA extraction, through precipitation in acetone, according to manufacturer’s instructions. Proteins were resolved by SDS-PAGE and electro-blotted onto Immobilion-P membrane (Millipore), which were then incubated with the appropriate antibody (@HA: Sigma–Aldrich ref. H6533; @HPR: Agrisera ref. AS11 1797; @P15: Incarbone et al., 2017; @P19: kindly provided by K. Bouarab; @AGO2: Garcia et al., 2012; @AGO1: Qi et al., 2005). After incubation with secondary antibody, membranes were revealed with Roche Lumilight Plus substrate (ref. 1201519600) and autoradiographic films (Fujifilm).

Previous experiments conducted with the VSR P15 have shown that although 21nt miRNAs could be detected in isolated peroxisomes, they were below detection level in northern analysis following IP experiments (Incarbone et al., 2017; MI and PD unpublished observations). These differences observed between IP and peroxisomal isolation approaches, regarding the detection of P15-bound miRNAs, could be a consequence of the different version of P15 used in these experiments. Indeed, whereas small RNA piggybacking in peroxisomes was assessed in wild-type P15-expressing plants, IP relied on a 2xFlag-2xHA epitope-tagged version of this VSR (P15FHA), which may exhibit an altered small RNA-binding ability compared to the untagged P15. Therefore, in order to compare side-by-side the sensitivity of the two approaches, we generated transgenic lines expressing the same epitope-tagged version of P15, fused or not in C-terminal to the canonical PTS1 tripeptide, serine (S)-lysine (K)-leucine (L) (P15FHA^SKL and P15FHA, respectively). Addition of this artificial PTS1 should allow to target P15FHA^SKL to peroxisomes, as opposed to the P15FHA-expressing lines where the tag prevents the peroxisomal import of P15 (Incarbone et al., 2017), by masking the PTS1 sequence naturally present in C-terminal of this VSR (Dunoyer et al., 2002). Both 35S:P15FHA and 35S:P15FHA^SKL transgenes were introduced into the SUC:SUL reporter system, where an inverted-repeat (IR) construct, driven by the phloem-companion cell-specific AtSUC2 promoter, triggers RNAi of the endogenous SULHUR (SUL) mRNA and apparition of a chlorotic phenotype that expands 10-15 cells beyond the vasculature (Himber et al., 2003; Dunoyer et al., 2005). Of note, although the SUL IR is processed by DCL4 and DCL3 to generate 21nt and 24nt siRNAs, respectively, only the former is required for the appearance of the SUL-silencing phenotype (Dunoyer et al., 2007). We then selected a 35S:P15FHA^SKL/SUC:SUL transgenic line that accumulated a similar amount of epitope-tagged P15 than the one found in our 35S:P15FHA/SUC:SUL line (Figure 1D), and producing similar levels of SUL siRNAs than the one found in our SUC:SUL reference line (Figure 1C).

Neither P15FHA- nor P15FHA^SKL-expressing plants displayed a visible SUL-silencing phenotype, indicating that both proteins were equally competent in suppressing siRNA-mediated silencing (Supplementary Figure S1A). Accordingly, both proteins were able to bind 21nt SUL siRNAs, as revealed by their detection in both P15FHA IP fractions (Figure 1A) and P15FHA^SKL peroxisomal fractions (Figure 1C). Similarly, 22nt-long siRNAs deriving from the endogenous IR71 locus, and the 22nt-long miRNA miR173, were found associated to P15 using both approaches (Figures 1A,C). However, the results were drastically different when we analyzed P15-binding to several 21nt-long miRNAs. Indeed, and in agreement with our previous report (Incarbone et al., 2017), all 21nt-long miRNAs tested were below detection level in P15FHA IP fractions (Figure 1A). In sharp contrast, all 21nt-long miRNAs were consistently found in three biological replicates of peroxisomal isolation performed on P15FHA^SKL-expressing plants (Figure 1C). Given that these miRNAs were not detected in peroxisomal isolates of the P15FHA-expressing line, performed in parallel, this indicates that the miRNAs specifically detected in peroxisomes of plants expressing P15FHA^SKL result from P15-mediated piggybacking of these small RNAs. Moreover, detection in these peroxisomal isolates of the otherwise unstable miR160 passenger strand (miR160^∗), strongly suggests that small RNAs are imported as duplexes by P15.

Collectively, these results suggest that, despite an apparent lower binding capacity towards 21nt miRNAs, P15 is able to associate, at least to a certain extent, with this specific class of small RNAs. This association is strong enough to cause the import of detectable levels of miRNA into peroxisomes, along with the other classes of P15-bound small RNAs. Once delivered into these organelles, their confinement within a closed membrane structure preserves them from the ex vivo treatment performed during peroxisomal isolation, whether or not they remain bound to P15, thereby allowing their detection. By contrast, P15 interaction with 21nt miRNAs seems to be lost upon ex vivo conditions applied during IP experiments (Incarbone et al., 2017). These observations prompted us to consider peroxisomal isolation as an alternative approach to the widely used IP experiments, that may provide more information about VSR-associated cargoes in planta.

Intriguingly, despite efficient P15FHA^SKL-dependent piggybacking of small RNAs in peroxisomes (Figure 1C), the P15FHA^SKL protein was difficult to detect within peroxisomal isolates, using either antibodies raised against the HA tag or the P15 protein (Figure 1D and Supplementary Figure S1B). This was surprising given that (i) both hydroxypyruvate reductase (HPR), our peroxisomal marker, and wild-type untagged P15 were readily detectable in, respectively, this and previous analyses of peroxisomal fractions (Incarbone et al., 2017), and that (ii) P15FHA^SKL was found to accumulate to high levels in total protein extracts (Figure 1D and Supplementary Figure S1B). These observations suggest that either the P15FHA^SKL is quickly recycled to the cytoplasm following peroxisomal import and release of the bound small RNAs, and/or that addition of the 2xFlag2xHA tag to P15 triggers the prompt degradation of the P15FHA^SKL within peroxisomes by a currently unknown mechanism.

The more sensitive detection of P15-bound small RNAs in peroxisomal fractions compared to IP experiments (Figures 1A,C) prompted us to determine whether these observations were specific to P15, or if this trend could also be observed with another, unrelated, small RNA-binding VSR upon artificial targeting to peroxisomes. For this purpose, we decided to test the tombusviral P19. P19 has been extensively studied in the past and is potentially the best characterized VSR to date. The structure of P19 has been resolved and showed that a head-to-tail P19 homodimer binds preferentially 21nt-long small RNA duplexes with 2nt 3’ overhangs (Vargason et al., 2003; Ye et al., 2003). Accordingly, IP experiments revealed that P19 efficiently binds and sequesters 21nt siRNAs and miRNAs (Chapman et al., 2004; Schott et al., 2012), thereby preventing their loading into AGO effectors (Schott et al., 2012; Kontra et al., 2016; Incarbone et al., 2017).

In order to directly compare the results of IP versus peroxisomal isolation regarding detection of P19-bound small RNAs, we first generated a peroxisome import-competent version of P19, using the same rationale as for P15. Thus, we generated SUC:SUL transgenic lines expressing an HA-tagged version of P19, fused or not in C-terminal to the canonical PTS1 tripeptide, SKL (35S:P19HA^SKL/SUC:SUL or 35S:P19HA/SUC:SUL, respectively). Western blot analysis revealed that, despite lower accumulation of P19HA^SKL than P19HA in total fractions, only the former was detected in peroxisomal isolates (Figure 2B), supporting functional targeting of this protein to peroxisomes. Importantly, both constructs efficiently impaired the appearance of the SUL-silencing phenotype (Figure 2A), without affecting production of SUL siRNAs (Figure 2C), indicating that addition of the SKL tripeptide did not hinder P19 ability to suppress siRNA-mediated silencing. Moreover, expression of P19HA^SKL triggered the appearance of similar developmental defects (mostly characterized by leaf serrations) as the one observed in P19HA-expressing lines (Figure 2A). Given that this phenotype has been previously associated with a disruption of the miRNA pathway (Kasschau et al., 2003; Chapman et al., 2004; Jay et al., 2011), these observations suggest that both proteins were also efficient in suppressing miRNA-mediated silencing.

We next wondered whether addition of a PTS1 to the tombusviral P19 also leads to piggybacking of P19-bound small RNA into peroxisomes. In agreement with the absence of the SUL-silencing phenotype, the presence of developmental defects (Figure 2A), and the documented size-specificity of P19 (Vargason et al., 2003; Ye et al., 2003; Lakatos et al., 2006; Mérai et al., 2006), northern analysis revealed that both 21nt SUL siRNAs and miR159 were indeed readily detected in peroxisomal isolates of P19HA^SKL-expressing plants but not in control plants (Figure 2C), indicating specific P19-mediated piggybacking of these small RNAs into peroxisomes. More intriguingly, DCL3-dependent 24nt long siRNAs deriving from the exogenous SUL-IR transgene, the endogenous IR71 locus or the p4-siRNA REP2 loci, were also specifically detected in these fractions, suggesting that this small RNA size class can also be imported into peroxisomes along with P19HA^SKL. Although surprising given that (i) in vitro binding assays showed that P19 has a 22-fold lower affinity for 24nt siRNA than for its 21nt counterpart (Vargason et al., 2003), and (ii) that 24nt SUL siRNAs were not detected in previous P19 IP experiments (Incarbone et al., 2017), these results may further support the advantage of the peroxisomal isolation approach to detect unstable or labile interactions, compared to IP experiments.

To confirm these observations, we decided to perform a side-by-side comparison of the P19-bound small RNAs retrieved using these two approaches. Similarly, to the experiments described above in the case of P15 (Figure 1), both IP and peroxisome isolations were performed in triplicate from P19HA- or P19HA^SKL-expressing plants, grown in parallel in the same conditions and harvested simultaneously. In agreement with previous reports, IP experiments showed that, out of the 21nt and 24nt SUL-siRNAs produced in these plants, P19HA only efficiently binds to the former species (Figure 3A). In addition, analysis of the endogenous IR71-derived siRNAs, which are produced by DCL2, DCL3 and, to a lower extent, by DCL4, revealed that both 21nt and 22nt siRNAs can be immunoprecipitated by P19HA, although with an apparent bias toward the 21nt species (Figure 3A). Importantly, neither the SUL-, IR71-, nor REP2-derived 24nt siRNAs were found associated to P19HA in these IP experiments.

Similarly, to the 21nt siRNAs, 21nt miRNAs were also efficiently immunoprecipitated by P19HA, together with 1 or 2 nucleotide-shorter derivatives (Figure 3A), which most likely correspond to the recently described 3′-end trimming products of P19-bound miRNAs (Kontra et al., 2016). A similar trimming product was also observed for the miR160^∗ passenger strand in our IP fractions, indicating that both 2nt 3’ overhang extremities of the P19-bound miRNA duplexes can be attacked by the exonuclease activity responsible for this shortening. As previously observed (Moissiard et al., 2007; Incarbone et al., 2017), expression of P19 triggers a strong decrease in miR173 accumulation (Figure 3A). Although the mechanism behind this specific destabilization remains unknown, detection with the miR173 probe of several shorter bands in both total and IP fractions suggests that this specific miRNA duplex may be particularly sensitive to the P19-induced 3′end trimming enzymatic activity.

In parallel, the results obtained from northern analysis of the small RNAs retrieved in our peroxisomal isolates (Figure 3C) yielded interesting differences compared to the IP experiments. First, in addition to the 21nt and 22nt siRNA species, both SUL- and IR71-derived 24nt siRNAs were readily detected in peroxisomes of P19HA^SKL-expressing plants, as were the REP2-derived 24nt siRNAs (Figure 3C). Importantly, none of these siRNA species were detected in peroxisomal isolates of P19HA-expressing plants performed in parallel, supporting that their detection in peroxisomes of P19HA^SKL-expressing plants specifically results from their piggybacking into these organelles by the P19HA^SKL. In addition, western and mass spectrometry analysis of these peroxisomal extracts revealed that presence of these small RNA species in peroxisomes cannot be attributed to import of AGO proteins or other known silencing factors within these organelles upon P19HA^SKL expression (Supplementary Figure S2A and Supplementary Table S1). This indicates that P19 is able to bind 24nt siRNA species in vivo, but that this interaction is most likely lost during the ex vivo treatment applied during IP experiments, while it is maintained throughout the import into peroxisomes. Secondly, although miRNAs were, as expected, found to be efficiently imported into peroxisomes of P19HA^SKL-expressing plants, we observed a strong reduction in the accumulation of their 1-2nt-shorter derivatives in both total and peroxisomal RNA fractions (Figure 3C), suggesting that peroxisomal import of P19-bound miRNAs protects them from the 3′-end trimming activity present in the cytoplasm. Consequently, miR173 accumulation, which was at or below detection level in P19HA-expressing plants, is restored to wild-type levels in the presence of P19HA^SKL. Therefore, both P15 and P19 experiments (Figures 1, 3) support peroxisomal isolation as an alternative approach to obtain more information about VSR-associated small RNAs in planta than that provided by conventional IP experiments.

Of note, and while not as drastically different as what we observed above in the case of P15FHA^SKL (Figure 1D), detection of the P19HA^SKL within peroxisomal isolates was also difficult to achieve using both anti-HA or anti-P19 antibodies, and required the loading of a higher amount of peroxisomal protein than the one required to detect HPR (Figure 3D and Supplementary Figure S2B). Although beyond the scope of this paper, these observations suggest that addition of an HA tag (one for P19HA^SKL and two for P15FHA^SKL) seems to promote degradation of the tagged protein within peroxisomes, or their prompt export, by an unknown mechanism.

We next decided to assess whether peroxisomal targeting can also be used to determine RNA or protein interactors of the RNA silencing effector AGO2, which plays key roles during the antiviral silencing reaction against several RNA viruses (Harvey et al., 2011; Jaubert et al., 2011; Wang et al., 2011; Carbonell et al., 2012; Garcia-Ruiz et al., 2015). For this purpose, we generated, in an ago2-1 mutant background, transgenic lines expressing AGO2, fused or not in C-terminal to the canonical PTS1 tripeptide SKL (35S:AGO2^SKL/ago2-1 or 35S:AGO2/ago2-1, respectively). Unfortunately, despite high levels of AGO2^SKL accumulation in total fractions, we were unable to detect any import of this protein in peroxisomal fractions (Figure 4A), suggesting that addition of a C-terminal PTS1 does not lead to peroxisomal import of AGO2. While the reason for this lack of AGO2 import into peroxisomes remains to be formally determined, we believe this is most likely caused by the inaccessibility of the PTS1 signal to PEX5, as crystal structures of human and yeast Argonaute proteins have shown that their C-terminal ends are not exposed on the surface, but rather buried within the RNA binding cleft (Poulsen et al., 2013).

Interestingly, while 35S:AGO2^SKL-expressing plants were phenotypically undistinguishable from wild-type plants, we noticed that our 35S:AGO2/ago2-1 transgenic lines displayed an elongated, “zippy” leaf phenotype (Figure 4B), typically observed in plants deficient for TAS3 trans-acting (ta)-siRNA production (Adenot et al., 2006; Garcia et al., 2006). The production of TAS3 ta-siRNAs relies on a unique two-hit pathway involving the loading of miR390 into AGO7, cleavage of the TAS3 transcript in one of the two AGO7-miR390 binding sites, conversion by RDR6 of the cleaved TAS3 transcript into dsRNA and its subsequent dicing by DCL4 into ta-siRNAs (Axtell et al., 2006; Montgomery et al., 2008). However, miR390 has also been reported to be loaded into AGO2 (Montgomery et al., 2008; Takeda et al., 2008; Fatyol et al., 2015), possibly because of its 5′A, which has been shown to promote small RNA loading into this Argonaute (Mi et al., 2008).

In agreement with their elongated leaf phenotype, northern analysis revealed that the TAS3 5’D7(+) ta-siRNA accumulation was indeed strongly reduced in our 35S:AGO2 transgenic lines (Figure 4C). We hypothesized that this reduction could be a consequence of direct competition for miR390 loading between AGO7 and AGO2. This could result from the overexpression of AGO2, assuming that the AGO2/miR390 complex would be less efficient than the AGO7/miR390 complex to trigger TAS3 ta-siRNA production. Surprisingly, we found that miR390 steady-state level was also strongly reduced in our 35S:AGO2 transgenic lines, whereas the other miRNAs tested accumulated to wild-type levels (Figure 4C). As mutants impaired in TAS3 ta-siRNA production, such as rdr6, are not affected in miR390 accumulation (Fahlgren et al., 2006), this indicates a specific effect of AGO2 overexpression. Given that upon loading of the miRNA guide strand into AGO proteins, the complementary miRNA passenger strand (miR^∗) is rapidly degraded, we next hypothesized that AGO2 in our transgenic lines might be preferentially loaded with miR390^∗, thereby triggering degradation of the miR390 guide strand. The corollary of this hypothesis is that the miR390^∗ should be stabilized in AGO2 over-expressing compared to wild-type plants, as previously observed in the case of miR393^∗ loading into AGO2 during Pseudomonas syringae infection (Zhang et al., 2011). However, northern analysis revealed a strong reduction of miR390^∗ in our 35S:AGO2 transgenic plants, leaving open the molecular mechanism behind these observations.