Endomembrane-associated RSD-3 is important for RNAi induced by extracellular silencing RNA in both somatic and germ cells of Caenorhabditis elegans

RNA silencing signals in C. elegans spread among cells, leading to RNAi throughout the body. During systemic spread of RNAi, membrane trafficking is thought to play important roles. Here, we show that RNAi Spreading Defective-3 (rsd-3), which encodes a homolog of epsinR, a conserved ENTH (epsin N-terminal homology) domain protein, generally participates in cellular uptake of silencing RNA. RSD-3 is previously thought to be involved in systemic RNAi only in germ cells, but we isolated several deletion alleles of rsd-3, and found that these mutants are defective in the spread of silencing RNA not only into germ cells but also into somatic cells. RSD-3 is ubiquitously expressed, and intracellularly localized to the trans-Golgi network (TGN) and endosomes. Tissue-specific rescue experiments indicate that RSD-3 is required for importing silencing RNA into cells rather than exporting from cells. Structure/function analysis showed that the ENTH domain alone is sufficient, and membrane association of the ENTH domain is required, for RSD-3 function in systemic RNAi. Our results suggest that endomembrane trafficking through the TGN and endosomes generally plays an important role in cellular uptake of silencing RNA.

Through a previous genetic screen to search for genes involved in systemic RNAi, namely rsd (RNAi Spreading Defective) screen, rsd-3 alleles have been isolated 16 . rsd-3 encodes a homolog of epsinR (epsin-related protein), an ENTH domain protein involved in endomembrane trafficking in mammalian cells 17,18 . In the previous report, following dsRNA feeding, rsd-3 mutants showed defects in RNAi in the germline, but were not defective in RNAi in the soma 16 . Thus, RSD-3 is assumed to be a germ cell-specific component of systemic RNAi, and to date, the function of RSD-3 in systemic RNAi is poorly characterized. In the present work, we isolated new alleles of rsd-3, and revealed that RSD-3 is involved in systemic RNAi not only in germ cells but also in somatic cells, indicating that RSD-3 is a general component of systemic RNAi. In addition, we found that RSD-3 is required for importing dsRNA into cells rather than exporting from cells. We also found that RSD-3 is associated with endomembranes such as the trans-Golgi network (TGN) and endosomes, and membrane association of the ENTH domain of RSD-3 is required for function. From these results, we suggest that endomembrane trafficking through the TGN and endosomes generally plays an important role in cellular uptake of silencing RNA.

RSD-3 is required for systemic RNAi in both germ cells and somatic cells.
To understand the role of membrane trafficking in the intercellular transport of silencing RNA in C. elegans, we searched for mutants of membrane trafficking-related genes exhibiting defects in systemic RNAi spreading. We assayed various mutants of endocytosis, exocytosis or intracellular membrane trafficking-related genes for their sensitivity to RNAi induced by feeding transformed bacteria expressing dsRNA (feeding RNAi). Among 61 strains examined (Supplementary Table S1), we found that the RB733 strain showed resistance to feeding RNAi. The RB733 strain contains a deletion (ok498) in the ctbp-1 gene, a homolog of C-terminal binding protein 1 (CtBP1), which is involved in macropinocytosis in mammals 19 . We found that the RB733 strain showed resistance to RNAi when fed on dsRNA against both germline expressed gene pos-1 and soma expressed gene bli-3, an epithelial gene ( Supplementary Fig. S1). However, other ctbp-1 mutants (tm6130 and tm6188) did not phenocopy the RB733 strain ( Supplementary Fig. S1) and the defects of the RB733 strain were not rescued by expression of ctbp-1, suggesting the existence of a mutation other than ctbp-1(ok498) in the RB733 strain. We performed whole-genome sequencing of the RB733 strain and identified a large deletion in the rsd-3 gene. This mutation was confirmed by Sanger sequencing and revealed a 7809 bp deletion and 7 bp insertion located in the genomic region of rsd-3. We named this mutation tm9006 (Fig. 1a). rsd-3 and ctbp-1 are both located on chromosome X, 10.9 cM apart from each other. By crossing the RB733 strain with the wild type, we obtained segregants that contained only the ctbp-1(ok498) mutation or only the rsd-3(tm9006) mutation. Mutants that contained only rsd-3(tm9006) were resistant to both pos-1 and bli-3 feeding RNAi ( Fig. 1b-d), but mutants that contained only ctbp-1(ok498) showed a normal response to feeding RNAi ( Supplementary Fig. S1b,c), suggesting that the rsd-3(tm9006) mutation is essentially responsible for the systemic RNAi defects observed in the RB733 strain.
rsd-3 encodes a homolog of epsinR, an evolutionarily conserved epsin family protein 20,21 . RSD-3/epsinR contains an N-terminal ENTH domain, a well-characterized phosphoinositide binding module 20,21 , and motifs for binding clathrin and clathrin adaptor proteins in the C-terminal region (Fig. 1a). In mammalian cells, epsinR has been implicated in clathrin-mediated membrane trafficking between the trans-Golgi network (TGN) and endosomes 17,18,[22][23][24] . Previously, Tijsterman et al. performed a forward genetic screen named rsd (RNAi Spreading Defective) screen, and rsd-3 alleles such as pk2013 were isolated 16 . In their report, following feeding of dsRNA, rsd-3(pk2013) showed defects in transmitting the RNAi effect to the germline, but show a normal RNAi response in somatic tissues. We also confirmed the phenotypes of rsd-3(pk2013); i.e., rsd-3(pk2013) showed no RNAi phenotype when fed pos-1 dsRNA, but displayed RNAi phenotype when fed bli-3 dsRNA ( Fig. 1b and Supplementary  Fig. S2e). Because rsd-3(tm9006) exhibits systemic RNAi defects in both the germline and the soma, we examined whether RSD-3 actually plays a role in systemic RNAi in somatic tissue. We generated a transgene containing rsd-3 genomic sequences including 4 kb of upstream promoter sequences, the complete ORF and introns, fused in frame to GFP (see Fig. 2a for a schematic representation of the construct), and introduced this transgene into rsd-3(tm9006). We found that expression of RSD-3::GFP completely rescued the silencing defects in the photographed. Wild type animals that were fed on food with bli-3 dsRNA showed severe molting defects and resulted in larval arrest, whereas almost all rsd-3(tm9006) worms reached adulthood. A portion of rsd-3(tm9006) worms showed mild cuticle blister formation (inset, arrowhead). Scale bars, 500 μ m. (d) Bars represent the percentage of animals reached adulthood. Expression of RSD-3::GFP under the control of the rsd-3 promoter (see Fig. 2a for a schematic representation of the construct) rescues the phenotype of rsd- 3(tm9006). (e) rsd-3 mutants show resistance to unc-22 feeding RNAi. Bars represent the percentage of twitching animals. (f) rsd-3 mutants show resistance to elt-2 feeding RNAi. Bars represent the percentage of animals reached adulthood. (g) rsd-3(tm9006) worms show resistance to transgene-induced systemic RNAi. Expression of gfp hairpin RNA and mCherry in the neurons induces silencing of GFP signal (LET-858::GFP) in most cells of wild type animals, but not in rsd-3(tm9006) background (upper panels). snb-1p::mCherry indicates the neurons that express gfp hairpin RNA (lower panels). Scale bars, 100 μ m. Data are shown as mean ± SEM of three separate experiments. * P < 0.001 (Student's t-test, two-tailed). soma of rsd-3(tm9006) (Fig. 1d,e). To provide further proof for the gene identity of rsd-3, we isolated other rsd-3 mutants, tm6416 and tm6420 (Fig. 1a), and found that these mutants showed systemic RNAi defects not only in the germline (Fig. 1b) but also in the soma (Fig. 1d-f) as was the case with rsd-3(tm9006). From these results, we conclude that RSD-3 is required for systemic RNAi in somatic cells as well as in germ cells.
Because transgene-induced silencing information is also transported between cells 9 , we also examined the response of rsd-3(tm9006) to RNAi triggered by a transgene. We generated an RNAi-inducing transgene expressing gfp hairpin RNA and mCherry under the control of the panneuronal snb-1 promoter and introduced this transgene into let-858::gfp transgenic animals, which express nuclear-localized GFP in all somatic cells 25 . In the wild type, neuronally expressed gfp hairpin RNA spread to other tissues and potently silenced the GFP signal in most cells, although it did not affect GFP expression in the neurons as reported previously 26 (Fig. 1g, upper left panel). In contrast, GFP expression was not prominently silenced in any of the somatic cells with the rsd-3(tm9006) background (Fig. 1g, upper right panel). These results indicate that RSD-3 is involved in transport of both environmentally-and transgene-derived silencing signals between cells.
We also investigated why the originally identified rsd-3(pk2013) exhibited no defects in systemic RNAi in somatic tissues. In pk2013, Tc1 transposon (a 1.6-kb repeated DNA sequence) is inserted within the first exon of rsd-3, immediately after the start codon 16 . To determine whether the pk2013 mutation affects the expression or function of RSD-3 in the soma, we constructed a pk2013 mutation-containing rsd-3, amplified from the rsd-3(pk2013) genome, with a C-terminal fusion to GFP under the control of the rsd-3 promoter (genomic rsd-3_pk2013::gfp) (see Supplementary Fig. S2a for a schematic representation of the construct), and generated an extrachromosomal array in rsd-3(tm9006). We detected GFP expression in all somatic tissues, and found that the subcellular localization of the GFP signal was indistinguishable from that of normal RSD-3::GFP ( Supplementary  Fig. S2b-d and see below). Furthermore, expression of genomic rsd-3_pk2013::gfp completely rescued the systemic RNAi defects in the soma of rsd-3(tm9006) (Supplementary Fig. S2e). These results indicate that the pk2013 mutation did not strongly affect RSD-3 expression and function in the soma. As reported for some Tc1 insertions 27,28 , in somatic cells of rsd-3(pk2013), Tc1 may be excised from the genome, or may be removed from pre-mRNA by splicing, resulting in the production of an in-frame message to produce RSD-3 protein that is functional in vivo.
We tested this hypothesis by analyzing the sequence of rsd-3 cDNA derived from the Tc1-containing allele of rsd-3 in somatic cells. We used an extrachromosomal array of genomic rsd-3_pk2013::gfp described above, because transgene expression from an extrachromosomal array is typically restricted to somatic cells 25 . mRNA from the wild type animals, rsd-3(tm9006) and rsd-3(tm9006);Ex[genomic rsd-3_pk2013::gfp] were reverse transcribed, and rsd-3 cDNAs around the Tc1 insertion site were PCR amplified (see Supplementary Fig. S3a,b). As shown in Supplementary Fig. S3b, no PCR product was observed from rsd-3(tm9006) as expected. The size of the PCR product from rsd-3(tm9006);Ex[genomic rsd-3_pk2013::gfp] was almost the same as that from wild type, suggesting that most of Tc1 is spliced from mature mRNA of genomic rsd-3_pk2013::gfp or imprecisely excised from the extrachromosomal array ( Supplementary Fig. S3b). Then, we cloned the PCR product from rsd-3(tm9006);Ex[genomic rsd-3_pk2013::gfp] and sequenced five independent cDNA clones (#1~5). As shown in Supplementary Fig. S3c, of the five cDNA clones, three clones (#1~3) have 48 bp insertion sequences at the site corresponding to genomic Tc1 insertion site. These insertions do not lead to premature stop codons, and thus, the three clones maintain the normal rsd-3 translational reading frame. The predicted amino acid sequences translated from such mRNAs contain extra 16 amino acids at the site 11 amino acids downstream from the N-terminus. This insertion site is located upstream of the functionally important region of the ENTH domain (see below), and therefore, such insertion may not significantly affect the function of RSD-3. Meanwhile, two clones (#4 and #5) have a small deletion and/or insertion, resulting in a frameshift with a premature stop codon ( Supplementary Fig. S3c). These results suggest that the Tc1-containing allele of rsd-3 produces functional RSD-3 in addition to some non-functional RSD-3 in somatic cells.
Scientific RepoRts | 6:28198 | DOI: 10.1038/srep28198 pattern of the tagged RSD-3 protein very likely reflects those of endogenous protein. RSD-3::GFP fusion protein was expressed in all somatic tissues including the pharynx, neurons, excretory canal, coelomocytes, intestine, body-wall muscles, hypodermis and seam cells ( Fig. 2b-g). The RSD-3::GFP fusion protein was diffusely distributed in the cytoplasm and also localized in a punctate pattern in most tissues, indicating that RSD-3 associates with some organelles or vesicles (see the next section). Expression of RSD-3 was previously reported to be high in the coelomocytes 16 , but in our observation, the expression level of RSD-3::GFP in coelomocytes was comparable with those in other tissues. The discrepancy may be due to a difference in the length of rsd-3 upstream sequences used to generate the transgenes.
The subcellular localization of RSD-3 in the coelomocytes was also checked with double labeling of RSD-3::mCherry (see Supplementary Fig. S4a for a schematic representation of the expression construct) with endocytic compartment GFP-markers. RSD-3::mCherry partially overlapped with, or was juxtaposed to the early endosome marker 2xFYVE::GFP, the late endosome marker GFP::RAB-7 and the recycling endosome marker GFP::RME-1 ( Supplementary Fig. S4b-d, arrowheads). RSD-3::mCherry also formed ring-like structures surrounding the medial Golgi marker AMAN-2::GFP-positive puncta ( Supplementary Fig. S4e, arrowheads), but RSD-3::mCherry did not overlap with the lysosome marker LMP-1::GFP ( Supplementary Fig. S4f). These results indicate that the localization pattern of RSD-3 in the coelomocytes is similar to that in the hypodermis. In cultured mammalian cells, human epsinR also localizes primarily to the TGN with a minor portion to the endosomal compartment 17,23 . Thus, in addition to the sequence similarity, the subcellular localization of RSD-3 is very similar to that of human epsinR.

RSD-3 is not an integral part of the RNAi machinery.
Because the observed phenotypes of rsd-3 mutants might be due to disruption of cellular RNAi machinery itself rather than to a systemic RNAi defect, we examined whether RSD-3 plays a role in cell-autonomous RNAi. Tijsterman et al. showed that, by injection of pos-1 dsRNA into one gonad arm, cell-autonomous RNAi machinery is intact in germ cells of rsd-3(pk2013) 16 . We also injected various concentrations of pos-1 dsRNA into both gonad arms of rsd-3(tm9006) and found that RNAi efficiency in germ cells of rsd-3(tm9006) is comparable to that in the wild type ( Supplementary Fig. S5). This result confirmed that RSD-3 is not involved in cell-autonomous RNAi machinery in germ cells, in contrast to RSD-2 and RSD-6, which were also identified in the rsd screen 16 but turned out to be involved in cell-autonomous RNAi machinery, in particular, secondary siRNA amplification 26 .
To examine the involvement of RSD-3 in the RNAi mechanism in somatic cells, we generated an extrachromosomal array expressing gfp hairpin RNA and mCherry under the control of the muscle-specific myo-3 promoter. This transgene was introduced into myo-3p::gfp transgenic animals (ccIs4251), which express GFP in the muscle nuclei and mitochondria. In the wild type background, GFP was largely silenced in red muscle cells expressing the GFP-RNAi construct (Supplementary Fig. S6). GFP expression was reduced equally in red muscle cells with an rsd-3(tm9006) background ( Supplementary Fig. S6), suggesting that the RNAi pathway is not compromised in rsd-3(tm9006) muscle cells. These results indicate that RSD-3 is not involved in the RNAi machinery itself, but rather required for the systemic spread of silencing RNA in both somatic and germ cells.
RSD-3 is involved in the import but not the export of silencing RNA. When C. elegans is fed with dsRNA, the dsRNA is first taken up by intestinal cells from the intestinal lumen. The silencing RNA is then exported from intestinal cells to the pseudocoelom and subsequently imported into the germ and somatic cells 32 . To determine which step of feeding RNAi is impaired in rsd-3(tm9006), we performed tissue-specific rescue experiments. We used myo-3p::gfp transgenic animals (ccIs4251) and fed them with bacteria expressing GFP dsRNA. In the wild type background, GFP expression in muscle cells was significantly decreased, but in muscle cells with an rsd-3(tm9006) background, GFP was not silenced prominently after exposure to GFP dsRNA (Fig. 3a). We examined whether expression of rsd-3 cDNA in the body wall muscle or in the intestine rescues the feeding RNAi defect caused by rsd-3(tm9006). We found that expression of RSD-3 and mCherry under the control of the muscle-specific myo-3 promoter efficiently restored GFP feeding RNAi sensitivity in muscle cells of rsd-3(tm9006);ccIs4251 animals ( Fig. 3b). Especially, GFP was silenced in RSD-3(+ ) red muscle cells (Fig. 3b, arrows) but not in RSD-3(-) non-red muscle cells (Fig. 3b, arrowhead). On the other hand, feeding RNAi defect was not rescued in muscle cells of rsd-3(tm9006);ccIs4251 expressing RSD-3 and DsRed under the control of the intestine-specific ges-1 promoter (Fig. 3c). These results indicate that RSD-3 is not required for the first uptake of silencing RNA into intestinal cells from the intestinal lumen or export from intestinal cells to the pseudocoelom, but rather is required for subsequent import into each cell from the pseudocoelom. To confirm this conclusion, we injected pos-1 dsRNA (10 ng/μ l) into the pseudocoelom. In wild type animals, the injection induced efficient RNAi in the germline, and resulted in almost complete embryonic lethality in the F1 embryos (Fig. 3d). In contrast, rsd-3(tm9006) animals were partially resistant to the injection (Fig. 3d). This result supports the view that RSD-3 is required for the import of silencing RNA into cells from the pseudocoelom.
The ENTH domain is necessary and sufficient to rescue the systemic RNAi defects of rsd-3 mutants. Human epsinR possesses an N-terminal ENTH domain with affinity for membrane phosphoinositides, especially for phosphatidylinositol 4-phosphate (PI4P) that is mainly generated on the TGN membrane ( Fig. 1a) 17,24 . The unstructured C-terminal region of human epsinR contains motifs for binding clathrin and clathrin adaptor proteins, such as AP-1 (adaptor protein-1) and GGAs (Golgi-localized, Gamma-ear-containing, Arf-binding proteins), functioning at the TGN and endosomes (Fig. 1a) [33][34][35] . These domain/motifs have been proposed to target human epsinR to the TGN and endosomes. In addition, the ENTH domain of human or rat epsinR interacts with specific endosomal soluble NSF attachment protein receptors (SNAREs) [36][37][38] . Thus, mammalian epsinR has been implicated in clathrin-mediated membrane trafficking between endosomes and the TGN, and particularly functions in sorting of cargoes into transport vesicles 36,37,39 . RSD-3 possesses domain and motif structures similar to those of human epsinR, except for the methionine-rich domain of unknown function at the  extreme C-terminus of human epsinR (Fig. 1a). A sequence alignment of RSD-3 and several epsinR-related proteins shows that the ENTH domain of RSD-3 displays high sequence conservation, whereas the overall sequence conservation in the C-terminal region of RSD-3 is relatively low (Fig. 4a).
As a step toward understanding the role of RSD-3 in systemic RNAi, we examined which parts of RSD-3 are required for restoring feeding RNAi sensitivity of rsd-3(tm9006). We generated expression plasmids for ECFP-tagged full-length RSD-3 and RSD-3 truncations, namely the ENTH domain (RSD-3_ENTH; amino acids 1-189) and the C-terminal region (RSD-3_Cterm; amino acids 190-483) (Fig. 4b). We expressed these constructs separately in the hypodermis of rsd-3(tm9006) and examined the ability to restore sensitivity to bli-3 feeding RNAi. We also observed the distribution of these constructs in hypodermal cells. Expression of full-length RSD-3::ECFP in the hypodermis completely rescued the bli-3 feeding RNAi defects of rsd-3(tm9006) (Fig. 4b), and RSD-3::ECFP fluorescence was observed in a punctate pattern (Fig. 4c). Also, expression of RSD-3_ENTH::ECFP alone fully rescued the defects of bli-3 feeding RNAi in rsd-3(tm9006) (Fig. 4b). RSD-3_ENTH::ECFP had a relatively diffuse distribution in hypodermal cells, and also localized in the nucleus, but still localized in puncta just like full-length RSD-3::ECFP (Fig. 4d). In contrast, expression of RSD-3_Cterm::ECFP could not rescue the phenotype of rsd-3(tm9006) at all (Fig. 4b), and in hypodermal cells, while a small portion of RSD-3_Cterm::ECFP was localized in the puncta, most of it was distributed diffusely in the cytoplasm and nucleus (Fig. 4e). These results indicate that the ENTH domain is necessary and sufficient for RSD-3 function in systemic RNAi. It should be noted that, although the C-terminal region has some contribution to RSD-3 localization, only the ENTH domain can be sufficient for targeting RSD-3 to the appropriate subcellular compartments.
The region at the N-terminus of mammalian ENTH domain, which is unstructured when not bound to the lipid ligand, folds an α -helix referred to as helix 0 (α 0) upon binding to phosphoinositide 17,[40][41][42] . This additional helix is inserted between the lipid head groups due to the hydrophobic residues exposed on the outer surface of helix, and therefore, α 0 is required for lipid interaction of mammalian ENTH domain 40,41 . The N-terminal region of RSD-3 ENTH domain also contains several hydrophobic residues at the positions corresponding to the outer surface of α 0 helix (Fig. 4a, blue boxes), suggesting that the role of α 0 is conserved in C. elegans RSD-3. Meanwhile, mammalian ENTH domain proteins are thought to have other functions not directly related to membrane trafficking, such as transcriptional regulation in the nucleus 43 . Then, to examine whether membrane interaction is required for the function of RSD-3 ENTH domain in systemic RNAi, we performed rescue experiments with ECFP-tagged RSD-3_ENTHΔ α 0 (amino acids , in which the putative α 0 was deleted (Fig. 4b).
As expected, RSD-3_ENTHΔ α 0::ECFP rarely associated with puncta, but instead uniformly distributed in the cytosol and the nucleus in hypodermal cells (Fig. 4f), confirming that this truncation mutant cannot interact with membrane. We found that expression of RSD-3_ENTHΔ α 0::ECFP could not restore bli-3 feeding RNAi sensitivity of rsd-3(tm9006) at all (Fig. 4b). This result strongly suggests that RSD-3 functions in systemic RNAi by regulating membrane trafficking.

Discussion
Our data indicates that endomembrane-associated RSD-3 generally plays an important role in cellular uptake of silencing RNA. It is known that plasma membrane-localized ENTH domain proteins, such as epsin1/ EPN-1, regulate clathrin-mediated endocytosis 21,44 , whereas the TGN/endosome-localized epsinR regulates clathrin-mediated intracellular membrane trafficking. Because RSD-3 appears to be an ortholog of human epsinR, it is conceivable that RSD-3 is involved in intracellular membrane trafficking rather than endocytosis. Both human epsinR and Ent3p, an epsinR ortholog in Saccharomyces cerevisiae, are reported to be involved in retrograde transport from endosomes to the TGN 18,45 , thus RSD-3 may also function in this transport pathway. In addition, the systemic RNAi defects in rsd-3 mutants were partial, especially in the soma, indicating the existence of functionally redundant genes or pathways.
The ENTH domain, which lacks clathrin and clathrin adaptor protein-binding motifs, is necessary and sufficient for RSD-3 function in systemic RNAi, suggesting that RSD-3 function is not mediated by formation of clathrin-coated vesicles. Instead, because the ENTH domain of mammalian epsinR/Ent3p interacts with some SNAREs such as Vti1b/Vti1p and functions as a cargo adaptor [36][37][38][39]46 , RSD-3 may also function as a cargo adaptor for transporting some cargoes to the appropriate compartment. We generated mutants of vti-1, a homolog of Vti1b/Vti1p, but vti-1 mutants did not show systemic RNAi defects (our unpublished observation). Other SNAREs may function redundantly with vti-1.
Scientific RepoRts | 6:28198 | DOI: 10.1038/srep28198 The mechanism by which RSD-3 is involved in the uptake of silencing RNA into cells remains unclear. At present, we suggest the following two possibilities. First, RSD-3 may regulate SID-1 localization. Several plasma membrane transporters, such as GLUT4 (glucose transporter) and Fet3p/Ftr1p (iron transporter), are known to cycle between the plasma membrane and the TGN, via endosomes 47,48 . SID-1, which is considered as a plasma membrane-localized dsRNA transporter, may also cycle between the plasma membrane and intracellular compartments via retrograde transport from endosomes to the TGN, and RSD-3 may be involved in this transport pathway and required for the proper localization of SID-1. Second, RSD-3 may regulate the intracellular transport of endocytosed silencing RNA. In this case, SID-1 may transport silencing RNA into the cytosol at an intracellular compartment but not at the plasma membrane. In fact, SID-1::GFP was observed at intracellular compartments as well as the cell periphery 9 . In addition, endomembrane localization of SIDT1 and SIDT2, mammalian homologs of SID-1, was reported 49,50 . Growing evidence indicates that RNA silencing is connected to the endomembrane system 2 . Particularly, exogenous siRNAs are loaded into RNA-induced silencing complexes (RISCs) at the cytosolic membrane surface of the rough endoplasmic reticulum (ER) 51 . Thus, it is conceivable that endocytosed silencing RNA arrives at the appropriate compartment, such as ER, by RSD-3-mediated retrograde trafficking, then enters the cytosol through SID-1, and is immediately incorporated to the RNAi machinery. Such a mechanism may be important for the efficient transfer of exogenously delivered silencing RNA into the intracellular RNAi machinery.
Oligonucleotides hold outstanding promise as potential therapeutic agents, but a major concern is their poor accessibility to the targets within the cells. Recently, some studies using mammalian cells suggested that the uptake and trafficking pathway of oligonucleotide could affect the ultimate pharmacological effectiveness [52][53][54] . Thus, future studies to elucidate the role of RSD-3 in the trafficking of silencing RNA may help to develop an efficient drug delivery system in oligonucleotide therapeutics.

Identification of rsd-3(tm9006). RB733 genomic DNA was purified by phenol and chloroform extraction
including RNase A treatment. 1 μ g DNA was subjected to fragmentation and adaptor ligation using an Ion Xpress Plus Fragment Library Kit (Life Technologies), according to the manufacturer's instructions. The library was subjected to emulsion PCR using the Ion OneTouch 200 Template Kit v2 DL (Life Technologies), followed by bead enrichment. Then, whole-genome sequencing was performed with an Ion Torrent PGM system using the Ion PGM 200 Sequencing Kit and Ion 318 chip (Life Technologies). Using IGV (Integrative Genomics Viewer), we inspected the genome region where no read was aligned as the potential deletion candidates. Except for ok498, we found two potential deletions larger than 1 kb on chromosome X. One of the two genomic regions contains no protein-coding gene, thus we focused on another region, around chromosome X: + 6.73 cM. This mutation candidate was confirmed by Sanger sequencing and revealed a 7809 bp deletion and 7 bp insertion located in the genomic region of rsd-3.

RNAi experiments.
Feeding RNAi was carried out as described 60 . RNAi clones were transformed into E. coli HT115(DE3), and the bacteria was grown on 100 μ g/mL ampicillin and 1 mM isopropyl-beta-D-thiogalactopyranoside (IPTG). RNAi clones for pos-1, mex-3, lin-31, bli-3 and unc-22 were taken from the Ahringer RNAi library (GeneService). RNAi clones for elt-2 and GFP were generated as following: cDNA fragment of elt-2 was amplified from N2 cDNA. cDNA fragment of GFP was cut from pPD79.44 using NotI/BglII sites. Then, cDNA fragments of elt-2 or GFP were cloned into the L4440 (pPD129.36) vector, and transformed E. coli HT115 (DE3) with each construct. For pos-1 and mex-3 RNAi assay, more than twenty L1 larvae were fed with bacteria expressing pos-1 or mex-3 dsRNA, reared to adulthood and allowed to lay eggs for several hours. Then adult animals were removed and 24 hours later, the number of dead eggs and hatched larvae were counted. At least 300 eggs and larva were counted per strain. For lin-31 RNAi assay, L1 larvae were placed on plates seeded with bacteria expressing lin-31 dsRNA and 54 hr later, the percentage of animals with Muv (multi-vulva) phenotype was scored. At least 50 animals were counted per strain. For bli-3 RNAi assay, L1 larvae were placed on plates seeded with bacteria expressing bli-3 dsRNA and 48 hr later (wild type animals reach adulthood), the percentage of animals that reached adulthood was scored. At least 100 animals were counted per strain. For unc-22 RNAi assay, L4 larvae were placed on plates seeded with bacteria expressing unc-22 dsRNA and 48 hr later, the percentage of twitching animals was scored. At least 100 animals were counted per strain. For elt-2 RNAi assay, L4 larvae were placed on plates seeded with bacteria expressing elt-2 dsRNA, reared to adulthood and allowed to lay eggs for several hours. Then adult animals were removed and 54 hours later, the percentage of animals that reached adulthood was scored. At least 100 animals were counted per strain. For GFP RNAi, L1 larvae were fed with bacteria expressing GFP dsRNA, reared to L4 larvae and photographed. For injection RNAi experiments, pos-1 dsRNA was synthesized as following: pos-1 cDNA fragment attached with T7 promoter at both ends was amplified from N2 cDNA, and pos-1 dsRNA was transcribed using T7 polymerase (Ambion). Synthesized pos-1 dsRNA was treated with DNase I, purified by phenol and chloroform extraction, and then injected into both gonad arms or the pseudocoelom of more than twenty day 1 adult hermaphrodites for each strain. The hatching rate of embryos laid between 12 to 24 hr post injection was scored for each injected animal. At least 300 progenies were counted per strain. For transgene-induced RNAi, gfp hairpin RNA was expressed from extrachromosomal arrays and GFP fluorescence was photographed at the L4 stage.
Sequence analysis of cDNAs derived from genomic rsd-3_pk2013::gfp transgene. Total RNA was extracted from mixed-stage tm9006;tmEx3681[genomic rsd-3_pk2013::gfp] animals using TRIzol reagent according to the protocol supplied by the manufacturer (Invitrogen). First-strand cDNA was revers transcribed from the total RNA with an oligo-dT primer using SuperScript IV (Invitrogen). rsd-3 cDNAs in the vicinity of the Tc1 insertion site were PCR amplified using primers shown in Supplementary Fig. S3a. The cDNA amplification products were cloned using TOPO TA Cloning Kit (ThermoFicher Scientific). Five cDNA clones were sequenced in the vicinity of the Tc1 insertion site.
Microscopy. Worms on NGM (nematode growth medium) plates were imaged with a Leica DFC280 camera using Leica IM500 imaging software. For confocal images, animals in 20 mM azide were mounted on a 5% agar pad on a glass slide. Micrographs were taken on a Zeiss LSM 710 inverted confocal microscope with 488 and 543 lasers and images were processed using ZEN software (Carl Zeiss) except for Supplementary Fig. S6. For Supplementary Fig. S6, micrographs were taken on a Leica TCS SP8 inverted confocal microscope with 488 and 552 lasers and HyD detectors, and images were processed using LAS AF software (Leica).
Statistical analysis. The standard error of the mean (SEM) was used as the error bar for bar charts plotted from the mean value of the data. All the data were compared using two-tailed Student's t-test. Data were considered significantly different if P-values were lower than 0.05.