Identification of a novel conserved sorting motif required for retromer-mediated endosome-to-TGN retrieval

The function of the different organelles that comprise the secretory and endocytic pathways in eukaryotic cells is determined by the complement of resident proteins present within the respective organelle. The mechanisms that govern membrane protein localisation to the different organelles rely on intrinsic information in the membrane protein such as sorting motifs in the cytoplasmic domain and extrinsic factors such as coat-proteins that recognise the sorting motifs.

A well-studied example of these concepts is the sorting of membrane proteins by the clathrin-coated vesicle (CCV) adaptor proteins through recognition of the YXXΦ sorting motif. This tetra-peptide motif, comprising Tyr-X-X (where X is any amino acid)-bulky hydrophobic, is bound by the medium chain (μ) subunit of adaptor complexes resulting in the protein containing the YXXΦ motif being concentrated in CCVs (Ohno et al., 1995; Owen and Evans, 1998) (reviewed in Bonifacino and Traub, 2003). Another example of sorting motifs interacting with coat proteins is the recognition of acidic di-leucine motifs by Golgi-associated γ-ear-containing ARF-binding (GGA) proteins. GGA proteins bind to the consensus sequence, Asp-X-X-Leu-Leu (DXXLL) and function at the trans-Golgi network (TGN) to direct specific cargo proteins into CCVs (Puertollano et al., 2001; Misra et al., 2002). Sorting motifs, therefore, are often just two to four amino acids present in the correct molecular context. YXXΦ motifs must be at least seven amino acids from the transmembrane domain (Rohrer et al., 1996) whereas acidic di-leucine motifs are usually near the end of the cytoplasmic domain of proteins that traffic between the TGN and endosomes.

Membrane proteins that traffic in the post-Golgi endo-membrane system often contain multiple sorting motifs that function to direct the transport of the membrane protein between the various compartments such as the TGN, endosomes and the plasma membrane. A good example of this is the cation-independent mannose 6-phosphate receptor (CIMPR) which binds to lysosomal hydrolases in the TGN and is sorted into CCVs by GGA proteins in concert with the AP-1 clathrin adaptor (Ghosh et al., 2003; Doray et al., 2002). The cytoplasmic tail of the CIMPR, therefore, has both YXXΦ and DXXLL motifs. In the case of the CIMPR, the affinity of the GGAs for DXXLL motifs is augmented by casein kinase-mediated phosphorylation of a serine residue that immediately precedes the aspartate of the acidic di-leucine (Kato et al., 2002). Following the delivery of the CIMPR to an endosome, the CIMPR releases its ligand and is then recycled back to the TGN to undergo further rounds of hydrolase sorting. Retrieval of the CIMPR from the endosome to the TGN is mediated by the retromer complex (Seaman, 2004; Arighi et al., 2004; Carlton et al., 2004) and has also been shown to require the function of the TIP47 and PACS1 proteins (reviewed by Seaman, 2005; Bonifacino and Rojas, 2006).

Retromer was first described in yeast and is required for the endosome-to-Golgi retrieval of the vacuolar hydrolase receptor Vps10p, the yeast functional equivalent of the CIMPR (Seaman et al., 1997; Seaman et al., 1998). Retromer comprises five proteins in yeast, Vps35p, Vps29p, Vps26p, Vps5p and Vps17p, which are conserved in mammals (with the exception of Vps17p that has no clear homologue) (reviewed by Seaman, 2005). Retromer can recognise cargo such as Vps10p and the CIMPR through interactions with Vps35p or mammalian VPS35, respectively (Nothwehr et al., 2000; Arighi et al., 2004). Retromer can also reshape the membrane by the action of the sorting nexin components Vps5p and its mammalian homologue, sorting nexin-1 (SNX1) (Seaman et al., 1998; Carlton et al., 2004). Recent structural and biochemical studies of VPS29 have revealed that VPS29 is a phosphoesterase, which can dephosphorylate the serine preceding the acidic di-leucine (SDEDLL) motif in the CIMPR cytoplasmic tail (Collins et al., 2005; Damen et al., 2006). Interestingly, the cytoplasmic tail of sortilin, another protein that is retrieved by retromer (Seaman, 2004) contains an identical SDEDLL sequence. So, in addition to VPS35, VPS29 must also be able to interact with cargo proteins. In fact a third component of retromer, VPS26, might also be able to interact with cargo. The crystal structure of VPS26 has recently been solved, revealing an unexpected structural similarity between VPS26 and the arrestin family of proteins (Shi et al., 2006). As arrestins have been shown to bind cargo proteins in plasma membrane-associated CCVs (Zhang et al., 1997) it is possible that VPS26 could play a similar role within the retromer complex. Presently, however, no sorting motif for retromer has yet been identified.

In this study we have used a reporter-protein-based approach to identify the critical region of the CIMPR tail required for endosome-to-Golgi retrieval and for interaction with retromer. We have identified a highly conserved tri-peptide motif with the consensus Trp/Phe-Leu-Met/Val as being necessary for the endosome-to-Golgi retrieval of the CIMPR and sortilin. Mutation of the tri-peptide motif results in failure of the reporter protein to be retrieved properly causing the reporter protein to be rapidly degraded.

We have previously shown that a CD8-CIMPR-reporter protein (which contains the full length bovine CIMPR cytoplasmic domain) can be expressed in HeLa cells and will then cycle between endosomes and the TGN, the endosome-to-TGN retrieval being dependent upon retromer (Seaman, 2004). Transport from endosomes to the TGN can be monitored by an antibody-uptake assay using a monoclonal antibody against the lumenal domain of CD8. Colocalisation with the TGN marker protein, TGN46, is used to determine if the CD8-reporter protein has reached the TGN.

In order to establish that antibody uptake requires binding to the CD8 lumenal domain and does not occur by fluid-phase endocytosis, cells expressing the CD8-CIMPR reporter were mixed with untransfected HeLa cells and then incubated with antibody for 3 hours. After fixation, the cells were labelled with antibodies to SNX1 (which will be present in transfected and untransfected cells) and fluorescently labelled anti-mouse antibodies to reveal which cells had endocytosed the anti-CD8 monoclonal antibody (mAb). In Fig. 1A, the immunofluorescence images show that the anti-CD8 mAb was taken up only in two of the cells in the field whereas SNX1 is present in all the cells. Even when the exposure time is increased fivefold (right hand panel), the anti-CD8 antibody is visible in only two of the five cells. Fig. 1B shows the results of the antibody uptake assay repeated using either CD8-CIMPR-transfected or -untransfected cells grown on tissue culture plastic and incubated with anti-CD8 mAb before fixation and incubation with ¹²⁵I-protein-A to quantitatively determine the amount of anti-CD8 endocytosed. Only the cells transfected with the CD8-CIMPR reporter and incubated with the anti-CD8 mAb registered above background levels of endocytosed anti-CD8, demonstrating that there is no significant fluid-phase uptake of the anti-CD8 mAb.

The cytoplasmic domain of the CIMPR is approximately 160 amino acids in length which precludes alanine-scanning mutagenesis as a viable approach. Therefore, the sequences of the cytoplasmic domains of the CIMPR proteins from various species were compared (Fig. 2A), to identify highly conserved regions which would be most likely to contain retromer sorting motifs. It is clear that the most highly conserved regions of the CIMPR tail comprise amino acids 1-100 and 150-160. The region in the middle of the tail is less highly conserved. The various sorting motifs in the CIMPR tail that have been shown to bind to AP-1, AP-2, TIP47, PACS-1 and the GGA proteins (reviewed by Ghosh et al., 2003) are indicated by the black bars. In Fig. 2B, the schematic diagram shows the first set of truncations of the CIMPR tail. The Δ1-19ΔLL truncation was generated using the rat CIMPR tail (Reaves et al., 1996). The other truncation constructs each removed ∼70-90 amino acids of the CIMPR tail and therefore are broadly equivalent to each other in terms in the size of the remaining CIMPR tail. The Tail19-55 construct was generated after analysing the effects of the other truncations on the ability of the construct to be retrieved from the endosome to the TGN.

Using the antibody uptake assay, the ability of the different truncations of the CIMPR tail to be retrieved was assessed. In Fig. 3 the CD8-CIMPR reporter can be efficiently retrieved and colocalise with TGN46 at the TGN, consistent with our previously published findings (Seaman, 2004). Similar results were obtained for the Δ1-19ΔLL, the Δ75-163 and the Δ56-143 truncations. The Δ1-75 truncation, however, did not retrieve to the TGN and instead accumulated in a punctate compartment similar in morphology to endosomes and lysosomes. The region of the CIMPR tail that is present in all the truncations that did retrieve to the TGN, but absent in the Δ1-75 truncation, encompasses amino acids 19-55. A CD8-reporter construct (Tail19-55) was generated that contained just amino acids 19-55 of the bovine CIMPR tail. This was found to efficiently retrieve to the TGN and very extensively colocalised with TGN46. This region of the CIMPR tail therefore appears to be sufficient to direct retrieval of the CIMPR and should contain the retromer interacting motif.

The inability of the Δ1-75 truncation to undergo endosome-to-TGN retrieval is not due to an inability of the construct to be endocytosed. In Fig. 4A, endocytosed Δ1-75 truncation can be seen to accumulate in a compartment that is positive for VPS26 (shown enlarged in inset boxes). Previous studies by ourselves and others (Seaman, 2004; Arighi et al., 2004) have shown that loss of retromer function results in the rapid degradation of the CIMPR. If the Δ1-75 construct is not efficiently retrieved then a predictable consequence would be the rapid degradation of the Δ1-75 truncation. This was tested using a cycloheximide chase protocol which blocks the production of newly synthesized CD8-reporter constructs, enabling the monitoring of the fate of pre-existing reporter constructs. The stability of the CD8-CIMPR, the Δ75-163, Δ56-143 and Δ1-75 truncations was assessed (Fig. 4B). The truncations that retrieved to the TGN were stable during a 6-hour chase, whereas the Δ1-75 truncation is very unstable and was undetectable after 3 hours chase.

The data presented in Fig. 3 demonstrates that the region of the CIMPR tail that contains amino acids 19-55 is sufficient for endosome-to-TGN retrieval. Although this is still quite large, it is feasible to use alanine-scanning mutagenesis to identify potential sorting motifs present that are necessary for endosome-to-TGN retrieval. Therefore 10 mutants were generated in which successive amino acid triplets were mutated to alanine. As shown in Fig. 5A, the mutants cover the length of the 19-55 region with the exception of the YXXΦ motif, which was deliberately avoided as this motif would likely be necessary for the efficient endocytosis of the reporter constructs. The stability of the 10 different mutants was assessed using the cycloheximide chase protocol (Fig. 5B). Only mutant 7 (comprising amino acids Trp-Leu-Met) was found to be significantly unstable when compared to the Tail19-55 reporter. This construct was also predominately present in the lower molecular mass form.

The antibody uptake assay was used to determine the extent of endosome-to-TGN retrieval for all 10 mutants (see Fig. 6A). Consistent with mutant 7 being unstable and rapidly degraded (Fig. 5B), mutant 7 is unable to retrieve to the TGN. All the other alanine-scanning mutants were able to retrieve and colocalise with TGN46. When the localisation of endocytosed anti-CD8 in mutant 7 is compared with VPS26, there is again extensive colocalisation (Fig. 6B) as was seen for the Δ1-75 construct in Fig. 4A.

The mutation of the Trp-Leu-Met amino acids to alanines suggests that these three amino acids are necessary for the endosome-to-TGN retrieval of the CD8-CIMPR-reporter construct. It was, however, possible that only one or two of the amino acid triplet (Trp-Leu-Met) was necessary, and so each of the amino acids was mutated to alanine individually. Fig. S1 in supplementary material, shows cells expressing the W42A, L43A and M44A mutant analysed by antibody uptake. Mutating the Trp-Leu-Met triplet individually to alanines produces much more heterogeneous results. All three of the mutants are compromised in their ability to retrieve to the TGN, the L-A mutant being most severely blocked although this mutant also appeared to be extensively localised to the plasma membrane. The W-A mutant and the M-A mutant could both partially retrieve to the TGN to colocalise with TGN46 (indicated by the arrows). None of the individual mutants produced labelling patterns identical to that seen for mutant 7 (in which Trp-Leu-Met is mutated to Ala-Ala-Ala). The stability of the individual mutants was examined by cycloheximide chase (see Fig. S1B in supplementary material). Only the W-A mutant is as unstable as mutant 7.

The CIMPR is not the only known cargo of the retromer complex. Sortilin, a membrane protein that is homologous to the yeast vacuolar protein sorting receptor Vps10p, is also trafficked from endosomes to the TGN in a retromer-dependent manner (Seaman, 2004). Therefore the sequences of the cytoplasmic tails of sortilin proteins from various species were aligned to identify conserved regions that could contain putative retromer sorting motifs (Fig. 7A). It was noticed that there is a sequence, Phe-Leu-Val, that is biochemically very similar to the Trp-Leu-Met sequence in the CIMPR tail. Therefore, the Phe-Leu-Val sequence in a CD8-sortilin reporter was mutated to alanines. Additionally another CD8-CIMPR construct (WLM-AAA) was generated in which the Trp-Leu-Met was mutated to alanines in an almost full-length CIMPR tail. In the WLM-AAA mutant, amino acids 1-19 of the CIMPR tail are absent. These have been shown to be unnecessary for endosome-to-TGN retrieval (Fig. 3). Cells expressing CD8-CIMPR, CD8-WLM-AAA, CD8-sortilin and CD8-FLV-AAA were analysed by the antibody uptake experiment (Fig. 7B). The mutation of the Trp-Leu-Met triplet to Ala-Ala-Ala in the context of an almost full-length CIMPR tail was sufficient to prevent the endosome-to-TGN retrieval of the reporter protein. Similarly, mutation of the Phe-Leu-Val to Ala-Ala-Ala in the CD8-sortilin reporter prevented its retrieval.

The stability of the constructs shown in Fig. 7B was determined by cycloheximide chase (Fig. 7C). The CD8-WLM-AAA mutant and the CD8-FLV-AAA mutant are both unstable relative to the CD8-CIMPR and CD8-sortilin reporters, respectively. During these (and other) cycloheximide chase experiments it was noticed that the CD8-reporter constructs often migrated as a doublet on SDS-PAGE, which would chase to the upper band over time. The lumenal domain of CD8 has no N-linked glycosylation sites but is known to be O-glycosylated (Carmela-Pascale et al., 1992). Therefore, one possibility is that the lower migrating band seen in the cycloheximide chase experiments is the non-O-glycosylated form of the CD8 reporter. As O-glycosylation occurs in the Golgi complex (Van den Steen et al., 1998), absence of O-glycosylation might indicate that the CD8 reporter is retained in the endoplasmic reticulum (ER).

This was investigated by an antibody-uptake experiment performed in cells grown in dishes. After uptake of the antibody, the cells were washed, and lysed before the endocytosed anti-CD8 mAb was captured on protein-A-Sepharose. In Fig. S2A in the supplementary material, it can be seen that the CD8-CIMPR reporter that is present in post-Golgi (PG) compartments or at the cell surface (CS) migrates as a doublet in SDS-PAGE as does the total CD8 reporter present (lane T). The same is true for the CD8-WLM-AAA reporter. This demonstrates that the lower migrating form of the CD8 reporters is not retained in the ER as it is accessible to antibodies added to the media and therefore must have been present at the cell surface at some point during the experiment. As not all of the reporter construct will traffic to the cell surface during the course of the experiment, it is reasonable to expect the signal level for the post-Golgi and cell surface fractions of the CD8 reporter will be significantly less than the total CD8 reporter present in the cell at time of lysis. Additionally, cells expressing either CD8-CIMPR or CD8-WLM-AAA were fixed and labelled with the anti-CD8 mAb. The CD8 reporters exhibit staining patterns consistent with a post-Golgi localisation (Fig. S2B in supplementary material). There is no labelling of the nuclear membrane which is indicative of ER localisation.

If the Phe-Leu-Val motif is directing CD8-sortilin to be retrieved by the same machinery as the CIMPR tail, then it should be possible to replace the Trp-Leu-Met sequence in the CIMPR tail with Phe-Leu-Val. This construct (WLM-FLV) was generated in the context of the Tail19-55 reporter which is sufficient for retrieval (see Fig. 3). Cells expressing the WLM-FLV mutant were analysed by antibody uptake experiment. Fig. 8A shows that the WLM-FLV reporter is able to retrieve to the TGN and colocalise with TGN46 whereas mutant 7, in which the Trp-Leu-Met motif is mutated to Ala-Ala-Ala is not. Additionally the WLM-FLV reporter displays a similar stability to the Tail19-55 reporter and is not unstable like mutant 7 (Fig. 8B). This demonstrates that the Phe-Leu-Val motif in the sortilin tail can replace the Trp-Leu-Met motif in the CIMPR tail. Similarly, the Trp-Leu-Met motif in the CIMPR tail can replace the Phe-Leu-Val motif in the sortilin tail and results in a reporter construct with an identical localisation to the CD8-sortilin with wild-type tail (data not shown).

The Trp-Leu-Met motif in the CIMPR tail is necessary for the retrieval of the CD8-CIMPR and therefore is a good candidate for a retromer-interacting motif. We have examined the ability of retromer to interact with the different reporter constructs in vivo by native immunoprecipitation (IP) using the anti-CD8 mAb. Lysates prepared from the CD8-reporter-expressing cells were incubated with anti-CD8 that had been pre-bound to protein-A-Sepharose. The presence of both retromer and the clathrin adaptor AP-1 was determined by western blotting. Fig. 9A shows that the CD8-CIMPR, the Δ75-163, the Δ56-143 and (to a lesser extent) the Tail19-55 reporter were all able to associate with AP-1. The retromer components VPS35 and VPS26 were found to associate with the CD8-CIMPR and the Δ56-143 reporters and VPS26 was weakly associated with the Δ75-163 reporter. The CD8-CIMPR co-immunoprecipitated AP-1, and also VPS35 and VPS26, whereas the CD8-WLM-AAA mutant did not (Fig. 9B), even though comparable amounts of the CD8 reporter was immunoprecipitated. CD8-sortilin was only weakly associated with AP-1 and no retromer components were detectable. The co-IP of retromer with the CD8-CIMPR reporters is therefore dependent upon an intact Trp-Leu-Met motif. Reporters in which Trp-Leu-Met was deleted or mutated did not associate with retromer.

It is interesting that the association of retromer with the CIMPR essentially mirrors that of the clathrin adaptor AP-1. To determine if retromer and AP-1 were acting together, the distribution of the two complexes was investigated by immunofluorescence. Cells expressing CD8-CIMPR were labelled with antibodies against VPS26 and γ-adaptin. The labelling patterns for these two proteins are strikingly different with no obvious colocalisation observed (Fig. 9C). Additionally, cell lysates were immunoprecipitated with antibodies against either γ-adaptin or VPS26. The immunoprecipitates were analysed by western blotting. Antibodies against γ-adaptin did not immunoprecipitate VPS26 and vice versa (data not shown).

The lack of colocalisation of retromer and AP-1 is consistent with the two complexes functioning at different sorting steps, but in order to establish that retromer and AP-1 do not function in the same endosome-to-TGN pathway, an siRNA knockdown (KD) of retromer and AP-1 was performed. Lysates from control cells, VPS26 KD or μ1A KD were western blotted with antibodies against VPS26, VPS35, SNX1, the β1 subunit of AP-1, the transferrin receptor (TfnR) and actin (Fig. 10A). Loss of VPS26 results in instability of VPS35 and has been reported previously (Seaman, 2004; Arighi et al., 2004) whereas SNX1 remains stable. The μ1A subunit interacts with β1 (Page et al., 1999; Collins et al., 2002) therefore loss of β1 resulting from the μ1A KD is consistent with ablation of AP-1 function. There are no significant differences in the TfnR or actin levels in the lysates demonstrating the specificity of the siRNA knockdowns and also serving as loading controls.

When endosome-to-TGN retrieval was analysed in control and KD cells by antibody-uptake assay, the VPS26 KD resulted in a block in endosome-to-TGN retrieval (Fig. 10B) with the anti-CD8 antibody accumulating in a SNX1-positive endosome instead (Fig. 10C). Loss of AP-1 function due to μ1A KD did not prevent the endosome-to-TGN retrieval of the CD8-CIMPR reporter, which was able to colocalise with TGN46 and did not cause the CD8-CIMPR reporter to accumulate in endosomes. Loss of retromer or AP-1 did, however, produce one similar phenotype, as many VPS26 KD or μ1A KD cells became enlarged and multinucleated.

In this study, we have systematically dissected the cytoplasmic tail of the CIMPR in order to identify the sorting motif necessary for retromer-mediated endosome-to-TGN retrieval. The use of the CD8-reporter protein system avoids two potential pitfalls of using the full-length CIMPR for these studies. Firstly, the influence that the very large lumenal domain has on trafficking (Dintzis and Pfeffer, 1990; Dintzis et al., 1994; Conibear and Pearse, 1994) will be negated by using CD8 reporters as the lumenal domain of CD8 is small (∼20 kDa) and has no known targeting information. Secondly, as the CD8 transmembrane domain was present in all of the reporter constructs, any effects of the CIMPR transmembrane domain upon its localisation were avoided. This is significant, as for some TGN resident proteins (e.g. TGN38) the transmembrane domain contributes to the steady-state localisation (Ponnambalam et al., 1994).

The use of the CD8-reporter proteins in conjunction with the antibody-uptake assay is a direct method to assess the ability of the CD8 reporter to undergo endosome-to-TGN retrieval. There is no fluid-phase endocytosis of the anti-CD8 mAb (see Fig. 1), therefore the endocytosed anti-CD8 mAb accurately reflects the trafficking of the CD8 reporter in post-Golgi compartments. As the lumenal and transmembrane domains of all the constructs studied are identical, any differences in the trafficking of the various constructs can only be down to the engineered changes in the cytoplasmic tail. The ability of the CD8 reporters to be retrieved to the TGN is therefore a reflection of the interactions between the tail and sorting machinery such as retromer.

By truncating large regions of the CIMPR tail, the portion of the tail that is necessary for endosome-to-TGN retrieval was narrowed down to amino acids 19-55 (Fig. 3). This is consistent with studies of a lysozyme-CIMPR chimera in which TGN localisation was specified by the membrane-proximal third of the CIMPR tail (Conibear and Pearse, 1994).

The Δ1-75 construct which fails to retrieve is still efficiently endocytosed and colocalises with VPS26 (Fig. 4A). This is similar to the results of a retromer knockdown by siRNA in which the CD8-CIMPR does not efficiently retrieve but instead accumulates in a SNX1-positive endosome (Seaman, 2004). Loss of retromer function has also been shown to result in the rapid degradation of the CIMPR (Seaman, 2004; Arighi et al., 2004). The Δ1-75 reporter is very unstable, unlike the CD8-CIMPR, Δ75-163 and Δ56-143 constructs (Fig. 4B).

The Tail19-55 reporter was systematically analysed by site-directed mutagenesis, mutating three adjacent amino acids simultaneously to alanines (Fig. 5A). Mutant 7 in which the amino acids Trp-Leu-Met were mutated to alanines is both unstable and unable to retrieve to the TGN, and like the Δ1-75 mutant, accumulated in VPS26-positive structures (Fig. 5B, Fig. 6A,B). Mutation of the other residues in the Tail19-55 reporter had no effect upon the retrieval of the reporter to the TGN, strongly implicating the Trp-Leu-Met sequence in directing retrieval of the Tail19-55 reporter.

Mutation of the Trp-Leu-Met residues individually to alanine produced more heterogeneous results (Fig. S1 in supplementary material). The Trp-Ala mutant was the most unstable but in some cells was still able to partially retrieve to the TGN. The Leu-Ala mutant was stable but could not retrieve to the TGN and was localised much more to the cell surface suggesting that it could not efficiently endocytose. The improved stability of the Leu-Ala mutant relative to the Trp-Ala mutant could be due to the inability of the Leu-Ala mutant to endocytose efficiently as trafficking to lysosomes for degradation is likely to require endocytosis.

Results obtained from the mutation of the Trp-Leu-Met residues individually to alanine suggests that the three amino acids are likely to act in concert to direct the retrieval of the Tail19-55 reporter. The question remained, however, whether the Trp-Leu-Met sequence was necessary in the context of a full-length CIMPR tail. When an almost full-length CD8-CIMPR was mutated so that the Trp-Leu-Met sequence was converted to alanines, this construct (WLM-AAA) behaved identically to mutant 7, being both unstable and unable to retrieve. The cytoplasmic tail of sortilin (which is also dependent upon retromer for endosome-to-TGN retrieval) has a sequence Phe-Leu-Val, which is biochemically very similar to the Trp-Leu-Met sequence in the CIMPR tail (Fig. 7A,B). Mutation of the Phe-Leu-Val to alanines (CD8-FLV-AAA) results in the CD8 reporter failing to retrieve to the TGN and becoming very unstable (Fig. 7B and C). The fact that the CIMPR and sortilin both have a conserved sequence (Trp/Phe-Leu-Met/Val) is highly significant because the CIMPR and sortilin also have identical acidic di-leucine motifs that can serve as substrates for the VPS29 subunit of retromer (Damen et al., 2006) and can both promote recruitment of GGA2 to the TGN membrane (Hirst et al., 2007). The replacement of the Trp-Leu-Met sequence in the tail19-55 reporter with Phe-Leu-Val resulted in a construct (WLM-FLV) that was stable and could retrieve to the TGN (Fig. 8). Likewise, swapping the Phe-Leu-Val motif in sortilin for Trp-Leu-Met generated a reporter that displayed an identical localisation to the CD8-sortilin construct (data not shown). Therefore it seems likely that these two amino acid triplets comprise a novel endosome-to-TGN retrieval motif with the consensus Trp/Phe-Leu-Met/Val.

Through the stability experiments it became clear that the CD8 reporter migrated on SDS-PAGE gels as a doublet. As CD8 is O-glycosylated (Carmela-Pascale et al., 1992), the lower migrating form could represent a non-glycosylated protein that is retained in the ER. This does not, however, appear to be the case, as the lower migrating form of both CD8-CIMPR and CD8-WLM-AAA can be bound by exogenously added anti-CD8 mAbs (see Fig. S2 in supplementary material). The lower-migrating form of these two constructs may therefore be CD8 reporters that escaped O-glycosylation during passage through the Golgi. With time, the CD8-CIMPR will cycle repeatedly between endosome and TGN/Golgi and so chases to the upper band through acquisition of O-glycosylation whereas the CD8-WLM-AAA does not chase to the upper band as it cannot be retrieved to the TGN/Golgi. In this scenario, the ability to chase to upper band is a secondary indication that the respective CD8 reporter can be retrieved to the TGN/Golgi. It seems an unlikely coincidence that the constructs that migrate as the predominately lower molecular mass form (e.g. mutant 7, WLM-AAA and FLV-AAA) are also unable to retrieve to the TGN.

Does the Trp-Leu-Met motif specify interaction with retromer? From native IP, the CD8-CIMPR reporter can be seen to interact with both retromer and the clathrin adaptor AP-1. The Δ75-163 and Δ56-143 constructs (which can retrieve to the TGN) can also interact with retromer and AP-1 although it is of note that the Δ75-163 construct could weakly co-immunoprecipitate VPS26 but not VPS35. The WLM-AAA construct (which is almost full length) did not co-immunoprecipitate either retromer or AP-1 (Fig. 9A,B). Do retromer and AP-1 therefore co-operate in CIMPR retrieval? The data from the native IPs suggests this and there is evidence that places AP-1 in the endosome-to-TGN retrieval pathway (Meyer et al., 2000). Furthermore, the Trp-Leu-Met motif has also been shown to be part of an unconventional `di-leucine' signal (ETEWLM) which can bind AP-1 in vitro (Ghosh and Kornfeld, 2004), but it is worth remembering that mutation of the ETE triplet in the context of the Tail19-55 reporter (i.e. mutant 6) had no effect on retrieval (see Figs 5 and 6).

If retromer and AP-1 were functioning together in endosome-to-TGN retrieval one would predict that retromer and AP-1 should colocalise or co-immunoprecipitate and/or should have similar phenotypes after siRNA knockdown of the respective complex. However, as shown in Fig. 9C, retromer and AP-1 do not colocalise. Secondly, the siRNA knockdown (KD) experiment in Fig. 10 shows that loss of retromer function after VPS26 KD results in a block in the endosome-to-TGN retrieval of the CD8-CIMPR whereas loss of AP-1 (after μ1A KD) does not to block endosome-to-TGN retrieval of CD8-CIMPR.

It also seems highly unlikely that two large complexes such as AP-1 and retromer could both bind to essentially the same region of the CIMPR tail simultaneously, therefore if both retromer and AP-1 function in retrieval of the CIMPR then it must be via parallel pathways.

Finally, the Phe-Trp-Leu (FLV) sequence in the sortilin tail does not conform to an unconventional acidic `di-leucine' motif and therefore could not be involved in binding AP-1, but the FLV motif is necessary for endosome-to-Golgi retrieval of sortilin. For these reasons, we therefore favour a model in which the WLM motif in the CIMPR is necessary for retromer-mediated endosome-to-TGN retrieval. The interaction with AP-1 may occur primarily in the TGN, therefore mutation of the WLM motif will abolish interaction with AP-1 as there will be little or no CIMPR in the TGN to interact with AP-1.

Yeast two hybrid data has shown that VPS35 interacts with two distinct regions of the CIMPR tail encompassing amino acids 48-80 and 80-100 (Arighi et al., 2004). Curiously, however, these regions are largely absent from the Δ56-143 mutant that can co-immunoprecipitate VPS35 and VPS26. This truncation does, however, retain the SDEDLL motif which can be a substrate for VPS29 (Damen et al., 2006) suggesting perhaps that the retromer complex associates with the CIMPR tail through bipartite interactions of both VPS35 and VPS29. However, the interaction between the CIMPR tail and VPS35 must be critical for retrieval as the WLM-AAA construct has the SDEDLL motif but does not associate with retromer and does not retrieve. It is also noteworthy that the region(s) of the CIMPR tail that binds to TIP47 or PACS-1 (see Fig. 2A) are not essential for the endosome-to-TGN retrieval of the respective CD8 reporter (see Fig. 3). That does not rule out a role for TIP47 in endosome-to-TGN retrieval as TIP47 could be interacting with other endogenous proteins such as the CDMPR to perform its function in endosome-to-TGN retrieval (Diaz and Pfeffer, 1998). PACS-1 has been shown to associate with AP-1 and therefore may be more important in an AP-1-mediated sorting/trafficking step (Folsch et al., 2001).

There is much yet to be learned regarding the interaction between retromer and cargo proteins. Data presented here have identified a conserved motif essential for the retrieval of the CIMPR and sortilin and necessary for the interaction with retromer and, therefore, provides a key piece to the puzzle of understanding how retromer functions in endosome-to-TGN retrieval.

Most reagents were obtained from Sigma-Aldrich (Poole, Dorset, UK) with the following exceptions: restriction enzymes were purchased from New England Biolabs (Hitchin, Herts, UK). Effectene was obtained from Qiagen (Crawley, West Sussex, UK), Fugene6 from Roche (Lewes, East Sussex, UK) and ¹²⁵I-protein-A from Amersham Biosciences (St Albans, Herts, UK). The anti-CD8 monoclonal antibody used for antibody uptake experiments and native immunoprecipitations was produced by a mouse hybridoma cell-line obtained from the American Type Culture Collection (ATCC number: CRL-8014) through LGC Promochem (Teddington, Middlesex, UK). The polyclonal anti-CD8 used for western blotting was purchased from Autogen Bioclear (Calne, Wiltshire, UK). Anti-TGN46 was purchased from Serotec (Oxford, UK). Anti-γ-adaptin and anti-β1-adaptin for western blotting and/or native immunoprecipitations was generously provided by M. S. Robinson (University of Cambridge, UK). The monoclonal anti-γ-adaptin used for immunofluorescence was purchased from BD Bioscience (Cowley, Oxford, UK). Fluorescently labelled secondary antibodies were obtained from Molecular Probes (Paisley, Scotland, UK). Antibodies against mammalian VPS35 and VPS26 are described by Seaman (Seaman, 2004).

The cytoplasmic tail sequences of the CIMPR from various species were aligned using the ClustalW program available at the EMBnet website (http://www.ch.embnet.org/). The sequences were aligned according to homology. The output from the ClustalW program was shaded using the BOXSHADE program also available at the EMBnet website. The CIMPR tail sequences and their accession numbers are as follows: Australian echidna (Tachyglossus aculeatus - AAL23910), platypus (Ornithorhynchus anatinus - AAF68173), chicken (Gallus gallus - NP_990301), North American opossum (Didelphis virginiana - AAL23909), mouse (Mus musculus - AAM22159), rat (Rattus norvegicus - NP_036888), rabbit (Oryctolagus cuniculus - AAK71864), cow (Bos taurus - AAL23908), human (Homo sapiens - P11717), chimpanzee (Pan troglodytes - XP_518839), ring-tailed lemur (Lemur catta - AAK71866), dog (Canis familiaris - XP_541184) and zebrafish (Danio rerio - AAT42194). The sortilin cytoplasmic tail sequences were aligned in a similar fashion. The following sortilin sequences with accession numbers were used: mouse (Mus musculus - BC034129.1), rat (Rattus norvegicus - XM_342317.2), chimpanzee (Pan troglodytes - XM_513621.1), human (Homo sapiens - CAI13180), dog (Canis familiaris - XM_537041.2), cow (Bos taurus - XM_588956.2) and zebrafish (Danio rerio - NM_213230.1).

The CD8-CIMPR and CD8-sortilin constructs are described by Seaman (Seaman, 2004). The CD8-Δ1-19ΔLL construct was generated by subcloning the region of the rat CIMPR tail described by Reaves et al. (Reaves et al., 1996) into pGEX 4T-2 (using BamHI and EcoRI). This was then cut with BamHI, blunted and then cut with NotI and cloned into CD8 in pIRESneo that had been cut with AflII, blunted and then cut with NotI. To generate the CD8-Δ75-163, CD8-Δ56-143 and CD8-Δ1-75 constructs, the transmembrane and cytoplasmic tail of the CD8-CIMPR construct was first subcloned into pBluescript (Stratagene) using EcoRV and NotI. Δ75-163 was produced by cutting with BlpI, digesting with mung bean nuclease and then cutting with NotI, blunting and religating. Δ56-143 was generated by digesting with XmaI, blunting and religating. The Δ1-75 construct was produced by cutting CD8-CIMPR in pIRESneo with AflII and blunting with mung bean nuclease and then cutting with NotI. A BlpI(blunted)-NotI fragment was cloned into the AflII-NotI cut vector.

The Tail19-55 construct was produced by using PCR to amplify the region encompassing residues 19-55 of the bovine CIMPR tail. Primers were designed to introduce AflII and EagI sites at the 5′ and 3′ ends, respectively. The PCR product was cloned using pCRblunt (Invitrogen) and then subcloned (using the AflII-EagI sites) into CD8 in pIRESneo that had been cut with AflII and NotI. The Tail19-55 construct gained four amino acids (Arg-Ile-Asp-Asn) at the 3′ end that were coded by the vector before the stop codon. The EcoRV-EagI fragment of the CD8-Tail19-55 construct was subcloned into pBluescript (Stratagene) and this was then used for subsequent site-directed mutagenesis to produce the alanine scanning mutants. Site-directed mutagenesis was performed using the Quikchange kit (Stratagene) following manufacturer's instructions. The WLM-AAA construct was produced by subcloning a SacII fragment from the CD8-CIMPR tail in pBluescript into mutant 7 in pBluescript, which had been cut with SacII. The WLM-AAA construct in pBluescript was then subcloned into CD8 in pIRESneo using the EcoRV-NotI sites. The FLV-AAA sortilin mutant was produced by site-directed mutagenesis of the sortilin-tail in the pCRblunt vector before subcloning into CD8 in pIRESneo. All constructs were sequenced to confirm the truncation and/or mutagenesis had occurred as intended.

The anti-CD8 uptake assay was performed essentially as described in Seaman (Seaman, 2004). Cells stably expressing the CD8 reporters were incubated with anti-CD8 antibodies for 3 hours continuously prior to fixation with either 4% paraformaldehyde in PBS or methanol-acetone. The anti-CD8 used was from tissue culture supernatant from the hybridoma cells, which was diluted 1:5 with the cell culture medium. After fixation, the cells were permeabilised with 0.1% Triton X-100 in PBS (as necessary) and then labelled with anti-TGN46 or in some cases anti-VPS26 or anti-SNX1 followed by appropriate fluorescently labelled secondary antibodies. Microscopy was performed as described by Seaman (Seaman, 2004). To quantitatively measure the amount of endocytosed anti-CD8 mAb, ¹²⁵I-protein-A was used instead of fluorescently labelled secondary antibodies. The bound ¹²⁵I-protein-A was released by washing cells with 1 ml 1 M NaOH, 1% SDS, which was then transferred to a tube and the ¹²⁵I measured in a γ-counter.

Cells expressing the CD8 reporters were grown to ∼80% confluency in 140-mm diameter tissue culture dishes. After decanting the medium, the cells were transferred to ice, washed with 10 ml of ice-cold PBS, and left to drain for 5 minutes. 1 ml of lysis buffer [0.1 M MES-NaOH pH 6.5, 1 mM magnesium acetate, 0.5 mM EGTA, 200 μM sodium vanadate, 1% (w/v) digitonin with protease inhibitors] was added to the dish and the cells were then scraped with a cut rubber bung and transferred into a 1.5 ml tube on ice. The lysate was cleared by centrifugation for 5 minutes at 10,000 g at 4°C and the supernatant was transferred to a fresh tube. 60 μl of a 20% slurry of protein-A-Sepharose (pre-washed with lysis buffer) was added to the cleared lysate and the tube was transferred to a rotating wheel at 4°C for 30 minutes to preclear. After preclearing, the lysate was centrifuged for 5 minutes at 10,000 g at 4°C and the supernatant was transferred to a fresh 1.5 ml tube. Anti-CD8, prebound to Sepharose (1 ml of hybridoma culture supernatant was incubated with 100 μl of protein-A-Sepharose slurry, washed in lysis buffer and then divided between two 1.5 ml tubes containing the cell lysate) was added to the lysate which was then returned to the rotating wheel at 4°C for 90 minutes. After four washes with lysis buffer, the immunoprecipitates were dried in a speed-vac and subjected to SDS-PAGE. The digitonin in the lysis buffer can aggregate during the experiment, therefore the lysis buffer was passed through a sterile 0.2 μm filter immediately before the cells were lysed and before the immunoprecipitates were washed.

Knockdowns of VPS26 and μ1A using siRNA were performed as described by Seaman (Seaman, 2004), or Hirst et al. (Hirst et al., 2003) respectively.

Cells grown in six-well tissue culture dishes were treated with 100 μg/ml of cycloheximide dissolved in ethanol (after addition to the cells the concentration of ethanol in the medium was 1%). The 0 hour time point cells were incubated with just ethanol for 6 hours. At the end of the 6-hour period, the cells were washed with ice-cold PBS and then lysed in 1 ml of PBS containing 0.5% Triton X-100 with protease inhibitors. The lysate was cleared by centrifugation for 5 minutes at 10,000 g at 4°C and then transferred to a fresh 1.5 ml tube on ice. 20 μl of anti-CD8 hybridoma culture supernatant was added to each tube that was then placed on a rotating wheel for 2 hours at 4°C. 60 μl of a 20% protein-A slurry was then added and the tubes returned to the wheel for a further 60 minutes. The immunoprecipitates were washed three times with lysis buffer and then dried in a speed-vac before being subjected to SDS-PAGE. Two wells of a six-well tissue culture dish were used per time point for each CD8 reporter analysed.

SDS-PAGE and western blotting was carried out as previously described by Reddy and Seaman (Reddy and Seaman, 2001).

This work was funded by an MRC Senior Research Fellowship awarded to M.N.J.S. We wish to thank Scottie Robinson for anti-γ- and anti-β1-adaptin antisera and J. Paul Luzio for the rat CIMPR tail in pGEX construct. Thanks also to Scottie Robinson, Jenny Hirst, David Owen, Suzanne Gokool and Claire Skinner for useful discussions and comments on the manuscript.