An EHBP-1-SID-3-DYN-1 axis promotes membranous tubule fission during endocytic recycling

The ACK family tyrosine kinase SID-3 is involved in the endocytic uptake of double-stranded RNA. Here we identified SID-3 as a previously unappreciated recycling regulator in the Caenorhabditis elegans intestine. The RAB-10 effector EHBP-1 is required for the endosomal localization of SID-3. Accordingly, animals with loss of SID-3 phenocopied the recycling defects observed in ehbp-1 and rab-10 single mutants. Moreover, we detected sequential protein interactions between EHBP-1, SID-3, NCK-1, and DYN-1. In the absence of SID-3, DYN-1 failed to localize at tubular recycling endosomes, and membrane tubules breaking away from endosomes were mostly absent, suggesting that SID-3 acts synergistically with the downstream DYN-1 to promote endosomal tubule fission. In agreement with these observations, overexpression of DYN-1 significantly increased recycling transport in SID-3-deficient cells. Finally, we noticed that loss of RAB-10 or EHBP-1 compromised feeding RNAi efficiency in multiple tissues, implicating basolateral recycling in the transport of RNA silencing signals. Taken together, our study demonstrated that in C. elegans intestinal epithelia, SID-3 acts downstream of EHBP-1 to direct fission of recycling endosomal tubules in concert with NCK-1 and DYN-1.

Mammalian recycling endosomes consist of membrane tubules and discrete vesicles [1]. Similarly, the recycling endosomal network in the C. elegans intestine is highly tubular [15]. Lack of RAB-10 or EHBP-1 leads to the collapse of the endosomal meshwork, suggesting that RAB-10 and EHBP-1 function jointly to generate and/or maintain endosomal tubules [10,12]. In agreement with this observation, RAB-10 is located at the tips of newly formed tubules and functions in concert with the exocyst to direct the extension and tethering of membrane tubules [8]. Like other types of endosomes, recycling endosomes undergo fission to generate tubular carriers [9]. However, the mechanism coordinating recycling endosomal tubulation and membrane fission has been a long-standing mystery, primarily because the link between tubule biogenesis/maintenance and subsequent scission is unknown.
ACK (activated CDC42-associated kinase) is a non-receptor tyrosine kinase that was initially identified as a CDC42 effector protein [16]. ACK family kinases contain an N-terminal tyrosine kinase domain, an SH3 (src homology 3) domain, a CRIB (CDC42/RAC interacting binding) domain, and a highly variable C-terminal region [17]. ACK is found in clathrincoated pits [18] and interacts with the clathrin heavy chain [19]. Several ACK-interacting proteins have been reported, including the adaptor protein NCK [20][21][22]. During sperm differentiation, ACK is required for the subcellular distribution of Dock, the Drosophila homolog of NCK [22]. NCK harbors three SH3 domains and a C-terminal SH2 domain [23], and the interaction between ACK and NCK is mediated by the SH2 domain [24]. Importantly, NCK was found to interact with dynamin via the third SH3 domain [25], suggesting that NCK and dynamin could function in the same pathway.
The C. elegans genome encodes two ACK family tyrosine kinases: SID-3 and ARK-1 (ACK related kinase-1). Phylogenetic and functional analysis suggested that SID-3 is the sole homolog of the mammalian ACK [26]. SID-3 is broadly expressed and localizes to cytosolic puncta [26]. Functional analysis indicated that SID-3 and its mammalian homolog ACK encourage double-stranded RNA (dsRNA) uptake, and this efficacy appears to involve clathrin-mediated endocytosis [19,26]. In the present study, we identified SID-3 as a novel interactor of EHBP-1 and highlighted the necessity of EHBP-1 for the endosomal localization of SID-3. Loss of SID-3 led to a severe defect in the recycling of the clathrin-independent cargo hTAC-GFP. Our results also suggested that SID-3 functions downstream of RAB-10 to facilitate recycling in concert with NCK-1/NCK and DYN-1/dynamin. Accordingly, in sid-3 mutants, the detachment of membrane tubules from recycling endosomes largely did not occur, and the overexpression of DYN-1 effectively restored the recycling of hTAC-GFP. Together, our findings demonstrate that EHBP-1 represents a point of convergence between RAB-10-directed endosomal tubulation and SID-3-coordinated membrane fission and promotes tubular recycling carrier formation.

Loss of SID-3 causes endocytic recycling defects
In the current study, we focused on basolateral recycling transport in the C. elegans intestine, a polarized epithelial tube (Fig 1B). The apical membrane faces the digestive lumen and is responsible for nutrient uptake. The basolateral membrane faces the pseudocoelom and is responsible for the exchange of molecules between the intestine and other tissues. In the C. elegans intestine, the basolateral sorting endosomes are enriched in the recycling regulator RAB-10 [8,12]. RAB-10 is predominantly involved in the recycling of CIE cargos [10,27]. Through its effectors, EHBP-1 and CNT-1, RAB-10 bridges endosomes with F-actin and manipulates the level of endosomal PI(4,5)P2 (phosphatidylinositol 4,5-bisphosphate) to ensure proper recycling transport [5,12]. However, the mechanistic processes underlying recycling regulation are still poorly understood. To address this, we conducted a genome-wide RNAi screen to isolate genes that when silenced cause overaccumulation of the CIE cargo hTAC in the deep cytosol of intestinal epithelia (please see Materials and methods for details). After a preliminary screen and subsequent validation, we found that knockdown of SID-3/ACK resulted in prominent intracellular deposition of hTAC-GFP.
To confirm the role of SID-3 in recycling, we employed two mutant alleles: sid-3(ok973) and sid-3(gk436958). sid-3(ok973) harbors a premature termination codon at amino acid site 547 that leads to a 690-amino acid deletion at the C-terminus. sid-3(gk436958) carries a pointnonsense mutation at site 353, resulting in a deletion that spans part of the tyrosine kinase domain and the entire C-terminal region (Fig 1A). Through confocal stack imaging of intestinal epithelial cells from intact living animals (Fig 1B), we found that, in both sid-3 mutants, the number and size of hTAC-GFP-labeled intracellular structures increased significantly (~9-fold increase in total area) ( Fig 1C-1C''). This phenotype is typical of basolateral recycling mutants [5,12,[28][29][30]. We further assayed the accumulation of FGT-1, the worm homolog of another clathrin-independent recycling cargo, GLUT1 [8], and observed that SID-3 deficiency led to substantial intracellular aggregation of FGT-1-positive structures (~7.8-fold increase in total area) (S1A and S1A' Fig). We also examined the well-defined clathrin-dependent recycling cargo hTfR (human transferrin receptor) [10,31]. Loss of SID-3 caused accumulation of hTfR-GFP on the plasma membrane (~11.5-fold increase in total area) (S1B and S1B' Fig). Likewise, SID-3-depleted cells exhibited poor endocytic uptake of the clathrin-dependent retrograde cargo MIG-14 (~13.3-fold increase in Top total area) (S1B and S1B' Fig) [32]. Hence, our results indicate that, in addition to its previously described role in clathrin-dependent endocytosis [18,19], SID-3 is involved in the regulation of the recycling of clathrin-independent cargos.
Next, to assess where SID-3 functions in the recycling pathway, we compared the localization of hTAC-GFP with that of an early endosome marker, 2xFYVE [5]. hTAC-GFP was rarely localized in 2xFYVE-labeled early endosomes in the wild-type background (Pearson's Coefficient:~0.28) (Fig 1D and 1D'). However, in the absence of SID-3, the overlap between hTAC-GFP and 2xFYVE was substantially higher (Pearson's Coefficient:~0.603), indicating that cargo flow from early endosomes to recycling endosomes was blocked ( Fig 1D and 1D'). This observation also indicates that SID-3 modulates recycling at a step similar to that of RAB-10 and EHBP-1 [33].

Recycling transport remains unaffected in the absence of CDC-42 or ARK-1
Mammalian ACK is recognized as an effector of the Rho family small GTPase CDC42 [16,17,24]. Therefore, we investigated the effect of loss of CDC-42 function on endocytic recycling in C. elegans. Worms homozygous for the putative null allele cdc- 42(gk388) can only grow to the L3/L4 larval stages [34], this makes acquiring proper intestinal hTAC-GFP images challenging. Hence, we chose to use feeding cdc-42(RNAi) bacteria from the Arhinger library [35]. To test the knockdown efficiency of cdc-42 feeding RNAi, we utilized an animal expressing GFP-CDC-42. Through western blotting and fluorescence imaging, we observed that the cdc-42 feeding RNAi achieved a significant level of CDC-42 knockdown (S1C and S1C' Fig). To ascertain whether CDC-42/CDC42 directs the function of SID-3 during endocytic recycling, we examined the distribution of hTAC-GFP in CDC-42-knockdown cells (S1C and S1C' Fig). There were no differences in the subcellular localization of hTAC-GFP in CDC-42-deficient cells compared with control cells (Fig 1C-1C''). RME-1 has been extensively studied for its function in endocytic recycling in various tissues [11,15,36]. Consistent with the localization of hTAC-GFP, the number of GFP-RME-1-labeled recycling endosomes and the average intensity of GFP-RME-1 were not affected by the lack of CDC-42 (S1D and S1D' Fig) [15]. Moreover, there was only partial overlap between CDC-42 and RAB-10 (Pearson's Coefficient: 0.51; Mander's Coefficient:~0.56) (S1E and S1E' Fig). Together these results suggest that SID-3 is likely to act independently of CDC-42 during endocytic recycling in the C. elegans intestine.

EHBP-1 is required for the endosomal localization of SID-3
To assess the mechanism that determines the endosomal localization of SID-3, we examined the localization of SID-3-GFP in the absence of various recycling regulators. Remarkably, the lack of EHBP-1 led to substantial loss of SID-3 from punctate structures (Fig 3A and 3A'). A previous study showed that rab-10 and ehbp-1 mutants accumulated grossly enlarged early/ sorting endosomal vacuoles within intestinal cells [10,12]. Notably, SID-3 accumulated in enlarged endosomes and decorated the vacuoles in RAB-10-deficient cells, but no longer existed on the edges of the endosomal vacuoles labeled with the recycling regulator ARF-6 in ehbp-1(RNAi) cells (Fig 3A'', S3A Fig) [40], further supporting the hypothesis that EHBP-1 is required for the endosomal localization of SID-3. Consistent with the labeling of SID-3 on the vacuoles in RAB-10-depleted cells, there was no apparent interaction between SID-3 and the predicted constitutively active (GTPase-defective) form of RAB-10, RAB-10(Q68L) (S3B Fig). To directly evaluate the association of SID-3 with the membrane, we utilized ultracentrifugation to separate the cytosol from the membrane structures in worm lysates and assayed the amount of SID-3-GFP in each fraction by western blotting [41]. As expected, the membraneto-cytosol ratio of SID-3-GFP in ehbp-1(RNAi) animals dropped by~81% (Fig 3B and 3B').  We also determined whether RME-1 is required for SID-3 endosomal localization. In the rme-1(b1045) mutant background, in which intestinal vacuoles mainly consist of enlarged recycling endosomes [10,15], SID-3-GFP still labeled the recycling endosome vacuoles ( Fig 3A-3A''), indicating that the association of SID-3 with endosomes is not dependent on the presence of RME-1.
To investigate the role of different SID-3 domains in endosomal localization, we examined the distribution of GFP-tagged SID-3 truncations ( Fig 3H and 3H'). The fragment containing the C-terminal region (aa 545-1237) was enriched in the punctate structures ( Fig 3H). Surprisingly, the N-terminal fragment containing the tyrosine kinase, SH3, and CRIB domains (aa 1-544) was diffusely localized in the cytosol (Fig 3H). These results suggest that the central region of SID-3 downstream of the CRIB domain is mainly responsible for endosomal localization.

The kinase activity of SID-3 is indispensable for endocytic recycling
The SID-3 tyrosine kinase domain is essential for the transmission of RNAi signals [26]. To determine if kinase activity is required for the function of SID-3 in recycling transport, we prepared a kinase-dead form (K139A) and a constitutively active form (L509F) of SID-3 [24,42]. Like the wild-type SID-3, SID-3(K139A) also localized to EHBP-1-GFP-positive endosomal tubules (Pearson's Coefficient:~0.35) (S3C and S3C' Fig), indicating that kinase activity is not likely a prerequisite for SID-3 endosomal localization. We then tested the interaction of SID-3 (K139A) with the EHBP-1 C2-like domain. SID-3(K139A) displayed a comparable affinity for test ( ��� p<0.001). (B-B 0 ) The membrane-to-cytosol ratio of SID-3 decreased in ehbp-1(RNAi) animals. Membrane structures were separated from the cytosol of wild-type and ehbp-1(RNAi) worm lysates by ultracentrifugation at 100,000xg for 1 hour. SID-3-GFP in the supernatants and pellets were analyzed by western blotting using an anti-GFP antibody. The loading control was blotted by the anti-actin and anti-tubulin antibodies. Quantification of the membrane-to-cytosol ratio (P/S) of SID-3-GFP in wild-type and ehbp-1(RNAi) backgrounds. The SEMs from three independent experiments are shown ( ��� p<0.001, ns: no significance). the C2-like domain as wild-type SID-3 (S3D Fig). Next, we examined the ability of SID-3 mutant forms to complement SID-3-depleted cells. In the middle focal plane, GFP-RME-1 mainly labeled recycling endosomes close to the basolateral plasma membrane and only a limited amount of labeling was observed within the deep cytosol (total area:~88/unit region) (S3E and S3E ' Fig). In sid-3 mutants, a large number of GFP-RME-1-positive structures accumulated within the cytoplasm (~3.6-fold increase in total area), suggesting that SID-3 is essential for the integrity and distribution of recycling endosomes (S3E and S3E ' Fig).

EHBP-1 sequentially recruits SID-3, NCK-1, and DYN-1
In Drosophila, the SID-3 homolog ACK is required for the intracellular distribution of the adaptor protein Dock/NCK [22]. NCK harbors three SH3 domains and a C-terminal SH2 domain [23], and the interaction with ACK is mediated by the C-terminal SH2 domain [24]. Being a large GTPase, dynamin is capable of assembling into contractile helical polymers that sever the membrane to release the vesicle [43]. Interestingly, dynamin can be co-precipitated with anti-NCK antibody in mammals [25]. Consistent with a previous report [22], we found a significant reduction in GFP-NCK-1-labeled structures in sid-3(ok973) mutants (~2.5-fold decrease in total area) ( Fig 4A and 4A'), as would be expected if NCK-1/NCK fails to be recruited to endosomes. A similar reduction in DYN-1/dynamin labeling was observed in sid-3 (~6.5-fold decrease in total area) and nck-1 (~5.1-fold decrease in total area) mutants ( To determine whether NCK-1 and DYN-1 function in the regulation of recycling, we assessed the localization of hTAC-GFP in nck-1(ok694) and dyn-1 (RNAi) animals ( Fig 4C and 4C'). As expected, in NCK-1-or DYN-1-depleted cells, the aberrant distribution of hTAC-GFP completely resembled that resulting from a deficit of SID-3 (~11-to~15-fold increase in total area) ( Fig 1C-1C'').

Overexpression of DYN-1 alleviates the recycling defects in sid-3 mutants
Thus far, our findings indicate that the SID-3-NCK-1 complex is a critical link between EHBP-1 and DYN-1. To validate the necessity of DYN-1 for SID-3-mediated recycling transport, we examined the distribution of hTAC-GFP in sid-3(ok973) mutants expressing DYN-1-mCherry. The intracellular accumulation of hTAC was substantially reduced by the overexpression of DYN-1 (~10.5-fold decrease in total area) ( Fig 4E and 4E'). However, overexpression of SID-3 could not alleviate the intracellular overaccumulation of hTAC-GFP in dyn-1 (RNAi) animals (~1.26-fold increase in total area) (Fig 4E and 4E'). Because dynamin is known to function in membrane fission through membrane constriction [43,44], these observations suggest that the SID-3 deficiency-induced disruption of recycling could be due to defective endomembrane fission and the subsequent failure of recycling carrier formation.

RAB-10 functions upstream of SID-3 in the endocytic recycling pathway
RAB-10 and EHBP-1 work in concert to generate or maintain endosomal structures [5,12]. Loss of RAB-10 or EHBP-1 led to a notable breakdown of the RME-1-labeled basolateral endosomal meshwork, with tubular organization no longer being maintained (~7-fold decrease in total area) (Fig 5A and 5A'). To elucidate the genetic relationship between RAB-10 and SID-3 in the regulation of recycling endosome morphology, we examined the endosomal architecture in cells lacking both SID-3 and RAB-10. Similar to RAB-10-deficient cells, sid-3;rab-10 double-mutant cells had punctate structures labeled solely with RME-1, and the endosomal network no longer existed (~8.78-fold decrease in total area) (Fig 5A and 5A'), suggesting that RAB-10 acts upstream of SID-3 in the recycling pathway. In contrast, the pattern of RME-1-labeled structures in sid-3;nck-1 double mutants, which displayed a moderate reduction in endosomal tubules (~2.4-fold decrease in total area), closely resembled that in sid-3 mutants (Fig 5A and 5A'). In summary, our results suggest that RAB-10 functions upstream of SID-3 in the recycling pathway, possibly modulating the generation or maintenance of the tubular endosomal network together with EHBP-1.

Loss of SID-3 diminishes endosomal tubule fission
To better appreciate the mechanism by which SID-3 regulates basolateral recycling, we compared the distribution of hTAC-GFP with that of SID-3-mCherry and DYN-1-mCherry. We found that SID-3 and DYN-1 were often located on hTAC-GFP-positive endosomal tubules (~65% mCherry-puncta near tubular junctions) (Fig 5B and 5B'), prompting us to hypothesize that the SID-3-NCK-1-DYN-1 axis could participate in membrane tubule fission. To test  this hypothesis, we tracked endosomal dynamics and noted that membrane tubule dynamics consisted of four stages: budding, suspension, fission, and tubular carrier fusion with the opposite endosomal structures. In the wild-type background, hTAC-GFP-labeled tubules were highly dynamic, with frequent extension and detachment at the DYN-1-enriched endosomal domains (Fig 5C and 5C', S1 Video,~70.9% fission/budding events). However, in SID-3-depleted cells, the punctate labeling of DYN-1 was greatly reduced. Importantly, the remaining DYN-1-positive puncta rarely overlapped with the tubular recycling endosomes, and the extending tubules often underwent prolonged suspension and finally retracted (Fig 5C and  5C'; S2 Video,~26.08% fission/budding events).

Defects in basolateral recycling undermine the competency of feeding RNAi
Genetic analysis showed that lack of SID-3 leads to impaired RNAi efficiency [26]. Here we demonstrated that SID-3 plays a significant role in basolateral recycling. Thus, it is of interest to determine whether the reduced feeding RNAi efficiency in sid-3 mutants is due at least in part to the impairment in recycling. Moreover, if this assumption is valid, the loss of recycling regulators may cause feeding RNAi defects. To rigorously test this possibility, we assayed the effects of feeding RNAi of genes expressed in the skin (dpy-7) and muscle (unc-22) [26]. Remarkably, nck-1 mutants exhibited a silencing defect similar to that of sid-3 mutants ( Fig  6A,~2-6% affected). Likewise, EHBP-1-or RAB-10-depletion resulted in prominent defects in the silencing of dpy-7 and unc-22 genes; however, the degree of silencing was much higher compared with that in the sid-3 and nck-1 mutants (Fig 6A,~40-50% affected for dpy-7,~30-40% for unc -22). This suggests that, unlike EHBP-1 or RAB-10, SID-3 and NCK-1 are involved in multiple mechanisms underlying the transmission of ingested dsRNA. Indeed, our colocalization study showed that SID-3 was also located in RAB-7-labeled punctate endosomes (Fig  2A), suggesting that SID-3 might be involved in multivesicular body/late endosome-assisted dsRNA export [45][46][47]. In C. elegans, RAB-10 and its closest paralog RAB-8 are the homologs of yeast Sec4p. RAB-8 has been implicated in apical exocytosis [10,48], and RAB-10 and RAB-8 colocalize extensively in endosomes in intestinal cells, suggesting that these endosomes are involved in the sorting of basolateral and apical cargos [10,12]. However, we did not observe a distinct dpy-7 or unc-22 RNAi silencing defect in rab-8 mutant animals (Fig 6A), suggesting that this phenotype is associated with basolateral recycling, and apical exocytosis is not involved.
To discriminate whether apical endocytic dysfunction contributes to the inefficiency of feeding RNAi in recycling regulator mutants, we examined the effect of silencing of the intestine-expressed gene act-5 (percentage of surviving progeny), using the sid-1 mutant as the positive experimental control [26]. SID-1 is a transmembrane protein and functions as a dsRNAgated channel [49,50]. Prominently, SID-1 has been postulated to function in releasing endosomal dsRNA into the intestinal cytoplasm after apical endocytosis [51]. Therefore, loss of SID-1 would result in ineffective RNAi silencing of ACT-5 in the intestinal cells (Fig 6A).
To further assess the involvement of basolateral recycling in the efficiency of feeding RNAi, we prepared a transgenic strain that broadly expressed GFP and measured the extent of GFP silencing in response to gfp feeding RNAi. In a wild-type background, GFP silencing was effective in various tissues, including the intestine, coelomocytes (macrophage-like scavenger cells in the body cavity), and spermatheca (sperm storage organ) (Fig 6B and 6B'). Nevertheless, although silencing still occurred in intestines, loss of SID-3 resulted in a substantial decrease in silencing in coelomocytes. A comparable phenotype was observed in nck-1, ehbp-1, and rab-10 mutants (Fig 6B and 6B'), validating the requirement of basolateral recycling for ingested RNAi signal transmission. The closest paralog of RAB-10, RAB-8, is specifically involved in apical exocytosis in intestinal epithelia [48]. We observed extensive GFP silencing in rab-8 mutants, further demonstrating the necessity and specificity of basolateral recycling during feeding RNAi (Fig 6B and 6B').

SID-2 is trapped in endosomal structures of recycling mutants
SID-2 interacts with negatively charged dsRNAs and directs the apical endocytosis of ingested dsRNAs from the intestinal lumen [51,52]. In addition, analysis of SID-1 mosaic animals suggested that ingested dsRNAs could be transported to the pseudocoelom, likely via endosomal transport [51,53]. These intriguing results led us to speculate whether endocytosed SID-2 could traffic through the recycling compartments en route to the basolateral side. It is worth highlighting that, in addition to being enriched in the apical membrane, SID-2-GFP was also located in scattered intracellular puncta (Fig 6C and 6C'). In sid-3 mutants, there was significant intracellular accumulation of SID-2-GFP in the punctate structures (~12-fold increase in total area) (Fig 6C and 6C'). Moreover, in the absence of RAB-10 or EHBP-1, SID-2-GFP overaccumulated in cytosolic aggregates (~22 to~24-fold increase in total area) (Fig 6C and 6C',  S5A and S5A' Fig), suggesting that the transport of SID-2 requires the basolateral recycling pathway. In agreement with this observation, SID-2 partially overlapped with RAB-10 (Pearson's Coefficient:~0.675) and SID-3 (Pearson's Coefficient:~0.707) in the basolateral endosomal structures (Fig 6D and 6D').
To clarify whether the intracellular accumulation of SID-2 is due to defective apical exocytosis, we assayed the transport of the Golgi-derived apical secretory cargo PGP-1-GFP (ATPbinding cassette transporter) [48] in sid-3(ok973) cells. The distribution of PGP-1-GFP was normal in sid-3(ok973) cells (S5B and S5B' Fig), suggesting that the overaccumulation of SID-2-GFP is not due to apical secretion defects.

Discussion
A previous study showed that SID-3 is required for dsRNA endocytosis [26]. Here we identified SID-3 as a novel regulator mediating endocytic recycling in the C. elegans intestine. SID-3 is localized in basolateral endosomes and participates in an interactive cascade with the RAB-10 effector EHBP-1, NCK-1, and DYN-1. EHBP-1 governs the endosomal localization of SID-3, and SID-3 functions in concert with NCK-1 and DYN-1 to promote the fission of membrane tubules in recycling endosomes (Fig 7).
A series of studies in C. elegans identified multiple SID proteins involved in systemic RNAi, including SID-1, SID-2, SID-3, and SID-5 [54]. SID-1 and its mammalian homolog SIDT2 have been implicated in the release of endosomal dsRNA into the cytoplasm [51,55,56]. Genetic analysis demonstrated that loss of SID-3 leads to impaired RNA mobility, suggesting that SID-3 acts to direct extracellular dsRNA uptake [26]. Accordingly, in mammalian cells ACK/SID-3 is associated with clathrin-coated pits in the plasma membrane [18,19]. It is worth emphasizing that SID-3 also resides on endosome-like structures in C. elegans pharynx and intestine [26]. Transgenic overexpression of SID-3 in sid-3 mutants co-expressing pharyngeal gfp-dsRNA resulted in increased silencing of GFP in body-wall muscles and a significant decrease of GFP silencing in the pharynx [26], implying that SID-3 plays a role in dsRNA export. The results of our investigation support this putative function of SID-3. The recycling regulator EHBP-1 recruits SID-3 to basolateral recycling endosomes. SID-3 then cooperates with NCK-1 and DYN-1 to promote membrane tubule fission and subsequent tubular carrier formation, which may facilitate the transport of the ingested RNAi signals to the pseudocoelom. A similar process involving EHBP1L1 and dynamin occurs during apical-directed transport in epithelia [57]. Specifically, the EHBP1L1-Bin1-dynamin complex localizes to the recycling endosomes and promotes the generation of vesicular cargo carriers [57].
In response to gfp feeding RNAi, a moderate decrease in nucleus-located SUR-5-GFP silencing was observed in the sid-3 mutant intestine [26], which was inconsistent with the enhancement of GFP silencing in the pharynx [26]. In this respect, it is noteworthy that gfp-dsRNA in the pharynx was derived from pharyngeal transgene expression. In the case of gfp feeding RNAi, dsRNA in the intestinal cells could arise from multiple sources, including ingested dsRNA that is apically endocytosed and subsequently released into the cytosol by endosomal SID-1, and dsRNA that is transported to the pseudocoelom and then recaptured via endocytosis [51,58]. Although there is still a lack of evidence, dsRNA derived from the pseudocoelom has been postulated to contribute to intestinal silencing [58]. Loss of SID-3 leads to trapping of SID-2 in the basolateral endosomes, which may cause a decrease in dsRNA in the pseudocoelom and subsequent defects in silencing in the intestine. An alternative explanation is that disruption of recycling could indirectly impede the preceding apical endocytosis and therefore attenuate RNAi efficiency in intestinal cells. Of note, in contrast to the SUR-5-GFP silencing deficiency observed in sid-3 mutants, our gfp feeding RNAi experiments showed that the efficiency of GFP silencing in sid-3 mutant intestines was comparable to that in wild-type animals. This phenotypic discrepancy suggests that the sur-5 gene coding sequence may affect the efficiency of RNA silencing by an as yet to be determined mechanism. Further experiments need to be performed to rigorously test these speculations.

C. elegans strains and maintenance
All C. elegans strains were derived from Bristol strain N2 and grown at 20˚C on nematode growth media (NGM) plates seeded with E. coli strain OP50 following standard protocols. A complete list of strains and transgenes (with DNA injection concentrations) used in this study can be found in the S12 Table. Feeding RNAi assays RNAi-mediated silencing was performed following the feeding protocol described by Timmons and Fire [59]. RNAi constructs used in this study were from the Ahringer library [35]. For most experiments, synchronized L1 stage animals were cultured and F1 adults were scored, unless stated otherwise. For the dpy-7 phenotype, only those animals that were strongly dumpy were counted as dumpy. For the unc-22 phenotype, only those animals that were incapable of moving upon rapping the plate were scored as paralyzed.

Genome-wide RNAi screen
Previous studies showed that rab-10 mutant animals are viable and superficially normal in growth and development [5,12], and that rab-10 mutants accumulate the recycling cargo hTAC-GFP in enlarged structures in the deep cytosol of intestinal cells [10]. To identify additional recycling regulators, a large-scale RNAi genetic screen was performed using rab-10 (RNAi) as the positive control to identify candidate genes whose expression knockdown can lead to the hTAC-GFP overaccumulation phenotype. After the initial screen and follow-up validation, RNAi-mediated knockdown of 54 candidates was found to lead to intracellular hTAC-GFP aggregation, including SID-3, ARX-7, GGTB-1, EXOC-7, HUM-2, and SEC-10. Previous studies indicated that the mammalian homolog of HUM-2/myosin functions with Rab10 to mediate the delivery of GLUT4 storage vesicles to the plasma membrane [60]. SEC-10 and EXOC-7 were implicated in the regulation of hTAC recycling transport in the C. elegans intestine [8]. arx-7 encodes a subunit of the Arp2/3 complex, which is also known to be crucial for endocytic trafficking [61]. GGTB-1 is a homolog of human RABGGTB (Rab geranylgeranyltransferase β subunit), which presumably participates in the attachment of a geranylgeranyl group to the cysteine at the C-terminus of Rab [62].

Membrane fractionation assay
Worms expressing intestinal SID-3-GFP were synchronized and cultured on 1 mM IPTG NGM plates seeded with HT115 strains containing a control vector L4440 or ehbp-1 RNAi vector. Animals of mixed stages were washed off with M9 buffer, and the worm pellet was resuspended in 500 μl lysis buffer (25 mM Tris-HCl PH 7.5, 100 mM NaCl, 1 mM EDTA, 0.5% NP-40, 1 mM PMSF, 1 mM Na3VO4, 10 mM NaF and protease inhibitors). The worms were then disrupted utilizing an automatic grinding machine (Jingxin Inc., Shanghai, China). The lysates were cleared by centrifugation at 1000xg for 20 min at 4˚C. The post-cleared lysate (50 μl) was centrifuged at 100,000xg for 1h. Supernatants were collected, and pellets were reconstituted in the same volume of lysis buffer as that of the supernatant. Equivalent amounts of supernatants and pellets were subjected to immunoblotting using anti-actin, anti-tubulin, and anti-GFP antibodies.

Confocal microscopy
Live C. elegans animals were mounted on 2% agarose pads with 100 mM levamisole. Multiwavelength fluorescence images (GFP, mCherry, and DAPI channels) and mono-fluorescence images (GFP channel) were acquired at 20˚C using a Nikon C2 laser scanning confocal microscope (Nikon, Tokyo, Japan) equipped with a 100×N.A. 1.2 oil-immersion objective and NIS-Elements AR 4.40.00 software. Z-series of optical sections were obtained using a 0.8-μm step size. For tracking the dynamics of endosomal tubules, animals were loaded into a spinning disc confocal microscope (Olympus IX83, Japan) equipped with a 100x/1.4 oil (WD 013 mm, DIC slider) objective, multiple lasers (405nm, 488nm, 561nm) with the corresponding filters and EMCCD and scMOS cameras. In the C. elegans intestine, a polarized epithelial tube, the apical membrane faces the lumen, and the basolateral membrane faces the pseudocoelom. The imaging plane close to the basal membrane was defined as "Top", and the imaging plane where the apical membrane and lumen can be observed was defined as "Middle" (deep cytosol of the intestinal cells) (Fig 1B). Synchronized young adult animals (24 hours after the L4 stage) were used for mono-fluorescence imaging (GFP channel). Synchronized L3 stage animals were selected for multi-wavelength fluorescence imaging (GFP, mCherry, and DAPI channels) and spinning disc dynamics imaging (GFP and mCherry channels). Images taken in the DAPI channel (blue color) were used to identify broad-spectrum intestinal autofluorescence caused by lipofuscin-positive lysosome-like organelles [10].

Imaging analysis
Fluorescence data from the GFP channel were analyzed by Metamorph software version 7.8.0.0 (Universal Imaging, West Chester, PA). The "Integrated Morphometry Analysis" module of Metamorph was used to measure the fluorescent intensity (mean intensity) and fluorescence area (total area) within unit regions (automatic local background subtraction). From a total of 10 animals of each genotype, "mean intensity" and "total area" were sampled in wholecell regions of a total of 20 intestinal cells picked at random. In this study, "total area" was used as a composite index of endosomal size and quantity. Typical endosomes cover less area than endosomal structures that overaccumulate in the cytosol. Colocalization images were collected using the open-source Fiji (Image J) software [64]. An object-based plugin, JACoP (Just Another Co-localization Plugin), was used to evaluate Pearson's correlation coefficients for GFP and mCherry signals; the entire imaging area of 12 animals for each genotype was analyzed.