Vps9d1 regulates tubular endosome formation through specific activation of Rab22A

Endocytosis is a crucial cellular event that is required for the uptake of extracellular materials and plasma membrane proteins and lipids. The endocytosed molecules are sorted into endosomes and then ultimately degraded in lysosomes or recycled back to the plasma membrane (Maxfield and McGraw, 2004). Several types of endocytosis have been reported, and clathrin-mediated endocytosis (CME) is the best-characterized type. Other types of endocytosis do not require clathrin to form vesicles and are collectively referred to as clathrin-independent endocytosis (CIE) (Mayor and Pagano, 2007). One of the characteristic features of the CIE pathway in HeLa cells is the presence of tubule-shaped recycling endosomes (known as ‘tubular endosomes’), which are derived from CIE vesicles and function as a platform for the recycling of CIE cargos, including major histocompatibility complex class I (MHC-I), CD55, and CD147 (also known as BSG) (Grant and Donaldson, 2009; Maldonado-Báez et al., 2013b). Maintenance of tubular structures has been shown to be required for proper recycling of CIE cargos, and several important membrane trafficking regulators, including Arf6, EHD1, MICAL-L1 and several Rab GTPases, have previously been reported to be involved in the regulation of tubular endosomes (Radhakrishna and Donaldson, 1997; Caplan et al., 2002; Jovanovic et al., 2006; Sharma et al., 2009; Rahajeng et al., 2012; Finicle et al., 2018; Farmer et al., 2021; Wunderley et al., 2021; Higashi et al., 2022). One of them, Rab22A, has been shown to be essential for the initial step in tubular endosome formation (Maldonado-Báez et al., 2013a; Etoh and Fukuda, 2019).

Rab22A is a member of the Rab family of small GTPases that regulate intracellular membrane traffic in all eukaryotes (Zhen and Stenmark, 2015; Pfeffer, 2017; Homma et al., 2021), and two Rab22 isoforms, Rab22A and Rab22B (also known as Rab31), are present in vertebrates (Diekmann et al., 2011). In order for a Rab protein to function in membrane traffic, it must be activated by a guanine-nucleotide-exchange factor (GEF), which replaces the GDP bound to the inactive Rab with GTP and enables the active Rab to localize to a specific membrane compartment or vesicle (Barr and Lambright, 2010). The active Rab then recruits specific effector molecules that mediate various steps in membrane traffic, including budding, transport, tethering, docking and fusion. Thus, in order to fully understand Rab22A-mediated tubular endosome formation, it is necessary to identify the upstream and downstream regulators of Rab22A, that is, the Rab22A-GEFs and Rab22A effectors, respectively. Among the Rab22A effectors or Rab22A-binding proteins previously reported, KIF13A and its interaction with Rab22A have been shown to be required for tubular endosome formation (Shakya et al., 2018; Etoh and Fukuda, 2019; Patel et al., 2021; Thankachan and Setty, 2022). By contrast, nothing is known about the Rab22A-GEF that specifically regulates tubular endosome formation, although several vacuolar protein sorting 9 (VPS9) domain-containing proteins (hereafter referred to as VPS9 proteins), including Rinl, have been shown to exhibit Rab22A-GEF activity in addition to Rab5-GEF activity in vitro (Delprato et al., 2004; Woller et al., 2011). However, since tubular endosomes are involved in the recycling of CIE cargos (Rab22A-dependent) and not of CME cargos (Rab5-dependent), we hypothesized that an as-yet-unidentified or uncharacterized Rab22A-specific GEF should be present.

In this study, we comprehensively screened for Rab22A-GEFs by using a GDP-locked Rab22A mutant as bait, and we identified VPS9 domain containing 1 (Vps9d1), the only uncharacterized GEF domain-containing protein (Ishida et al., 2016), as the most likely candidate. The results showed that overexpression of Vps9d1 induced enlargement of endosomes positive for Rab22A or Rab22B (hereafter referred to collectively as Rab22A/B), but not of endosomes positive for Rab5 isoforms (Rab5A, Rab5B or Rab5C, hereafter referred to collectively as Rab5A/B/C), and that it possessed GEF activity toward Rab22A, but not toward Rab5A, in vitro. Moreover, knockout of Vps9d1 in HeLa cells greatly impaired both tubular endosome formation and CIE cargo recycling. Our findings indicate that Vps9d1 is the first functional Rab22A-GEF that regulates tubular endosome formation in mammalian cultured cells.

To identify a Rab22A-GEF that functions in tubular endosome formation, we turned our attention to reports showing that RabGEFs often physically interact with their substrate GDP-Rabs (Delprato et al., 2004; Delprato and Lambright, 2007; Tamura et al., 2009; Mori et al., 2013; Homma and Fukuda, 2016). Accordingly, we performed yeast two-hybrid assays to test for possible interactions between ∼40 putative GEF domain-containing proteins found in humans and mice and a constitutively negative (CN) form of Rab22A (Rab22A-S19N, referred to as Rab22A-CN hereafter). The results of the yeast two-hybrid assays showed that only one of the putative RabGEFs tested, Vps9d1, a previously uncharacterized VPS9 protein (Fig. 1A), strongly interacted with Rab22A-CN and did not interact with a GTP-locked, constitutively active (CA) form of Rab22A (Rab22A-Q64L, hereafter named Rab22A-CA) (Fig. 1B; Fig. S1). The interaction between Vps9d1 and Rab22A-CN was also observed in co-immunoprecipitation assays using cultured mammalian cells (Fig. 1C). To determine whether Vps9d1 specifically interacts with Rab22A, we performed additional yeast two-hybrid assays to test for possible interactions between Vps9d1 and each of all the mammalian Rabs, testing CN (Ser/Thr-to-Asn mutation) and CA (Gln-to-Leu mutation) mutants of Rab1–43. Since Vps9d1 contains a coiled-coil (CC) domain in its N-terminal region and a VPS9 domain in its C-terminal region, we divided Vps9d1 into two parts (named Vps9d1-N and Vps9d1-C) (Fig. 1A) and used them as prey in the yeast two-hybrid assays. Although CC domains have often been found to serve as Rab-binding domains (Fukuda et al., 2008; Lee et al., 2009), and several GEFs are known to interact with Rabs other than their substrate Rabs (Tamura et al., 2009; Kinchen and Ravichandran, 2010; Kloer et al., 2010; Knödler et al., 2010; Pusapati et al., 2012), none of the Rabs interacted with Vps9d1-N under our experimental conditions, whereas several Rabs, including Rab22A, interacted with Vps9d1-C (Fig. 1D). Based on the growth rate of the yeasts, Vps9d1-C strongly interacted with the CN mutants of Rab5A/B/C, Rab17, Rab22A and Rab24; relatively weakly interacted with the CN mutant of Rab22B; but only marginally interacted with the CN and CA mutants of Rab40B, Rab40C and Rab40AL (Fig. 1D). Because GEFs generally recognize GDP-bound Rabs, we selected seven Rabs whose CN mutants interacted with Vps9d1 as candidate Vps9d1 substrates, and we noted that all of them belonged to the Rab5 subfamily according to a phylogenetic analysis of mouse Rabs (Homma et al., 2021). Actually, all previously characterized VPS9 proteins have been shown to possess GEF activity toward Rab5 subfamily members; for example, Rabex-5 (also known as RABGEF1) and Varp (also known as ANKRD27) function as a Rab5-GEF and a Rab21-GEF, respectively (reviewed in Ishida et al., 2016 and references therein). Comparison with the Rab-CN binding specificity of Rabex-5 and Varp previously reported (Tamura et al., 2009; Mori et al., 2013) revealed that only Vps9d1 interacted with both Rab22A and Rab22B (Fig. 1B,D).

Since the results of the above-described yeast two-hybrid assays implied that Vps9d1 functions as a GEF for one or more of the seven candidate Rabs, we next evaluated the effect of Vps9d1 overexpression on endosome morphology in COS-7 cells, which do not contain tubular endosomes by nature, as a second screening step. Since Rabex-5 overexpression has been reported to induce enlargement of Rab5A-positive endosomes, presumably by facilitating homotypic fusion of early endosomes (Zhu et al., 2007), we hypothesized that overexpression of Vps9d1 would induce enlargement of endosomes where its substrate Rabs were present. When we overexpressed monomeric Strawberry (mStr)-tagged Vps9d1 in COS-7 cells stably expressing each of the EGFP-tagged candidate Rabs, only Rab22A/B-positive dots, not Rab5A/B/C-positive dots, were found to be significantly enlarged by mStr–Vps9d1 overexpression (Fig. 1E,F and data not shown). These results, together with the fact that Rab22A-CA induces the formation of enlarged endosomes (Mesa et al., 2001; Kauppi et al., 2002), strongly suggested Rab22A/B as the most likely candidates for Vps9d1 targets.

Several amino acid residues in the VPS9 domain of Rabex-5 that are highly conserved among the VPS9 proteins are known to be essential for its GEF activity (Delprato et al., 2004). Since Tyr354 of Rabex-5 is one such residues, and because the corresponding tyrosines of several VPS9 proteins have also been shown to be essential for GEF activity (Tamura et al., 2011; Kajiho et al., 2011, 2012), we expected that Tyr628 of Vps9d1 would be required for its GEF activity (Fig. 2A). As expected, yeast two-hybrid assays showed that a Vps9d1-Y628A mutant (hereafter referred to as Vps9d1-YA) was completely devoid of binding activity toward the CN mutants of the seven candidate Rabs (Fig. 2B, bottom panel), and overexpression of mStr-tagged Vps9d1-YA in COS-7 cells, unlike overexpression of the wild-type (WT) Vps9d1 (Vps9d1-WT), failed to induce enlargement of Rab22A/B-positive dots (Fig. 2C,D). The absence of Rab22A/B-dependent activity of Vps9d1-YA was not attributable to its lower protein expression level, because the expression levels of mStr–Vps9d1-WT and -YA were almost the same (Fig. 2E). Taken together, these results suggest that the enlargement of Rab22A/B-positive dots by Vps9d1 is attributable to activation of Rab22A/B via the VPS9 domain.

We next measured the Rab22A-GEF activity of purified recombinant Vps9d1 protein in vitro (Fig. 3A) by using a guanine nucleotide analogue, 2′/3′-O-(N-methylanthraniloyl)guanosine 5′-diphosphate (Mant-GDP). Consistent with the results of the cellular assays described above, Vps9d1 significantly facilitated GDP release from Rab22A in a time-dependent manner (Fig. 3B), but it did not enhance GDP release from Rab5A (Fig. 3C). In sharp contrast, a C-terminal domain of Rabex-5 (named Rabex-5-C) did not show any Rab22A-GEF activity under our experimental conditions (Fig. 3D), although Rabex-5-C strongly promoted the release of Rab5A-bound GDP, as described previously (Fig. 3E; Delprato and Lambright, 2007; Zhang et al., 2014; Shin et al., 2017; Lauer et al., 2019). Consistent with our observation, Rabex-5-C has previously been shown to bind to GDP-Rab5, but not to GDP-Rab22, in pulldown assays (Zhu et al., 2009). Since the Rab5-GEF activity of Rabex-5 has been reported to be regulated by an autoinhibition mechanism acting through the C-terminal CC domain, which restricts binding to substrate Rab5 (Delprato and Lambright, 2007; Zhang et al., 2014; Lauer et al., 2019), Rabex-5-C has been shown to exhibit stronger Rab5-GEF activity than the full-length protein. To investigate whether a similar regulatory mechanism exists in regard to Vps9d1, we tested the Rab22A-GEF activity of Vps9d1-C. Unexpectedly, however, Vps9d1-C showed neither Rab22A-GEF activity nor Rab5A-GEF activity (Fig. 3F,G), indicating that the N-terminal domain of Vps9d1 is also required for its Rab22A-GEF activity. The lack of Rab22A-GEF activity of Vps9d1-C was confirmed by the finding that overexpression of mStr-tagged Vps9d1-C in COS-7 cells failed to induce enlargement of Rab22A/B-positive endosomes (Fig. S2). Taken together, these results strongly indicate that Vps9d1 functions as a specific GEF for Rab22, but not as a Rab5-GEF, and that the full-length protein is required for Rab22A-GEF activity of Vps9d1.

In a final set of experiments, we investigated the possible involvement of Vps9d1 in Rab22-dependent membrane trafficking events. Since Rab22A/B have been shown to be required for tubular endosome formation in HeLa cells (Weigert et al., 2004; Shakya et al., 2018; Etoh and Fukuda, 2019), we hypothesized that Vps9d1 regulates tubular endosome formation through activation of Rab22. To test our hypothesis, we established Vps9d1-knockout (KO) HeLa cells and a rescued cell line (Fig. 4A, lane 2), and examined their tubular endosome structures by visualizing the known tubular endosome markers endogenous MICAL-L1 and EGFP–Rab10 (Sharma et al., 2009; Etoh and Fukuda, 2019). As expected, there were fewer and shorter MICAL-L1-positive and EGFP–Rab10-positive tubules in the Vps9d1-KO cells (Fig. 4B, left middle panels), and the phenotype was completely rescued by re-expression of Vps9d1-WT, but not by re-expression of Vps9d1-YA (Fig. 4B, right middle and far right panels, and Fig. 4C–E). However, the level of expression of Vps9d1-YA protein was much lower than the level of Vps9d1-WT protein expression when they were stably expressed in Vps9d1-KO HeLa cells (Fig. 4A, lanes 3 and 4). The lack of a rescue effect of Vps9d1-YA was unlikely to be attributable to its lower expression level because the expression level of Vps9d1-YA was higher than that of the endogenous protein (Fig. 4A, lane 1). Moreover, tubular endosome formation in Vps9d1-KO cells was not restored by transiently overexpressing Vps9d1-YA, even though its expression level was comparable to that of Vps9d1-WT (Fig. S3A,B).

More importantly, overexpression of Rab22A-CA (i.e. forced activation of Rab22A) in Vps9d1-KO HeLa cells clearly restored tubular endosome formation (Fig. 4F,G). Overexpression of wild-type Rab22A (Rab22A-WT) also appeared to slightly increase the number of MICAL-L1-positive tubules, but the difference was not statistically significant (Fig. 4G). Moreover, consistent with the results of the in vitro GEF assays, KO or knockdown (KD) of Vps9d1 had no effect on Rab5 localization at all, although KD of Rabex-5 caused the punctate Rab5 signals to disappear (Fig. S3C), further confirming the Rab22-specific GEF activity of Vps9d1 in living cells. We therefore expected there to be less active Rab22A in Vps9d1-KO cells, and we observed reduced Rab22A–KIF13A interaction, which has been shown to be required for tubular endosome formation (Shakya et al., 2018), in Vps9d1-KO cells (Fig. 4H, bottom panel). Nevertheless, we did not observe any clear differences between the membrane association or intracellular distribution of Rab22A in WT and Vps9d1-KO cells (Fig. 4I and data not shown). This discrepancy may be explained by the presence of an additional Rab22A-GEF that promotes membrane association of Rab22A even in the absence of Vps9d1, or it could be the result of inactive Rab22A not being efficiently dissociated from the membranes.

We also investigated the effect of Vps9d1-KO on the trafficking of Rab22A-dependent CIE cargo proteins CD55 and CD147 in tubular endosomes by performing antibody uptake assays in HeLa cells (Eyster et al., 2009; Maldonado-Báez et al., 2013a; Higashi et al., 2022). In the control WT cells, antibodies against CD55 and CD147 were incorporated into the cells and could be observed as numerous dots, as described previously (Eyster et al., 2009; Higashi et al., 2022), and their signals were well colocalized with EGFP–Rab10-positive tubules (Fig. 5A, top rows), whereas in Vps9d1-KO cells, the CD55 and CD147 dots were rather dispersed, and no colocalization between them and EGFP–Rab10 was observed (Fig. 5A, middle rows). Since some of the CD147 dots in the Vps9d1-KO cells were colocalized or closely associated with EEA1 (Fig. S4A), CD147 proteins that were not transported to tubular endosomes are likely to have been trapped in early endosomes. These phenotypes were completely rescued by re-expression of Vps9d1 in Vps9d1-KO cells (Fig. 5A, bottom rows). Lastly, we investigated the recycling of internalized CD147, and the results showed that both internalized CD147-positive dot numbers and total CD147 fluorescence signals were significantly reduced in a time-dependent manner in WT and rescued cells, whereas both were unaltered over time in Vps9d1-KO cells (Fig. 5B–D), strongly suggesting that CIE cargo recycling is impaired in Vps9d1-KO cells. By contrast, recycling of transferrin (Tf), a known CME cargo, was unaffected even in the Vps9d1-KO cells (Fig. S4B,C). Thus, Vps9d1 is likely to mediate tubular endosome formation and CIE cargo protein recycling, especially cargo sorting into tubular endosomes, through activation of Rab22A.

In this study, we have for the first time characterized the substrate Rabs of Vps9d1, the only uncharacterized VPS9 family protein among the ten VPS9 proteins present in humans and mice (Ishida et al., 2016), and we have found that Vps9d1 functions as a major Rab22A-GEF that is required for tubular endosome formation in HeLa cells. Although the number of long tubular endosome structures was dramatically decreased in Vps9d1-KO cells (Fig. 4B–E), relatively short tubules were still observed, suggesting that additional Rab22-GEFs function in the absence of Vps9d1. RIN family members, which possess both Rab22-GEF and Rab5-GEF activities (Kajiho et al., 2011; Woller et al., 2011), might be promising candidates for the additional Rab22-GEF that supports the formation of short tubular endosomes and the endosomal localization of Rab22A in Vps9d1-KO cells.

Although Vps9d1 significantly facilitated GDP release from Rab22A in vitro, its Rab22A-GEF activity appeared to be weaker than the Rab5A-GEF activity of Rabex-5-C (compare Fig. 3B and E). Whereas the Rab5-GEF activity of Rabex-5 is regulated by an autoinhibition mechanism through its C-terminal CC domain, which restricts binding to substrate Rab5 (Delprato and Lambright, 2007; Zhang et al., 2014; Lauer et al., 2019), because Vps9d1-C completely lacks Rab22A-GEF activity in vitro (Fig. 3F), it is likely that no such inhibitory regulation mechanism exists in regard to Vps9d1. How the N-terminal domain of Vps9d1 supports Rab22A-GEF activity is unknown, but since Vps9d1 contains a CC domain in its N-terminal region (Fig. 1A), Vps9d1 might facilitate GDP release from Rab22A more efficiently by interacting with certain molecules via its CC domain. To investigate this possibility, further research attempting to identify novel Vps9d1 binding partners will be necessary in the future.

How does Vps9d1 specifically activate Rab22A on early endosomes? RabGEFs have been proposed to be the main determinants of specific Rab membrane targeting (Blümer et al., 2013), and thus Vps9d1 would be expected to localize on early endosomes; however, under normal fixation conditions, Vps9d1 appeared to localize in the cytosol in both COS-7 and HeLa cells (Fig. 1E; Fig. S3A). After removing the cytosolic compartment, however, a portion of the Vps9d1 was clearly found to be localized on dotted structures, often closely associated with (or located adjacent to) EEA1-positive dots (Fig. S3D, arrowheads), suggesting that this membrane-associated Vps9d1 locally activates Rab22A. The Vps9d1 dots were actually often colocalized or closely associated with Rab22A/B dots, but not with Rab5A dots (Fig. S5, arrowheads). The mechanism by which Vps9d1 associates with endosomal membranes is the next important issue to be clarified in future studies. Identifying a protein that binds to the CC domain of Vps9d1 may provide the first clue to clarifying this issue.

In addition to its role in tubular endosome formation, Rab22A has been suggested to regulate the sorting of CIE cargo proteins (Maldonado-Báez et al., 2013a; Higashi et al., 2022) and their transport via dimerization of KIF13A (Shakya et al., 2018; Patel et al., 2021). Consistent with its role in CIE cargo sorting and transport, two representative CIE cargos, CD55 and CD147, were internalized even in the absence of Vps9d1, but they were not sorted to EGFP–Rab10-positive structures (Fig. 5A). Thus, these CIE cargos are unlikely to be efficiently recycled back to the plasma membrane in tubular endosome-deficient Vps9d1-KO cells. Since both Rab22A and Vps9d1 are widely expressed in human and mouse tissues (NCBI data base; gene IDs: 19334, 57403, 72325 and 9605), and Rab22A regulates the CIE recycling pathway in certain immune cells and cancer cells (Cebrian et al., 2016; Johnson et al., 2017; Mayorga and Cebrian, 2019; Qi et al., 2019), the Vps9d1–Rab22A axis is likely to play an important role in the CIE cargo recycling pathway in other cells in addition to HeLa cells.

In conclusion, we have identified Vps9d1 as the first functional Rab22A-GEF that regulates both tubular endosome formation and CIE cargo recycling. The tubular endosome-defective Vps9d1-KO cells established in this study will serve as an ideal tool for analyzing the function and significance of tubular endosomes in the future.

Details of the materials, including antibodies, primers and plasmids, used in this study are summarized in Table S1. All other general reagents used in this study were analytical grade or the highest grade commercially available.

cDNAs encoding mouse and human Vps9d1 were amplified from Marathon-Ready brain and testis cDNAs (Clontech-Takara Bio, Shiga, Japan) by PCR using specific primers and subcloned into appropriate vectors listed in Table S1. cDNAs encoding mouse Vps9d1-Y628A (Vps9d1-YA) (Fig. 2A), Vps9d1-N and Vps9d1-C (Fig. 1A) were similarly obtained by PCR and subcloned into appropriate vectors. The Vps9d1 expression plasmids used in this study have been deposited in the RIKEN BioResource Research Center (https://dnaconda.riken.jp/search/depositor/dep005893.html; RDB19825–RDB19826).

The yeast strain, medium, culture conditions and transformation protocol used were as described previously (James et al., 1996). Yeast two-hybrid assays were performed by using pGBD-C1-RabsΔCys-CA (or -CN) as bait and pAct2-Vps9d1-N, -C or pGAD-C1/pAct2-Varp, -Rabex-5, -Gapex-5 (GAPVD1), -Rin1, -Rin2, -Rin3, -Rinl, -Als2, -Als2cl, and -other putative GEFs as prey (Table S1). Yeast cells were grown at 30°C for 1–2 d on the growth medium (SC-LW; synthetic complete medium lacking leucine and tryptophan) and at 30°C for 1 week on the selection medium (SC-AHLW; SC medium lacking adenine, histidine, leucine and tryptophan).

COS-7 cells [obtained from RIKEN BioResource Center (BRC), cat# RCB0539], HeLaM cells (referred to as HeLa cells throughout this study; RIKEN BRC, cat# RCB5388), and PlatE cells (a kind gift from Dr Toshio Kitamura, The University of Tokyo, Japan; Morita et al., 2000) were grown at 37°C in Dulbecco's modified Eagle's medium (D-MEM; 044-29765; FUJIFILM Wako Pure Chemical, Osaka, Japan), supplemented with 10% fetal bovine serum (MP Biomedicals, Irvine, CA, USA), 100 U/ml penicillin G and 100 μg/ml streptomycin (Meiji Seika Pharma, Tokyo, Japan), in a 5% CO₂ incubator. Plasmids and siRNAs were transfected into these cultured cells by using Lipofectamine 2000 and RNAi MAX (Thermo Fisher Scientific, Waltham, MA, USA), respectively, according to the manufacturer's instructions.

pMRX vectors and pLP/VSVG were cotransfected into PlatE cells, and medium containing retroviruses was collected after filtration through a 0.45 μm syringe filter (Merck Millipore, Burlington, MA, USA). COS-7 and HeLa cells were infected by using the medium containing retroviruses and 8 μg/ml polybrene (Sigma-Aldrich, St. Louis, MO, USA). Uninfected cells were removed by using a 2 μg/ml concentration of puromycin (Merck Millipore) or a 5 μg/ml concentration of blasticidin S (FUJIFILM Wako Pure Chemical).

A Vps9d1-KO HeLa cell line was established by using the CRISPR/Cas9 system. In brief, a single-guide RNA (sgRNA) sequence was designed by using SYNTHEGO (https://design.synthego.com/#/). Annealed oligonucleotides that encode the sgRNA for human Vps9d1 (Table S1) were subcloned into the pSpCas9(BB)-2A-puro vector (#48139; Addgene). The resulting vector was transfected into HeLa cells, and 24 h later, 2 μg/ml puromycin was added to the medium to remove untransfected cells. Vps9d1-KO cells were subcloned by limiting dilution, and their gene disruption was confirmed by both immunoblotting and genomic sequencing. Two nucleotide deletions (5′-ccaGTGG--TCAGCTCTCTGGAA-3′) and one nucleotide insertion (5′-ccaGT[G]GGCGTCAGCTCTCTGGAA-3′) were found in the sgRNA target sequence in the Vps9d1-KO cells (PAM sequence, lowercase).

Cells were lysed with a lysis buffer [50 mM HEPES-KOH, pH 7.2, 150 mM NaCl, 1% Triton X-100 and a protease inhibitor cocktail (Roche, Penzberg, Germany)] containing 1 mM EDTA. Samples were boiled in an SDS sample buffer (62.5 mM Tris-HCl, pH 6.8, 2% 2-mercaptoethanol, 10% glycerol and 0.02% Bromophenol Blue). Proteins in the samples were separated by 7.5% or 10% SDS-PAGE and transferred to a polyvinylidene difluoride (PVDF) membrane (Merck Millipore) by electroblotting. After blocking with 1% skim milk (diluted in PBS containing 0.1% Tween 20) for 30 min, blots were incubated at 4°C overnight with the primary antibodies listed in Table S1, and then at room temperature for 1 h with appropriate horseradish peroxidase (HRP)-conjugated secondary antibodies. For HRP-conjugated primary antibodies, blots were incubated at room temperature for 1 h. Immunoreactive bands were detected by using enhanced chemiluminescence and the ChemiDoc Touch imaging system (Bio-Rad; Hercules, CA, USA). Original blot data are shown in Fig. S6.

FLAG–Rab22A-CA, FLAG–Rab22A-CN and T7–Vps9d1 were separately overexpressed in COS-7 cells, and the cells were lysed with lysis buffer (see Immunoblotting section above) containing 1 mM MgCl₂. FLAG–Rab22A was isolated with anti-FLAG M2 agarose beads (Sigma-Aldrich). The beads coupled with either FLAG–Rab22A-CA or FLAG–Rab22A-CN were incubated at 4°C for 1 h with COS-7 lysates containing T7–Vps9d1 in the presence of 0.5 mM GTPγS or 1 mM GDP, respectively. After washing the beads with a washing buffer (50 mM HEPES-KOH, pH 7.2, 150 mM NaCl, 1 mM MgCl₂ and 0.1% Triton X-100) three times, proteins bound to the beads were boiled with the SDS sample buffer and then immunoblotted as described above.

Cells expressing EGFP- and/or mStr-tagged proteins were fixed with 4% paraformaldehyde, permeabilized with 50 μg/ml digitonin in PBS for 5 min, and incubated with 3% bovine serum albumin (BSA) diluted in PBS for 30 min. Next, the cells were incubated with primary antibodies for 1 h and then with appropriate Alexa Fluor-conjugated secondary antibody for 1 h (Table S1). Finally, the cells were mounted on a slide glass with Prolong Diamond containing DAPI (Thermo Fisher Scientific). To remove cytosolic components, cells were treated with 50 μg/ml digitonin for 45 s and washed with PBS before fixation. All samples were examined with a confocal fluorescence microscope (Fluoview 1000; Evident/Olympus, Tokyo, Japan) equipped with a Plan-Apochromat 100×/1.45 NA oil-immersion objective lens or a Plan-Apochromat 60×/1.35 NA oil-immersion objective lens (Evident/Olympus). The images acquired were processed with ImageJ software (NIH, Bethesda, MD; version 1.53s; https://imagej.nih.gov/ij/index.html).

Glutathione S-transferase (GST)-Gly linker (gl)-tagged Rab5A and Rab22A were expressed separately in Escherichia coli strain JM109 and purified using the standard protocol. To remove the GST tag, GST–gl–Rab5A or GST–gl–Rab22A (120 μg) was treated with 0.06 U/μl of thrombin (GE Healthcare, Chicago, IL, USA). The sample was then incubated at 4°C for 1 h with glutathione MagBeads (GenScript, Piscataway, NJ, USA) and benzamidine-Sepharose 6B (GE Healthcare) to remove the GST tag and thrombin, respectively.

3×FLAG-tagged Vps9d1, Vps9d1-C or Rabex-5-C was transiently overexpressed in COS-7 cells, and the cells were lysed with lysis buffer containing 1 mM EDTA. After centrifugation at 17,900 g at 4°C for 10 min, the supernatant was recovered, and the 3×FLAG-tagged proteins in the supernatant were purified with anti-FLAG M2 magnetic beads (Sigma-Aldrich). In brief, the COS-7 cells lysates were incubated at 4°C for 1 h with the beads and then washed three times with a wash buffer (50 mM HEPES-KOH, pH 7.2 and 150 mM NaCl). The beads were incubated at 4°C for 1 h with 500 μg/ml 3×FLAG peptide (Sigma-Aldrich) with agitation. Finally, the supernatant containing the purified 3×FLAG-tagged proteins was collected, and the purified samples were analyzed by 10% SDS-PAGE followed by staining with Coomassie Brilliant Blue Rapid Stain (BIO-CRAFT, Tokyo, Japan).

The in vitro GEF activity of Vps9d1 and Rabex-5-C was measured by monitoring decreases in the fluorescence of 2′/3′-O-(N-methylanthraniloyl)guanosine 5′-diphosphate (Mant-GDP) (Sigma-Aldrich). Mant-GDP was loaded onto 1 μM of Rab5A and Rab22A as described previously (Zhu et al., 2001), and excessive free Mant-GDP was removed by using a 10K ultrafiltration spin column (Intégrale, Naruto, Japan). The Mant-GDP-loaded Rab samples were diluted with an exchange buffer (50 mM HEPES-KOH, pH 7.2, 150 mM NaCl and 0.5 mM MgCl₂), and the reaction was started by adding 1 μM of 3×FLAG–Vps9d1, 1 μM of 3×FLAG–Vps9d1-C or 0.5 μM of 3×FLAG–Rabex-5-C (or BSA as a control in the presence of 3×FLAG-peptide), and 1 μM of GppNHp (Sigma-Aldrich). The changes in fluorescence were measured (excitation at 355 nm and emission at 460 nm) at the times indicated in Fig. 3B–G, by using a Victor Nivo Multimode microplate reader (PerkinElmer, Waltham, MA, USA). The values shown in Fig. 3B–G were normalized by dividing each fluorescence intensity by the fluorescence intensity at 0 min.

HeLa cells co-expressing mStr–Rab22A and EGFP–KIF13A-tail (or EGFP alone) were lysed in lysis buffer containing 10 mM MgCl₂. After centrifugation at 17,900 g at 4°C for 10 min, the supernatants were incubated at 4°C for 1 h with 10 μl of glutathione-Sepharose 4B beads (wet volume; GE Healthcare) coupled with 1 μg of GST-tagged GFP nanobody (Etoh and Fukuda, 2019). The samples were washed with a washing buffer (50 mM HEPES-KOH, pH 7.2, 150 mM NaCl, 10 mM MgCl₂ and 0.1% Triton X-100) three times, boiled with SDS sample buffer, and then immunoblotted with the antibodies indicated in Fig. 4H.

HeLa cells expressing EGFP–Rab27A-C219A/C221A (a cytosolic marker) were homogenized with a 25-gauge needle in a fractionation buffer [0.32 M sucrose, 10 mM HEPES-KOH, pH 7.2, 1 mM MgCl₂ and protease inhibitor cocktail (Roche)]. After centrifugation at 3000 g at 4°C for 10 min, the supernatants were centrifuged at 100,000 g at 4°C for 1 h twice to precipitate the membrane fraction. Equal proportions of supernatant and membrane fractions were boiled in SDS sample buffer and immunoblotted with the antibodies indicated in Fig. 4I.

HeLa cells were incubated at 37°C for 2 h in medium containing 5 μg/ml primary antibody in a 5% CO₂ incubator. After antibody internalization, the cells were washed with PBS three times and then for 30 s with a low-pH buffer (0.5% acetic acid, pH 3.0 and 0.5 M NaCl) to remove the antibody remaining at the plasma membrane. After washing the cells with PBS three times and then with medium three times, the cells were incubated with medium at 37°C for 0, 15 or 30 min in a 5% CO₂ incubator, and the antibody recycled back to the plasma membrane was removed by washing with PBS and the low-pH buffer, as described above. Samples were fixed with 4% paraformaldehyde and incubated for 1 h with unlabeled goat anti-mouse IgG before permeabilization. The internalized primary antibody was visualized and analyzed as described in the Immunofluorescence analysis section above.

HeLa cells were pre-incubated at 37°C for 1 h in serum-free medium in a 5% CO₂ incubator. The cells were then incubated on ice for 30 min in serum-free medium containing 5 µg/ml Alexa Fluor 594-conjugated Tf (Alexa594–Tf; Thermo Fisher Scientific). After Tf binding to the cell surface, the cells were washed sequentially with PBS and medium three times and incubated at 37°C in a 5% CO₂ incubator for the times indicated in Fig. S4. After cell fixation, the internalized Alexa594–Tf was visualized and analyzed as described in the Immunofluorescence analysis section above.

The amino acid sequences of mouse Rab1–43 were obtained from the NCBI database. The multiple sequence alignment was drawn by using the ClustalW program version 1.83 (Thompson et al., 1994) set at the default parameters and by the neighbor-joining method.

The number of cells containing at least one >2 μm dot was manually counted (50 cells were analyzed in each experiment; Figs 1F, 2D). The number of cells containing at least one >20 μm tubule in maximum intensity z-projected images was also manually counted (20–30 cells were analyzed in each experiment; Fig. 4C,G). To determine the number of tubules longer than 20 μm (Fig. 4D) and total tubule length per cell (Fig. 4E), maximum intensity z-projected images were skeletonized by using Lpx Filter2d plugins (filter=lineFilters; linemode=lineExtract) in the LPixel ImageJ plugins package (https://lpixel.net/products/lpixel-imagej-plugins/) (Kuki et al., 2017). The skeletonized tubules were then automatically measured by using ImageJ (30 cells were analyzed in each experiment). The number of CD147-positive dots and total fluorescence intensity of CD147 were automatically measured by using ImageJ (20 cells were analyzed in each experiment; Fig. 5C,D), and the values were normalized to those at 0 min. For the Alexa594–Tf recycling assay, total fluorescence intensity was automatically measured by using ImageJ (20 cells were analyzed in each experiment; Fig. S4C), and the values were normalized to the values obtained at 5 min. The experiments were independently repeated three times for each analysis, and quantitative data are expressed as the mean±s.e.m. One-way ANOVA and Tukey's test (for multiple comparisons) or two-tailed unpaired Student's t-test (for comparison between two samples) were performed with R software or Prism4 (GraphPad software, version 4.0a), and the following criteria were used for statistical significance: *P<0.05, **P<0.01 and ***P<0.001.

We thank Riko Kinoshita and Kentaro Haga for initial Vps9d1 experiments, Kazuyasu Shoji for technical assistance, Dr Toshio Kitamura for kindly donating materials, and all members of the Fukuda laboratory for helpful discussions.

The peer review history is available online at https://journals.biologists.com/jcs/lookup/doi/10.1242/jcs.260522.reviewer-comments.pdf