Retromer is a vital element of the endosomal protein sorting machinery and comprises two subcomplexes that operate together to sort membrane proteins (cargo) and tubulate membranes. Tubules are formed by a dimer of sorting nexins, a key component of which is SNX1. Cargo selection is mediated by the VPS35–VPS29–VPS26 trimer, which additionally recruits the WASH complex through VPS35 binding to the WASH complex subunit FAM21. Loss of function of the WASH complex leads to dysregulation of endosome tubulation, although it is unclear how this occurs. Here, we show that FAM21 also binds to the SNX1-interacting DNAJ protein RME-8. Loss of RME-8 causes altered kinetics of SNX1 membrane association and a pronounced increase in highly branched endosomal tubules. Building on previous observations from other laboratories, we show that these tubules contain membrane proteins that are dependent upon WASH complex activity for their localization to the plasma membrane. Therefore, we propose that the interaction between RME-8 and the WASH complex provides a means to coordinate the activity of the WASH complex with the membrane-tubulating function of the sorting nexins at sites where retromer-mediated endosomal protein sorting occurs.
The endocytic system is a major hub for membrane protein sorting in the mammalian cell. Membrane proteins (cargo) in an early endosome are destined for one of a number of fates – recycling to the plasma membrane (either directly or through the recycling endosomes), retrieval to the trans-Golgi network or degradation in the lysosome (reviewed in Seaman, 2008). Sophisticated sorting mechanisms are required if all cargoes are to reach their correct destination, and one important aspect of this is the partitioning of cargo into distinct tubular transport intermediates before onward trafficking. This requires the concerted action of several macromolecular protein complexes, which function to both select and package cargo and to generate transport intermediates.
The retromer complex, which is composed of two loosely associated subcomplexes [the sorting nexin (SNX) dimer and the cargo-selective complex (CSC)], is a well-established example of protein machinery that combines the roles of membrane deformation and cargo selection. The SNX dimer, currently understood to comprise a combination of SNX1 or SNX2 with SNX5 or SNX6 (Wassmer et al., 2007), binds to phosphoinositol 3-phosphate through its phox homology domains and induces membrane curvature by the action of its C-terminal BAR (Bin-Amphiphysin-Rvs) domains (Kurten et al., 2001; Cheever et al., 2001; Yu and Lemmon, 2001; Cozier et al., 2002; Carlton et al., 2004). The tubules generated by the SNX dimer are stabilised by EHD1 (also known as RME-1), which associates with the CSC (Gokool et al., 2007b; Zhang et al., 2012). Although SNX5 and SNX6 do not drive membrane tubulation (van Weering et al., 2012), both interact with the p150Glued component of dynactin (Wassmer et al., 2009; Hong et al., 2009) and, therefore, link retromer-mediated protein sorting with microtubules through dynein.
The retromer CSC is composed of VPS26, VPS35 and VPS29 and associates transiently with the SNX dimer (Swarbrick et al., 2011) but relies upon Rab7a and SNX3 for interaction with the endosomal membrane (Rojas et al., 2008; Seaman et al., 2009; Harterink et al., 2011; Vardarajan et al., 2012; Harrison et al., 2014). This association is negatively regulated by the Rab GTPase-activating protein TBC1D5 (Seaman et al., 2009). All three subunits of the CSC are required for the assembly and stability of the complex (Collins et al., 2005; Gokool et al., 2007a; Restrepo et al., 2007; Zhao et al., 2007), and cargo recognition is mediated by the VPS35 and VPS26 subunits (Nothwehr et al., 1999; Nothwehr et al., 2000; Arighi et al., 2004; Fjorback et al., 2012). Retromer was first identified as a mediator of endosome-to-Golgi retrieval (Seaman et al., 1997; Seaman et al., 1998), and its cargoes include Vps10p and the cation-independent mannose-6-phosphate receptor (CI-MPR, also known as IGF2R) (reviewed in Seaman, 2012). However, the recent characterisation of the retromer-interacting WASH complex has implicated retromer in the recycling of receptors to the plasma membrane (Gomez and Billadeau, 2009; Derivery et al., 2009; Temkin et al., 2011; Zech et al., 2011; Piotrowski et al., 2013; Steinberg et al., 2013 and reviewed in Seaman et al., 2013).
The WASH complex is a pentameric protein complex [comprising WASH1, strumpellin (also known as KIAA0196), KIAA1033 (also known as the strumpellin and wash interacting protein, SWIP), FAM21 and CCDC53] that is responsible for the generation of branched actin networks on endosomes and the creation of microdomains into which specific cargoes can be concentrated (Puthenveedu et al., 2010; Derivery et al., 2012).
The precise functions of the WASH complex proteins strumpellin, KIAA1033/SWIP and CCDC53 are currently unknown; however, both strumpellin and its binding partner KIAA1033 have been implicated in neurodegenerative diseases. Mutations in strumpellin are causative of an autosomal dominant form of hereditary spastic paraplegia (HSP), and a mutation in KIAA1033 has been linked to autosomal recessive intellectual disability (Valdmanis et al., 2007; Ropers et al., 2011). Depletion of either strumpellin or KIAA1033 also causes an upregulation of endosomal tubulation (Derivery et al., 2009; Harbour et al., 2010). Although the mechanism that underlies this phenotype is unknown, this indicates a link between the WASH complex and endosomal membrane deformation. Interestingly, the spastin protein that is encoded by the HSP gene SPG4 has also been shown to regulate tubulation of endosomal membranes (Allison et al., 2013).
The largest WASH complex member FAM21 comprises a globular head domain (∼200 amino acids) and an unstructured ‘tail’, which is ∼1,100 amino acids in length. The head domain binds to and assembles with the other components of the WASH complex, and the tail domain participates in multiple interactions with the VPS35 component of retromer through a series of repeated elements, termed leucine-phenylalanine-acidic (LFa) motifs (Harbour et al., 2012; Jia et al., 2012; Helfer et al., 2013). The WASH complex, thus, depends upon the retromer CSC to associate with the membrane (Harbour et al., 2010). The FAM21 tail interacts with an array of other proteins in addition to the retromer CSC (Harbour et al., 2012), such as the actin-capping proteins CAPZa and CAPZb, and the cargo adaptor SNX27 (Hernandez-Valladares et al., 2010; Temkin et al., 2011; Steinberg et al., 2013). Here, we report an interaction between the WASH complex and the DNAJC13 protein, alternatively termed receptor-mediated endocytosis-8 (RME-8).
The DNAJ family of proteins is implicated in regulating protein folding through an associated chaperone (reviewed in Qiu et al., 2006; Sterrenberg et al., 2011), and RME-8, indeed, interacts with the heat shock chaperone Hsc70 (Chang et al., 2004). Although RME-8 was first identified as a protein that is required for endocytosis in Caenorhabditis elegans, it is clear that this is due to its involvement in the post-endocytic transport steps (Zhang et al., 2001; Fujibayashi et al., 2008). Loss of RME-8 disrupts the trafficking of both CI-MPR and the epidermal growth factor receptor (EGFR) (Girard and McPherson, 2008; Shi et al., 2009; Popoff et al., 2009). RME-8 also binds SNX1 through its C-terminus, further supporting the hypothesis that RME-8 plays a role in the endosomal transport machinery (Shi et al., 2009; Popoff et al., 2009).
In this study, we show that RME-8 interacts with the WASH complex through the FAM21 tail domain. Consistent with previously published reports (Popoff et al., 2009), we find that depletion of RME-8 results in a profound increase in retromer-positive endosomal tubules, a phenotype similar to – but more severe than – the effects of destabilisation of the WASH complex. We further show that the kinetics of the association of SNX1 with membrane are impaired in RME-8-depleted cells; therefore, we propose that RME-8 acts to coordinate the function of the WASH complex and the retromer SNX dimer.
RME-8 interacts with the FAM21 tail
In a search for novel binding partners of the retromer CSC and its interactors, we identified RME-8 in association with the FAM21 tail by using a native immunoprecipitation experiment. Immunoprecipitations were performed from HeLa cells that had been transfected with GFP-tagged EHD1, VPS35, TBC1D5 (the RQ mutant, R169A and Q204A, which lacks catalytic activity), SNX3 or the ‘tail’ region of FAM21 (Fig. 1A). Numbers 1–8 indicate the proteins that were identified by mass spectrometry. Band 2, which migrated at ∼220 kDa, was found exclusively in association with the FAM21 tail and was identified as RME-8. The known FAM21 tail interactors (Harbour et al., 2012) CCDC93, CCDC22, CAPZa and CAPZb (bands 3, 4, 6 and 7, respectively) were detected in the same experiment, confirming the validity of this approach for accurately identifying FAM21 tail interactors. Supplementary material Table S1 shows the results of a second set of immunoprecipitations in which the whole sample was analysed by mass spectrometry, rather than selected bands. RME-8 was found specifically in association with the FAM21 tail.
To confirm the identification of RME-8, the immunoprecipitation experiment was repeated in cells that had been treated with siRNA targeting either TBC1D5 or RME-8 (Fig. 1B). Again, the band representing RME-8 was present only in association with the FAM21 tail. Silencing of RME-8 by RNAi abolished this band, whereas the siRNA knockdown of TBC1D5 had no effect.
It is now well established that retromer interacts with multiple elements in the FAM21 tail (Jia et al., 2012; Helfer et al., 2013). We, therefore, aimed to compare the interaction profile of RME-8 with that of retromer and other proteins that bind the FAM21 tail. We transfected cells with RFP-tagged FAM21 tail truncations that have been previously employed by Helfer and colleagues (Helfer et al., 2013), which divide the FAM21 tail into three approximately equal segments: U1 (residues 228–490), U2 (residues 490–934) and U3 (residues 934–1341). Subsequent immunoprecipitation of these fragments using RFP-trap beads was followed by SDS-PAGE and western blotting for interacting proteins (Fig. 1C). Our results revealed information about a range of proteins that bound to the FAM21 tail: the retromer-associated cargo adaptor SNX27 displayed similar interaction preferences to the retromer CSC proteins whereas FKBP15 and CAPZa interacted exclusively with the U3 construct, and CCDC22 interacted predominantly with the U1 region. RME-8 itself was capable of interaction with several elements of the tail domain, binding preferentially with U1 and minimally with U3; however, its strongest binding preference was for the full-length FAM21 tail.
We next sought to establish whether RME-8 interacts directly with the FAM21 tail in vivo, and used the approach of depleting binding partners of the FAM21 tail (identified in supplementary material Table S1) and then testing for retention of the RME-8–FAM21 interaction. Cells that stably expressed the full-length GFP-tagged FAM21 tail domain were treated with siRNA against VPS35, FKBP15, RME-8, CCDC22 or CCDC93. The GFP–FAM21 tail was immunoprecipitated, and its binding partners were detected by western blot (Fig. 1D). The amount of RME-8 that was detected in association with FAM21 tail was not markedly affected by any condition except for siRNA knockdown of RME-8. Loss of RME-8 did not affect any other interactors of the FAM21 tail. Interestingly, both the immunoprecipitations and cell lysates generated in this experiment indicated that CCDC22 and CCDC93 are mutually dependent because knockdown of either one caused depletion of the other. It is probable, therefore, that these proteins bind to each other. The SNX27 protein, although detected in low abundance in immunoprecipitations of the FAM21 tail (supplementary material Table S1), was also depleted by RNAi in a similar experiment (Fig. 1E) but had no effect upon the binding of RME-8 or any other interactors of the FAM21 tail.
To establish whether the interaction between RME-8 and the FAM21 tail is direct, we performed further experiments. We first repeated the experiment shown in Fig. 1D that incorporated a triple knockdown of FKBP15, VPS35 and CCDC22, and found that more RME-8 co-immunoprecipitated with the GFP-tagged FAM21 tail construct from cells with the triple knockdown than from control cells (Fig. 1F). This indicates that, not only are the depleted proteins not required for this interaction, but that they might compete with RME-8 for binding to the FAM21 tail.
A well-characterised interactor of RME-8 is Hsc70 (Chang et al., 2004). To determine whether this protein is necessary for the FAM21–RME-8 interaction, we depleted Hsc70 in cells that expressed GFP–FAM21-tail and performed an immunoprecipitation under the conditions described above (Fig. 1G). Loss of Hsc70 enhanced the FAM21–RME-8 interaction, once again indicating competition for binding sites. We next transfected HeLa cells with the GFP-tagged FAM21 tail and variants of the most readily identified FAM21 tail interactors (see supplementary material Table S1). CCDC22 and CCDC93 were co-transfected, as we had found that these proteins might require each other for stability. Immunoprecipitation was performed with an antibody against GFP, and the immunoprecipitated proteins were analysed by using western blotting (Fig. 1H). RME-8 co-immunoprecipitated only with the FAM21 tail and FKBP15. We showed that FKBP15 is not necessary for the RME-8–FAM21 tail association; therefore, the detection of RME-8 in this immunoprecipitation is most probably due to the presence of the WASH complex, which interacts with FKBP15 through FAM21.
RME-8 interacts with an intact WASH complex
Following its initial identification in C. elegans, RME-8 was characterised in Drosophila and Arabidopsis, and in rodent and human cell lines (Zhang et al., 2001; Chang et al., 2004; Silady et al., 2004; Girard et al., 2005; Silady et al., 2008; Fujibayashi et al., 2008). Human RME-8 is 2,243 amino acids long, it comprises a DNAJ domain that spans residues 1322–1388 and has four conserved regions (shown in light blue), termed ‘IWN repeats’ (Zhang et al., 2001), which contain seven invariant amino acid residues and are of unknown function (Fig. 2A). In RME-8 of C. elegans, the region spanning positions 1388–1950 contains a domain that is necessary and sufficient for binding to SNX1 (Shi et al., 2009).
To confirm the in vivo association between the WASH complex and RME-8, we transiently transfected HeLa cells with either full-length RME-8, or N-terminal or C-terminal truncated versions that have been employed previously (Fujibayashi et al., 2008) (Fig. 2A). Immunoprecipitation of these RME-8 fragments and western blotting for WASH complex components revealed that strumpellin and WASH1 associated with full-length RME-8 and a truncation protein that lacked the C-terminal 425 amino acid residues (ΔC425) but not with RME-8 that lacked the N-terminal 453 amino acid residues (the ΔN453 truncation; Fig. 2B). Therefore, the N-terminus of RME-8 is required for an interaction with the FAM21 tail, although expression of this region of RME-8 alone is insufficient to immunoprecipitate the WASH complex (C.L.F. and M.N.J.S., unpublished). When transfected cells were examined by immunofluorescence (Fig. 2C), full-length or ΔC425 RME-8 displayed a punctate distribution and partially colocalised with SNX1; however, ΔN453 RME-8 adopted a diffuse cytosolic distribution and was never observed on SNX1-positive puncta, consistent with previously published reports of RME-8 localisation (Fujibayashi et al., 2008).
To demonstrate that the pool of FAM21 that interacts with RME-8 is part of an intact WASH complex, we examined whether WASH complex components were able to co-immunoprecipitate RME-8. Tagged versions of the core WASH complex components strumpellin, KIAA1033, WASH1 and FAM21, plus the FAM21 tail, were transfected into HeLa cells and immunoprecipitated (Fig. 2D). Western blotting for associated proteins showed that endogenous RME-8 co-immunoprecipitates with each of the transfected WASH complex components; thus, RME-8, presumably, interacts with an intact WASH complex.
Membrane association of RME-8
The finding that the RME-8 N-terminus is required for both membrane association and interaction with FAM21 led us to examine which of the proteins that interact with RME-8 might regulate its endosomal localisation. Experiments to examine the colocalisation of endogenous RME-8 with the WASH complex revealed that stably expressed GFP–WASH1 colocalised strongly with both FAM21 and RME-8. The antibody against RME-8 was then used to examine the association of the endogenous protein with the membrane under a range of knockdown conditions (Fig. 3B). As RME-8 interacts with SNX1 through its C-terminus, rather than its N-terminus (Shi et al., 2009; Popoff et al., 2009), we considered it unlikely that SNX1 recruits RME-8 to the endosomal membrane. As expected, depletion of SNX1 and SNX2, both individually and simultaneously, did not displace RME-8 from the membrane. RNAi-mediated silencing of SNX3, which recruits the retromer CSC to endosomes and can associate with RME-8 (Vardarajan et al., 2012), resulted in perinuclear clustering of endosomal compartments but did not affect RME-8 membrane association. Association of RME-8 with the membrane was also unaffected upon knockdown of the WASH complex members FAM21 or strumpellin, and, in a similar manner, depletion of the FAM21 tail interactors FKBP15, SNX27, CCDC22 and CCDC93 had no effect either (C.L.F. and M.N.J.S., unpublished). Therefore, separate mechanisms govern the recruitment to the membrane and FAM21 binding of RME-8, both of which are dependent upon the RME-8 N-terminal region. It has been reported elsewhere that RME-8 binds membrane lipids directly (Xhabija et al., 2011).
We observed that, when RME-8 was silenced under these conditions, SNX1 was present on numerous long and branched tubules throughout the cell, a phenotype similar to the tubulation reported by Popoff and colleagues (Popoff et al., 2009). Closer characterisation revealed that, in common with the tubules generated by WASH complex destabilisation (Gomez and Billadeau, 2009; Harbour et al., 2010), these structures are also positive for components of the retromer CSC (Fig. 3C).
RME-8 regulates membrane tubulation
Upregulated endosomal tubulation is a well-established effect of depletion or suppression of various core WASH complex members (Gomez and Billadeau, 2009; Derivery et al., 2009; Harbour et al., 2010). To determine whether the phenotype we observed was linked specifically to loss of RME-8, rather than a generalised effect of depleting an FAM21-tail-binding protein, we performed siRNA knockdown of a range of proteins that interact with the FAM21 tail. The tubulation phenotype was observed only in RME-8-depleted cells (supplementary material Fig. S1). We next performed siRNA knockdown of RME-8 using each of the individual oligonucleotides from the SMARTpool. As the commercially available antiserum against RME-8 was unable to detect the protein in crude cell lysates, the efficacy of RME-8 knockdown was assayed by immunoprecipitation of the GFP–FAM21-tail followed by western blot analysis. RME-8 was present in control cells but not in any of the siRNA-treated conditions (Fig. 4A). When examined by immunofluorescence, cells containing numerous SNX1-positive tubules were observed in all RME-8-depleted conditions but not in the control (Fig. 4B).
The percentage of cells with SNX1-positive tubules under each condition was assessed by using a double-blind procedure across three independent experiments, ∼300 cells were scored per condition (Fig. 4C). Approximately 20% of control cells contained SNX1-positive tubules, and in cells that had undergone siRNA knockdown of RME-8 this value was invariably higher. The difference was found to be significant in four out of the five knockdown conditions.
Previous studies have reported the presence of tubules that are positive for both SNX1 and the retromer cargo CI-MPR in RME-8-depleted cells (Popoff et al., 2009). We aimed to extend this observation by examining the distribution of various cargo proteins in cells subjected to RME-8 knockdown (supplementary material Fig. S2). CI-MPR was, indeed, present in tubules in these cells, as was the transferrin receptor. GLUT1 (also known as SLC2A1) and α5 integrin, both of which require the WASH complex for retrieval to the plasma membrane (Zech et al., 2011; Steinberg et al., 2013), also decorated tubular domains, as did the retromer cargo SLC11A2 (also known as DMT1-II, Tabuchi et al., 2010).
We hypothesised that SNX1 is responsible for tubule generation in RME-8-depleted cells, and tested this by imaging tubules in cells that stably expressed SLC11A2–GFP (Seaman et al., 2009) and lacked both RME-8 and SNX1. Minimal SLC11A2 was observed on tubules in control cells and was never present on tubules in SNX1-depleted cells; however, SLC11A2 localised extensively to SNX1-positive tubules in RME-8 depleted cells (Fig. 4D). Tubules were rarely visible in cells that had been treated with siRNA against both RME-8 and SNX1; however, a small number of cells displayed a similar phenotype to cells that had been depleted of RME-8 only. It, therefore, appears that SNX1 makes an important contribution to this phenotype but is not exclusively responsible for driving tubule formation, possibly due to functional overlap with SNX2 (Rojas et al., 2007).
Both the WASH complex and the retromer SNX dimer have been reported to interact with microtubules (Gomez and Billadeau, 2009; Wassmer et al., 2009; Hong et al., 2009; Monfregola et al., 2010), and it is probable that the microtubule cytoskeleton plays an important role in regulating membrane morphology, providing both the force and the structural support necessary for endosomal tubule extension. Co-labelling for SLC11A2–GFP and tubulin in RME-8-depleted cells revealed that the tubular domains generated by loss of RME-8 align with microtubules (Fig. 4E). To confirm that microtubules are required for tubule formation and extension in these cells, we treated either control or RME-8-depleted cells with the microtubule depolymerizing agent nocodazole (Fig. 4F). This caused a complete loss of SNX1-positive tubules, even where RME-8 had been depleted, indicating that this phenotype depends upon the microtubule cytoskeleton.
When we examined WASH complex components in RME-8-depleted cells, we observed a marked increase in localisation to SNX1-positive tubules (supplementary material Fig. S3A–C). This is in contrast to wild-type cells, where the WASH complex is typically restricted to the vesicular endosome (Derivery et al., 2012). FKBP15, which interacts with the FAM21 tail (supplementary material Fig. S3D), did not display such an obvious shift in distribution, possibly due to an overall lower abundance of this protein in the cell. In two separate experiments, we quantified the extent to which the WASH complex localised to tubules under control and RME-8-depleted conditions, and found that there was a clear shift to a tubular distribution (supplementary material Fig. S3E). When this knockdown was performed in cells expressing hepatocyte growth factor-regulated tyrosine kinase substrate (HRS, also known as HGS) tagged with GFP, HRS showed a similar redistribution to that of the WASH complex (supplementary material Fig. S3F), suggesting that numerous components of the endocytic system are affected by loss of RME-8.
In order to delineate more clearly the difference between typical retromer-positive tubules and the RME-8-knockdown phenotype, we quantified the number of tubule branch points in 40 control and 40 RME-8-depleted cells that had been labelled for SNX1 (Fig. 5A). We found a large difference in the mean number of branch points per cell between the different conditions (2.35 in control cells but 13.55 in RME-8-knockdown cells). Structures considered as branch points are labelled in the example images. This experiment was then repeated, labelling the cells for VPS35 (Fig. 5B), and we found an average of 5.35 branch points per cell in the controls but 17.75 under conditions of RME-8 knockdown. We, therefore, suggest that the tubular network observed in RME-8-knockdown cells represents a profound change in the morphology of the endocytic system. Closer examination of these tubules was permitted by the use of super-resolution direct stochastic optical reconstruction microscopy (dSTORM) (Fig. 5C). In keeping with previous data, this showed that VPS26 is present on short tubular transport intermediates in control cells but decorates long and branched tubules in cells lacking RME-8.
We hypothesised that this extensive network is sustained because of the hyper-stability of the constituent tubules, and we investigated this using live-cell microscopy. Control or RME-8-knockdown cells that stably expressed GFP–SNX1 were imaged for 60 seconds (supplementary material Movies 1 and 2); Fig. 5D shows example images captured during the first 30 seconds. SNX1-positive structures in control cells were highly dynamic, usually persisting for no more than a few seconds. However, in RME-8-depleted cells, tubules were present for an extended period of time and, typically, displayed little movement. This finding is in agreement with the previous report of Popoff and colleagues (Popoff et al., 2009).
Loss of RME-8 does not destabilise the WASH complex
As well as upregulating endosomal tubulation, loss of strumpellin or KIAA1033 causes destabilisation of the WASH complex itself (Gomez and Billadeau, 2009; Derivery et al., 2009; Harbour et al., 2010; Harbour et al., 2012; Gomez et al., 2012). To establish whether the tubulation observed in RME-8-depleted cells is the result of a similar mechanism, we examined the cellular levels of strumpellin, FKBP15, SNX1 and VPS26 in cells that had been subjected to RME-8 knockdown (Fig. 6A). RME-8 knockdown had no effect upon any of these proteins; therefore, enhanced tubulation is unlikely to occur as the result of a deficit of retromer or WASH complex members. WASH complex destabilisation was specifically examined by comparing the effects of RME-8 knockdown with those of KIAA1033 knockdown in cells that stably expressed GFP–WASH1 (Fig. 6B). Where KIAA1033 had been depleted, GFP–WASH1 did not co-immunoprecipitate strumpellin or FAM21, indicating loss of the intact WASH complex. No deficit in the assembly of the WASH complex was observed in RME-8-depleted cells; thus, the tubulation phenotype must occur through a different mechanism.
We next investigated the possibility that loss of the WASH complex impairs the association of RME-8 with SNX1. We examined the ability of GFP–SNX1 to co-immunoprecipitate RME-8 in cells that had been depleted of KIAA1033; the destabilisation of the WASH complex was assayed by the western blotting of lysates for strumpellin and FAM21 (Fig. 6C). Knockdown of KIAA1033 did not affect the amount of RME-8 that was found in association with GFP–SNX1. Although an intact WASH complex was not required for the RME-8–SNX1 interaction, it might be necessary for some other aspect of RME-8 regulation, such as its localisation to specific endosomal subdomains.
Loss of spastin depletes RME-8
Mutations in the WASH complex component strumpellin are causative of HSP, indicating a potential link between aberrant endosomal tubulation and neurodegenerative disease (Valdmanis et al., 2007; Harbour et al., 2010). This is supported by a recent study that reported that loss of the microtubule severing protein spastin (mutations in which are implicated in a high proportion of HSP) upregulates endosomal tubulation and causes mistrafficking and degradation of the transferrin receptor (Allison et al., 2013). As there are clear similarities between the effects of loss of spastin function and depletion of RME-8, we investigated a potential link between the two proteins. We first examined whether knockdown of RME-8 caused a corresponding loss of spastin by western blotting of cellular lysates (Fig. 6A). RME-8-depleted lysates displayed no spastin deficit. As we did not have an antibody that could be used to detect RME-8 in cell lysates, we performed an immunoprecipitation experiment in cells that expressed the GFP-tagged FAM21 tail construct, and examined whether spastin knockdown reduced the amount of RME-8 that was co-immunoprecipitated (Fig. 6D). There was a pronounced reduction in the amount of RME-8 that was detected in association with the FAM21 tail in cells that had been depleted of spastin; however, the levels of strumpellin, VPS35 and SNX27 that were associated with the FAM21 tail were unaffected. It, therefore, appeared that loss of spastin either caused depletion of RME-8 or affected the ability of RME-8 to participate in protein–protein interactions. We next performed an immunoprecipitation from the GFP–SNX1 stable cell line, and, once again, examined the amount of RME-8 that was pulled down (Fig. 6E). In keeping with our previous finding, loss of spastin caused a marked reduction in the amount of RME-8 that co-immunoprecipitated with GFP–SNX1. Finally, the cellular levels of RME-8 were examined by immunofluorescence (Fig. 6F). RME-8 fluorescence was reduced in cells that had been treated with siRNA against spastin, possibly due to degradation or loss of expression of RME-8.
RME-8 regulates the dynamics of SNX1 on the membrane
DNAJ-domain-containing proteins can have important roles in regulating the dynamics of macromolecular complexes – for example, the auxilins are required for the disassembly of clathrin coats (Ungewickell et al., 1995; Greener et al., 2000; Umeda et al., 2000; Hirst et al., 2008, and reviewed in Eisenberg and Greene, 2007). Our findings indicated that RME-8 might perform a regulatory function in relation to the spatial or temporal regulation of the SNX dimer or the WASH complex. We therefore used fluorescence recovery after photobleaching (FRAP) to examine SNX1 dynamics in RME-8-depleted cells. Control or RME-8-depleted cells that stably expressed GFP–SNX1 were examined by live-cell microscopy; small areas containing tubular domains were then subjected to photobleaching, and the recovery of GFP–SNX1 was monitored over time. Fig. 7A shows representative control and RME-8-knockdown cells. Approximately 35 cells per condition were examined, and the fluorescence recovery profiles were normalised and analysed. The average fluorescence recovery profiles indicate a clear disparity between control and RME-8-knockdown cells (Fig. 7B), with recovery during the 10–15 seconds immediately post-bleach markedly slower in RME-8-depleted cells. This observation is supported by the mean t½ values, which represents the time taken post-bleach for GFP–SNX1 fluorescence to recover to 50% of the final fluorescence intensity (Fig. 7C). The mean t½ was 7.8 seconds in control cells but 14.3 seconds in RME-8-depleted cells. Analysis of these data by Student's t-test showed that this difference was highly significant.
Having observed that depletion of RME-8 causes a deficit in the exchange of SNX1 between the membrane-associated and cytoplasmic pools, we next examined cells treated with the phosphatidylinositol 3-kinase inhibitor wortmannin, which has been shown to displace SNX1 from membranes (Cozier et al., 2002; Rojas et al., 2007). SLC11A2–GFP-expressing cells were treated with siRNA against RME-8 and incubated with wortmannin for 0, 10, 20 or 30 minutes. The extent of SNX1 localisation to endosomal membranes was then examined by immunofluorescence (Fig. 7D) that revealed a small, but noticeable, difference between control and knockdown cells. SNX1 was observed on SLC11A2-positive membranes in RME-8-depleted cells, even following prolonged exposure to wortmannin, in contrast with control cells, where little SNX1 was observed on membranes after wortmannin treatment. Images captured at the 20 minutes timepoint were analysed using automated microscopy software, and the spot intensity of SNX1 labelling was normalised against the mean SLC11A2–GFP fluorescence under each condition. This revealed a ∼50% increase in SNX1 fluorescence intensity in RME-8-depleted cells compared with the control. Taken together, these data further support the interpretation that RME-8 contributes to the regulation of SNX1 activity – in cells lacking RME-8, SNX1 might form part of larger, less dynamic, complexes that are slow to dissociate from the membrane.
In this study we have identified an interaction between the DNAJ protein RME-8 and the tail domain of the FAM21 protein in the WASH complex. Loss of RME-8 function results in a profound tubulation of endosomes that leads to an extensive branched network of tubules, a phenotype that is similar to, but more extensive than, the endosomal tubulation that results from loss of the WASH complex. RME-8 has previously been shown to bind SNX1, and we show that the kinetics of SNX1 association and dissociation with membranes is altered when RME-8 function is lost. Our findings, therefore, indicate that RME-8 regulates endosomal protein sorting by coordinating distinct elements of the membrane trafficking machinery to regulate formation of endosomal tubules.
RME-8 binds the WASH complex
We have shown by co-immunoprecipitation that RME-8 binds FAM21, interacting preferentially with the first third of the tail domain. Conversely, the FAM21 head domain, which can assemble with WASH complex components, does not co-immunoprecipitate RME-8. Because other core members of the WASH complex were also able to co-immunoprecipitate RME-8, we conclude that RME-8 interacts with an intact and, presumably, functional WASH complex. The first 453 amino acids of the RME-8 N-terminal region were necessary </emph>for this interaction to be maintained. We show that the same region of RME-8 is required for association of RME-8 with the membrane as Fujibayashi and colleagues have previously reported (Fujibayashi et al., 2008); however, neither loss of the retromer CSC – which recruits the WASH complex to the endosome – nor the WASH complex itself caused RME-8 to dissociate from the membrane, possibly because RME-8 can bind to lipids directly (Xhabija et al., 2011).
By using mass spectrometric analysis of immunoprecipitations with the FAM21 tail (see supplementary material Table S1), it was possible to identify every protein that associated with the FAM21 tail (under the conditions employed for the immunoprecipitation). This allowed each of the FAM21-tail-interacting proteins to be tested for a role in mediating the association between RME-8 and the FAM21 tail. Knockdown of each of the proteins that had been found to interact with the FAM21 tail (with the exception of RME-8 depletion) did not prevent RME-8 co-immunoprecipitating with the FAM21 tail; therefore, we conclude that this association is most likely to be direct. Indeed, the triple knockdown of VPS35, FKBP15 and CCDC22 led to an increased association between the FAM21 tail and RME-8, possibly because proteins that might compete with RME-8 for binding to the FAM21 tail were no longer present.
As we were able to demonstrate bi-directional interactions between RME-8 and the WASH complex but not between RME-8 and SNX1, we cannot state that RME-8 is able to interact with the WASH complex and SNX1 simultaneously. However, as RME-8 requires its N-terminus to bind FAM21 and uses its C-terminus to interact with SNX1, it is possible that one of the key roles for RME-8 is to link these two elements of membrane traffic machinery.
If RME-8 does, indeed, coordinate elements of the WASH and retromer complexes by associating with FAM21, we would expect RME-8 and FAM21 to be conserved to a similar extent through evolution. An analysis of the conservation of RME-8 (Koumandou et al., 2013) concluded that it is well conserved but with numerous secondary losses. Retromer was first identified in yeast and is ubiquitously conserved (Seaman et al., 1997; Seaman et al., 1998; Koumandou et al., 2011). The WASH complex, although widely conserved and likely to be ancient in origin (Derivery and Gautreau, 2010), is not conserved in yeast – which is also true for RME-8. We performed a search for FAM21 homologues in the species that had been included in the study of Koumandou and colleagues (Koumandou et al., 2013) and found that, in the higher eukaryotes, RME-8 and FAM21 are generally co-retained (supplementary material Table S2). FAM21 homologues in these species typically included an extended tail domain, indicating a conserved interaction and cooperative function for FAM21 and RME-8 throughout evolution. It is also noteworthy that the expression profiles of RME-8 and FAM21A in human are similar, consistent with a functional interaction (see supplementary material Fig. S4).
The importance of RME-8 in membrane trafficking
First identified in a C. elegans screen for mutants that displayed deficits in yolk protein uptake (Zhang et al., 2001), RME-8 was subsequently shown to be important in the post-endocytic transport steps – in particular, in EGFR recycling or degradation and CI-MPR retrieval (Girard et al., 2005; Girard and McPherson, 2008; Shi et al., 2009; Popoff et al., 2009). We propose that a factor in the importance of RME-8 for membrane trafficking is its role in coordinating the WASH complex and the sorting nexins. Further supporting this argument, the WASH complex is implicated in many of the same transport steps as RME-8. Loss of the WASH complex has been reported to impair both retrieval of CI-MPR to the Golgi and degradation of EGFR (Gomez and Billadeau, 2009; Duleh and Welch, 2010). However, although both RME-8 and the WASH complex have been shown to be important in EGFR degradation, SNX1 does not appear to be required for this process (Carlton et al., 2004).
RME-8 and membrane tubulation
We have observed that the siRNA-mediated knockdown of RME-8 creates a network of highly branched membrane tubules, which accumulate components of the retromer CSC and contain numerous cargo proteins that require retromer for their steady-state localisation. Our observations agree with, and extend, those of Popoff and colleagues (Popoff et al., 2009).
In light of the striking tubulation phenotype in cells that lacked RME-8, it is tempting to speculate that RME-8 must have an important role in regulating its binding partner SNX1. We have demonstrated, both by FRAP and by inhibition of phosphatidylinositol 3-kinase, that SNX1 activity is, indeed, altered in the absence of RME-8 – SNX1 was slower both to detach from and to recover onto the endosomal membrane. This might indicate that, under these conditions, SNX1 maintains an abnormally close association with its binding partners or with the membrane, causing the tubulation activity of its BAR domain to be upregulated or prolonged.
There is increasing evidence that abnormal membrane deformation has relevance to neurodegeneration. Mutations in the WASH complex members strumpellin and KIAA1033 are causative of neurodegenerative disease (Valdmanis et al., 2007; Ropers et al., 2011), whereas the HSP proteins spastin, atlastin, REEP1 and RTN2 interact to regulate the membrane morphology of the endoplasmic reticulum (reviewed in Hu et al., 2011). Recently, a Parkinson-disease-causing mutation in RME-8 has been reported (Vilarino-Guell et al., 2013), strengthening the link between endosomal tubulation and neurodegenerative disease. Another recent report has shown that spastin, because of its microtubule-severing capabilities and interaction with elements of the ESCRT complex, is also important for endosomal membrane morphology (Allison et al., 2013). We have found that loss of spastin causes a decrease in the levels of RME-8. Although further work will be needed to elucidate the mechanism by which this occurs, it is possible that partial loss of RME-8 function is a contributing factor to the tabulation that results from knockdown of spastin.
RME-8 links actin-regulating and membrane-tubulating components of trafficking machinery
The retromer CSC, and several other proteins, interact with multiple elements of the FAM21 tail, whereas RME-8 displays a strong preference for the tail region proximal to the FAM21 head (Fig. 8A). RME-8 also binds minimally to the final portion of the tail, which is the site of interaction for the FKBP15 and CAPZa1 proteins. Fig. 8B presents a model for the WASH and retromer complexes at endosomal transport domains. The SNX dimer is present on tubular transport intermediates, where it drives membrane tubulation and maintains an interaction with microtubules through dynactin and dynein. The retromer CSC, recruited to the membrane by SNX3 and Rab7a, packages cargo proteins into tubules and, as such, is also present on the tubular portions of the endosomal membrane. The WASH complex, however, is typically limited to the vesicular endosome, where it generates an actin network that both stabilises and defines the transport microdomain. RME-8 can act as a link between SNX1 and the WASH complex. We suggest that this imparts both temporal and spatial regulation by ensuring that all requisite elements of the transport machinery are present at the same domain, simultaneously regulating the membrane association of SNX1.
The finding that RME-8 interacts with FAM21 and might, thus, link the WASH complex with the retromer SNX dimer could help resolve unanswered questions concerning the WASH complex. In particular, the role of the WASH complex in membrane tubulation might now prove to be linked to RME-8 function. Furthermore, loss of RME-8 could be a contributing factor in the tubulation resulting from spastin depletion. The task now will be to examine the role of RME-8, if any, in this phenotype and to dissect the protein's precise mechanism of action. Ultimately, this could yield important insights both into the WASH complex and into the broader process of membrane tubulation.
MATERIALS AND METHODS
The RME-8 antibody used for immunofluorescence was a gift of Peter McPherson (McGill University, Montreal, QC, Canada). Antisera against RME-8, CCDC22, WASH1 and spastin were purchased from Sigma-Aldrich; the antibody against CCDC93 was purchased from Source Bioscience; an antibody against mCherry was purchased from Clontech; antibodies against SNX27, Hsc70 and GLUT1 were purchased from Abcam; a polyclonal antibody against CAPZa1 (F-actin capping protein) was from Millipore; polyclonal antibodies against VPS26 and SNX1 have been described previously (Seaman, 2004); antisera against strumpellin, FAM21 and FKBP15, used for western blotting, have been described previously (Harbour et al. 2010; Harbour et al. 2012); the antibody against FAM21, used for immunofluorescence, was purchased from Santa Cruz Biotechnology; the rabbit antibody against VPS29 was generated using a peptide corresponding to the C-terminus of VPS29 as an antigen and has been described previously (C. F. Skinner, A structure-based functional analysis of the retromer Vps29 and Vps26 proteins, PhD thesis, University of Cambridge, 2007).
The antibody against VPS35, used for both western blotting and immunofluorescence was purchased from Santa Cruz Biotechnology. For immunofluorescence, antibodies against the following proteins were used: WASH1 (C-terminus; Millipore), FKBP15 or CI-MPR (Abcam); SNX1 or α5 integrin (BD Biosciences); tubulin (Sigma-Aldrich); and transferrin receptor (Zymed). The monoclonal antibody against GFP and the fluorescently conjugated secondary antibodies used for immunofluorescence were from Invitrogen. The polyclonal antibody against GFP that was used for western blotting and immunoprecipitation experiments has been described elsewhere (Seaman et al., 2009). Horseradish-peroxidase-conjugated Protein-A and secondary antibodies against mouse IgGs were obtained from Millipore and Sigma-Aldrich, respectively.
Plasmids expressing the RFP-tagged FAM21 tail fragments have been described previously (Helfer et al., 2013). GFP-tagged RME-8 was a gift from Kiyotoshi Sekiguchi (Osaka University, Osaka, Japan). GFP–FKBP15 was a gift from Fumio Nakamura (Yokohama City University, Yokohama, Japan). GFP-tagged strumpellin, KIAA1033, FAM21 and FAM21 tail have been described previously (Harbour et al., 2010; Harbour et al., 2012). GFP–VPS35 has been described elsewhere (Gokool et al., 2007a). CCDC93–GFP and CCDC22–Myc were purchased from Origene. For GFP–WASH1, murine WASH1 cDNA was obtained from Source BioScience and amplified by using PCR, which incorporated BamH1 and Sal1 sites at the 5′ and 3′ ends, respectively. The PCR product was cloned first into PCR Blunt (Invitrogen) and then subcloned into pEGFP-C1 that had been cut with BamH1 and Sal1. This construct was used for transient transfection experiments.
Cell culture and stable cell lines
Adherent HeLa cells were cultured as described previously (Seaman et al., 2009). For live-cell experiments, CO2-independent medium (Gibco) containing 10% foetal calf serum (FCS), penicillin and streptomycin and L-glutamine was used. Cells that stably expressed GFP–SNX1 and GFP–FAM21-tail have been described previously (Harbour et al., 2012). SLC11A2–GFP- and HRS–GFP-expressing cells have been described elsewhere (Seaman et al., 2009). For the stable cell line expressing GFP–WASH1, the GFP-WASH1 construct was cloned into the pLXIN vector (Clontech) and virus (standard retrovirus envelope) was produced using Phoenix cells (Clontech). Virus particles were used to infect HeLa cells and a mixed stable population was selected using Geneticin (G418).
The transfection of cells for immunofluorescence microscopy was carried out using the Effectene kit (Qiagen) following the manufacturer's instructions. Cells prepared for immunoprecipitation experiments were transfected, as described previously (Helfer et al., 2013), using polyethylenimine (PEI; Polysciences).
siRNA transfection was carried out using Oligofectamine (Invitrogen) as previously described (Harbour et al., 2010). All siRNAs used in this study were purchased from Dharmacon (Thermo Scientific). Knockdown of spastin was performed using siGENOME oligonucleotide 1 as described previously (Allison et al., 2013). All other knockdowns were performed using ON-TARGET siRNA SMARTpools, with the exception of the individual oligonucleotides that were used in experiments to quantify tubulation in RME-8-depleted cells.
Native immunoprecipitation experiments were carried out as described previously (Seaman, 2007; Seaman et al., 2009; Harbour et al., 2012). For the immunoprecipitation of mCherry-tagged proteins, RFP-trap beads (Chromotek) were used. The immunoprecipitation experiments were performed as described previously (Breusegem and Seaman, 2014).
SDS-PAGE and western blotting
SDS-PAGE and protein transfer was carried out as previously described (Seaman, 2004). Western blotting was performed as described previously (C. L. Freeman, The Hereditary Spastic Paraplegia protein strumpellin and the WASH complex in neuronal and non-neuronal cells, PhD thesis, University of Cambridge, 2012).
Mass spectrometric analysis of selected gel bands was performed as described previously (Seaman et al., 2009).
Immunofluorescence microscopy and tubule quantification
Immunofluorescence microscopy was carried out as described elsewhere (Harbour et al., 2010; Harbour et al., 2012). For microtubule depolymerisation, cells were incubated at 37°C in a solution of 33 nM nocodazole (Sigma-Aldrich) in pre-warmed culture medium for 1 hour. Control cells were incubated in nocodazole-free media. For wortmannin treatment, cells were incubated at 37°C in warmed serum-free medium (negative control) or serum-free media containing 100 nM wortmannin (Sigma-Aldrich). Cellomics ArrayScan analysis software was used to perform spot intensity analysis of SNX1 labelling in these cells using six control and six RME-8 knockdown images.
Quantification of the endosomal tubulation phenotype in cells that had been treated with RME-8 siRNA was carried out using a double-blind procedure. Randomly selected images of each condition were captured across three experiments and assigned identifiers using an online randomizer (http://www.randomizer.org/form.htm). For experiments quantifying tubule branch points, black-and-white inverted images of 40 cells from each condition were assigned random identifiers and printed out for branch point scoring.
Labelling of cells for dSTORM was performed as for other immunofluorescence experiments. The coverglass was coated with glycine by incubation in a solution of 2 M glycine for 30 minutes at 37°C before seeding cells. Images were captured by using a custom-built STORM microscope with an Olympus IX71 microscope and Andor iXON DU897 EMCCD camera (see Metcalf et al., 2013). Samples were illuminated with a high-power 640-nm laser in a glucose-oxidase- and monoethanolamine-based switching buffer (Metcalf et al., 2013), and 20,000 frames were acquired with an exposure of 10 mseconds. Images were reconstructed using the rainSTORM software (Rees et al., 2013) resulting in super-resolution images with a mean localisation precision of 25 nm (Rees et al., 2012) and reconstructed with 25-nm pixels. Further processing of images was carried out using ImageJ.
Live-cell microscopy and FRAP
Wide field microscopy was performed using a Zeiss Axio Observer Z1 microscope and a Hamamatsu EMCCD camera. Images were captured continuously for 60 seconds; image analysis was performed using the Zen 2011 software. FRAP was performed as described previously (Kozik et al., 2013) using a Zeiss LSM710 incubated confocal microscope and Zen 2010 software. Fluorescence recovery curves were generated by calculating the average timepoints for all cells in each data set. Data normalization and analysis was carried out using Microsoft Excel, Prism and GraphPad.
We are very grateful to Peter McPherson, Kiyotoshi Sekiguchi and Fumio Nakamura for generously providing antisera against RME8, GFP-tagged RME-8 and the GFP-FKBP15 construct, respectively; we thank Robin Antrobus and Kamburpola Jayawardena for the mass spectroscopy analysis of native immunoprecipitations (Cambridge Institute for Medical Research); Daniel Metcalf (National Physical Laboratory, London, UK) optimised and carried out dSTORM; and Mark Bowen and Matthew Gratian (Cambridge Institute for Medical Research) offered invaluable assistance with live-cell microscopy. We thank our colleagues Folma Buss, Sophia Breusegem, Rachel Allison and Nicola Hodson for assistance during this project.
C.L.F. performed the majority of the experiments and wrote the manuscript. G.H. provided the GFP-WASH1 cells. M.N.J.S. performed initial native immunoprecipitations of the FAM21 tail, directed the experiments and edited the manuscript.
This work was funded by a Medical Research Council Senior Research Fellowship awarded to M.N.J.S. [grant number G0701444]. Deposited in PMC for release after 6 months.
The authors declare no competing interests.