TM9SF4 levels determine sorting of transmembrane domains in the early secretory pathway

Membrane proteins destined to the cell surface are inserted co-translationally in the membrane of the endoplasmic reticulum (ER). They are then transported by vesicular intermediates; first to the Golgi complex, then to the cell surface. Although both the ER and the Golgi complex are traversed by a continuous flow of proteins destined to the cell surface, each compartment maintains its specific composition (Teasdale and Jackson, 1996). This observation demonstrates the existence of specific mechanisms that ensure sorting of proteins in the secretory pathway. The identification of the mechanisms ensuring recognition and sorting of proteins in the secretory pathway has followed two main directions: first, identify the sorting motifs, i.e. elements of individual proteins recognized by the sorting machinery; second, identify the cellular mechanism(s) that recognizes each motif and modulates its intracellular transport.

Because the secretory pathway is traversed by a constant flow of secreted proteins, the original proposal was that proteins devoid of specific sorting motifs were secreted ‘by default’, whereas proteins localized in the ER or in the Golgi exhibited specific retention motifs. It has since become clear that several mechanisms also assist secretory transport by concentrating cargo proteins in transport vesicles. The elucidation of the sorting mechanisms of the secretory pathway has been a collective endeavour for the past decades. Sorting motifs have been characterized in the luminal, transmembrane and cytosolic domains of membrane proteins. In transmembrane domains, sorting motifs are capable of ensuring localization in the ER, in the Golgi, or at the plasma membrane (Sharpe et al., 2010), but the cellular mechanisms ensuring the recognition and sorting of transmembrane domains (TMDs) are still poorly understood (Cosson et al., 2013).

The TM9 family of proteins is characterized by nine transmembrane domains and a high degree of conservation across species (Benghezal et al., 2003). The first extensively studied member of the TM9 family is the Dictyostelium Phg1A protein. Phg1A ensures efficient cellular adhesion (Cornillon et al., 2000), owing to its specific role in intracellular transport in the secretory pathway: Phg1A associates with the glycine-rich TMD of SibA, a cell adhesion molecule, and facilitates its transport to the cell surface in Dictyostelium cells (Froquet et al., 2012; Perrin et al., 2015). Genetic inactivation of Phg1A results in loss of cell surface SibA and a concomitant loss of cellular adhesion. Other roles of Phg1A in intracellular killing of ingested Klebsiella bacteria, and in sensing of nutrients (Froquet et al., 2008), also indirectly reflect its role in intracellular transport. For example, in phg1A knockout cells, the Kil1 protein, a sulfotransferase normally found in the Golgi complex, is mislocalized and degraded, and this accounts for a loss of efficient intracellular killing (Le Coadic et al., 2013).

TM9SF4 is a human orthologue of Phg1A. It is overexpressed in tumour cells such as aggressively metastatic melanoma cells (Lozupone et al., 2009). It is highly expressed in hematopoietic progenitor cells and its downmodulation by hypoxia is associated with a loss of cell adhesion to fibronectin, indicating that variations in the levels of TM9SF4 occur in both pathological and physiological conditions (Paolillo et al., 2015). Similar to observations in Dictyostelium cells, in mammalian cells, TM9SF4 associates with a reporter protein exhibiting a glycine-rich TMD. Overexpression of TM9SF4 increases the cell surface localization of the reporter protein, and TM9SF4 genetic inactivation decreases it (Perrin et al., 2015). These observations suggest that in mammalian cells, like in Dictyostelium cells, TM9SF4 acts mainly by controlling the intracellular sorting and transport of TMDs.

The aim of the current study was to address some of the many unanswered questions concerning the role of TM9 proteins in intracellular transport: first, it is not clear at what stage of the secretory pathway TM9SF4 influences membrane sorting; second, it is not known whether TM9SF4 recognizes a very restricted set, or a wide range, of TMDs; third, the human TM9 family of proteins counts three additional members (TM9SF1– 3), but their function has essentially not been studied. Our results suggest that TM9SF4 ensures the sorting of a wide range of TMDs in the early secretory pathway. Its activity varies in two different cell types studied [human embryonic kidney (HEK) and HCT116 cells] and this accounts for differential sorting of TMDs in these two cell types.

In order to assess the effect of TM9SF4 on intracellular sorting of TMDs, we expressed, in HEK cells, mutant proteins composed of the Tac antigen with various mutations in its TMDs (Fig. 1A; Table S1). We incorporated several TMDs that have previously been shown to cause optimal localization in the ER. Five amino acids were replaced with glycines in the TMD of T-G5, and our previous results have shown that this protein is largely retained in the ER (Perrin et al., 2015). The TMD of T-R1 presents one arginine towards its middle, and this potentially charged residue was previously shown to ensure very efficient localization in the ER (Bonifacino et al., 1991). T-KKxx contains a cytosolic KKxx ER localization motif (Cosson and Letourneur, 1994).

In order to assess the presence of each Tac mutant protein at the cell surface, transfected HEK cells were stained before and after permeabilization to reveal the presence of the Tac antigen. Individual cells were then analysed to quantify the relative abundance of the Tac protein at the cell surface. The Tac antigen with its largely hydrophobic TMD (Tac) was mostly present at the cell surface (Fig. 1B,C). As previously reported (Perrin et al., 2015), T-G5 was also detected at the cell surface, although a majority of the signal was detected intracellularly in a reticular and perinuclear compartment typical of the ER (Fig. 1D,E). T-R1 and T-KKxx were undetectable at the cell surface and they were mostly present in the ER (Fig. 1F–I), where they were colocalized with an ER-targeted GFP fusion protein (Fig. S1).

As previously described, overexpression of TM9SF4 did not affect the cellular localization of Tac, but increased significantly detection of T-G5 at the cell surface (Fig. 1D,E). On the contrary, both T-R1 and T-KKxx remained undetectable at the cell surface in cells overexpressing TM9SF4 (Fig. 1F–I). However, intracellular distribution of T-R1 appeared modified in cells overexpressing TM9SF4: in these cells, T-R1 was largely present in a juxta-nuclear compartment reminiscent of the Golgi complex (Fig. 1F), and this observation is analysed in more detail below.

In order to determine more precisely whether TM9SF4 overexpression did induce relocalization of proteins to the Golgi complex, we labelled the Golgi complex with an antibody against giantin, a well-characterized Golgi marker. A small amount of T-G5 was detected in the Golgi complex in control cells, and this amount increased in cells overexpressing TM9SF4 (Fig. 2A). T-R1 was mostly localized in the ER and only a small fraction was detected in the Golgi complex. However, upon overexpression of TM9SF4, a large fraction of T-R1 was seen in the Golgi complex (Fig. 2C). T-KKxx was exclusively localized in the ER and absent from the Golgi, even in cells overexpressing TM9SF4 (Fig. 2E). These observations were quantified and confirmed by comparing the relative intensity of the signal detected in the nuclear envelope, an easily identified domain of the endoplasmic reticulum and the Golgi complex (Fig. 2B,D,F). These results suggest that TM9SF4 modifies the intracellular sorting of TMDs at the ER-Golgi interface, favouring the relocation of ER-retained TMDs to the Golgi complex. On the contrary, proteins localized in the ER by a different type of sorting motif, such as a KKxx cytosolic motif, are not subject to sorting by TM9SF4.

These results were confirmed by the observation that overexpression of TM9SF4 significantly increased the fraction of T-R1 exhibiting sugars matured in the Golgi [39.2% in cells overexpressing TM9SF4 vs 22.9% in wild-type (WT) HEK cells; N=4 independent experiments; P<0.01, Student’s t-test] (Fig. S2).

Our previous studies showed that TM9SF4 interacts with glycine-rich TMDs (Perrin et al., 2015). To test whether TM9SF4 was able to specifically interact with a broader range of TMDs, we co-expressed TM9SF4 fused to the β-galactosidase enzyme (TM9SF4-βGal) with variants of the Tac protein exhibiting different TMDs. The Tac protein was then immunoprecipitated and the percentage of the β-galactosidase co-precipitated was determined. As previously reported (Perrin et al., 2015), T-G4 and T-G5 associated significantly with TM9SF4, compared with the background signal measured with Tac (Fig. 3A). A significant level of interaction was also detected between TM9SF4 and T-R1 by this method. In the hydrophobic lipid bilayer, an arginine is probably not charged, and it would mostly stand out as a hydrophilic residue. Indeed, the presence of a glutamine residue (uncharged but strongly hydrophilic) in the TMD of Tac (T-Q1) was sufficient to induce an interaction with TM9SF4 (Fig. 3A). The presence of a cluster of four hydrophilic threonine and serine residues (T-T3S) did not, however, induce a significant interaction (Fig. 3A), indicating that not all TMDs containing hydrophilic residues are capable of interacting efficiently with TM9SF4.

The T-G5 and T-R1 proteins analysed in this study are mainly localized in the ER (Fig. S1), whereas TM9SF4 primarily localizes in the Golgi complex (Fig. S3A,B) (Perrin et al., 2015). The interaction between TMDs and TM9SF4 could thus take place either in the ER or in the Golgi complex.

To test the ability of TMDs to interact with TM9SF4 in the ER, we exchanged the extracellular domain of TM9SF4 with the extracellular domain of the δ-chain of the T-cell receptor, a modification sufficient to confer ER retention to the resulting δ-TM9SF4 chimera (Klausner et al., 1990). As previously shown, the luminal domain of the δ-chain of the T-cell receptor does not interfere with interactions between TMDs, neither inducing nor hindering them (Cosson et al., 1991). A Flag-tagged version of this chimera (δ-TM9SF4-Flag) was indeed localized largely in the ER, although a fraction of the protein was also present in the Golgi complex (Fig. S1C,D). δ-TM9SF4-βGal interacted more efficiently with T-G4 and T-R1 than with Tac (Fig. 3B), and the levels of association observed were increased compared with the association between T-G4 and T-R1 and TM9SF4. Taken together, these results demonstrate that TM9SF4 interacts with a large set of TMDs and suggest that this interaction can take place in the ER.

In order to extend these observations to a wider range of TMDs, we also tested the effect of TM9SF4 on the localization of a TMD containing aspartate (T-D1), a potentially negatively charged amino acid residue (Fig. 1A; Table S1). As previously observed (Bonifacino et al., 1991), T-D1 was mainly localized in the ER in HEK cells (Fig. S1; Fig. 4A). However, T-D1 was largely relocated in the Golgi complex upon overexpression of TM9SF4 (Fig. 4A,B). Surprisingly, no significant interaction was detected between TM9SF4 and T-D1 (Fig. 4C). However, we did detect an interaction between the ER-localized δ-TM9SF4-βGal and T-D1 (Fig. 4D). These results indicate that TM9SF4 also controls the access of T-D1 to the Golgi complex. They reinforce the suggestion that the interaction is transient and can take place in the ER.

As TM9SF4 overexpression increases Golgi localization of T-R1, genetic inactivation of TM9SF4 would be expected to have the opposite effect: it should decrease the amount of T-R1 in the Golgi complex. This assumption is unfortunately difficult to test in HEK cells, because in these cells the amount of T-R1 detected in the Golgi complex is very low. In HEK cells devoid of TM9SF4, we did detect a significant further decrease in the Golgi localization of T-R1 compared with WT HEK cells (Fig. S4). However, the magnitude of this change was limited, and we are uncertain of its significance.

Interestingly, when expressed in HCT116 cells, T-R1 was present both in the ER and in the Golgi complex (Fig. 5). This indicates that different cell types can sort differently the same TMD. It also provided us with a situation in which the effect of a genetic inactivation of TM9SF4 could be more reliably tested. Using CRISPR/Cas9 targeting, we generated two independent TM9SF4 knockout cells in HCT116 cells (Fig. S5). In these cells, T-R1 was mostly localized in the ER (Fig. 5). These results indicate that, in HCT116 cells, endogenous TM9SF4 levels are sufficient to ensure a significant localization of T-R1 in the Golgi complex.

Previous studies have suggested that different TM9 proteins play partially redundant functions (Benghezal et al., 2003). Human TM9SF2 and TM9SF4, like Dictyostelium Phg1A, belong to the subgroup I of TM9 proteins, whereas human TM9SF1 and TM9SF3 belong to the subgroup II, along with Dictyostelium Phg1B (Benghezal et al., 2003). We chose to study TM9SF1, a member of the subgroup II of TM9 proteins. TM9SF1 exhibited a significant interaction with glycine-rich TMDs T-G4 and T-G5, as well as with T-T3S, but not with other TMDs exhibiting putatively charged residues, or glutamine (Fig. 6). This pattern of association is different from that observed with TM9SF4 and suggests that TM9SF4 and TM9SF1 proteins recognize partially overlapping, but distinct, sets of TMDs.

We next tested the ability of TM9SF1 to alter the localization of TMDs in the secretory pathway. When co-expressed with T-G5, T-R1 or T-D1, TM9SF1 failed to modify their relative localization in the ER and in the Golgi complex (Fig. 7). It also did not increase the cell-surface localization of T-G5 (Fig. S6).

In order to better delineate the function of TM9SF1, we next analysed its intracellular localization. Like TM9SF4, Flag-tagged TM9SF1 largely colocalized with giantin in the Golgi complex and was not significantly detected in the ER (Fig. 8A, white arrowheads). However, contrary to TM9SF4, TM9SF1 was also present in Golgi-proximal giantin-negative compartments that could represent endosomal compartments (Fig. 8A, black arrowheads). To test this hypothesis, we attempted to colocalize TM9SF1 with internalized fluorescent transferrin, a marker of endosomal compartments. TM9SF1, but not TM9SF4, was partially detected in transferrin-positive endosomal compartments (Fig. 8B,C).

Taken together, these results suggest that TM9SF1 does not play a critical role in intracellular transport and sorting of TMDs in the secretory pathway, but leave open the possibility that it might sort and transport a subset of TMDs between other intracellular compartments; for example, between the Golgi complex and endosomes.

Molecular mechanisms ensuring the sorting of TMDs in the secretory and endocytic pathways are still largely unknown (reviewed in Cosson et al., 2013). The two best-characterized TMD-sorting mechanisms implicate Rer1 and Erv14. Rer1 resides in the Golgi complex. It associates with some hydrophilic TMDs that have escaped the ER, and ensures their selective retrieval back to the ER. Erv14 has been proposed to associate with long hydrophobic TMDs in the ER and to escort them along the secretory pathway towards the cell surface. The current study suggests that TM9SF4 might represent a third sorting mechanisms for TMDs in the early secretory pathway. The combined action of these three different sorting systems (and possibly others) presumably determines the fate of individual TMDs in the early secretory pathway.

TM9SF4 associates with TMDs containing multiple glycine residues (T-G5), an arginine (T-R1) or an aspartic acid (T-D1). In HEK cells, TM9SF4 overexpression relocates a significant portion of these three proteins from the ER to the Golgi complex. Although, for T-G5, this relocation leads to an increase in its surface localization, for T-R1 and T-D1, no surface localization was detected in any condition. These results suggest that TM9SF4 acts primarily at the level of the early secretory pathway (ER-Golgi interface), rather than at later stages (intra-Golgi transport or Golgi-to-surface transport). TM9SF4 overexpression did not modify the ER localization of a protein exhibiting a cytosolic KKxx motif, confirming the fact that it does not affect the whole population of ER-localized transmembrane proteins.

The simplest interpretation of our observations is that TM9SF4 associates with a subset of TMDs in the ER, and ensures their incorporation in vesicles exiting the ER and destined to the Golgi complex. In agreement with this interpretation, some of our results suggest that the interaction between TMDs and TM9SF4 can take place in the ER. Indeed, relocating TM9SF4 to the ER by exchanging its luminal domain with an ER-retained CD3-δ luminal domain increased interactions with target TMDs. These results are, however, only indicative, and do not exclude the possibility that TM9SF4 interacts productively with TMDs in the Golgi complex as discussed below. It is also worth noting that TM9SF4 exhibits an LYRTL sequence in the cytosolic loop connecting TMD3 and TMD4. This is similar to the IFRTL sequence found in Erv14 that ensures its concentration into COPII-coated vesicles at ER exit sites (Powers and Barlowe, 2002).

A second interpretation of our observations is possible, and not mutually exclusive from the first one: TM9SF4 may reside in the Golgi complex, where the bulk of the protein is indeed detected. TM9SF4 could capture and retain selectively some incoming TMDs (such as T-R1), limit their return to the ER and favour their localization in the Golgi complex. Indeed, as previously observed in yeast cells (Letourneur and Cosson, 1998), T-R1 is efficiently localized in the ER by its TMD, but acquires modifications typical of the Golgi complex, suggesting that it continuously cycles between these two compartments. For some TMDs (T-G5), capture in the Golgi by TM9SF4 would increase their transport along the secretory pathway to the cell surface. For others (T-R1, T-D1), further transport might be impossible and this interaction would result in their accumulation in the Golgi complex. This suggestion is compatible with our previous observation that a proximity ligation assay suggests that TM9SF4 interacts with glycine-rich TMDs in the Golgi complex (Perrin et al., 2015). It is also supported by previous phenotypic analysis of phg1A knockout cells in Dictyostelium: in these cells, the trans-Golgi sulfotransferase Kil1 is rapidly degraded, suggesting that Phg1A plays a role in its retention in the Golgi complex (Benghezal et al., 2006; Le Coadic et al., 2013). The possibility that TM9SF4 may function as part of a Golgi-retention mechanism is intriguing: to date no protein-based mechanism ensuring retention of TMDs in the Golgi complex has been identified, and it is usually assumed that interactions with the lipid bilayer account for sorting of TMDs in the Golgi complex (Banfield, 2011). Further experiments will, however, be necessary to determine whether TM9SF4 indeed ensures selective retention of TMDs in the Golgi complex.

Although the majority of this study was performed in HEK cells, sorting of T-R1 was also analysed in HCT116 cells. Remarkably, in HCT116 cells, sorting of T-R1 is different from that observed in HEK cells: a significant fraction of T-R1 was localized to the Golgi complex in HCT116 cells, a situation similar to that observed in HEK cells overexpressing TM9SF4. In order to verify whether differences in TM9SF4 activity account for this difference, we genetically inactivated TM9SF4 in HCT116 cells. This alteration restricted T-R1 to the ER, as observed in HEK cells not overexpressing TM9SF4. These observations suggest that the sorting activity of TM9SF4 accounts for significant differences in the sorting of TMDs in these two cell types. The levels of expression of TM9SF4 can vary in response to physiological stimuli (e.g. hypoxia) (Paolillo et al., 2015), or in pathological situations (e.g. cancer cells) (Lozupone et al., 2009). Our results suggest that these variations in the levels of TM9SF4 affect sorting of TMDs in different cell types, and in different physiological or pathological situations. By modulating cell surface expression or intracellular localization of proteins, variations in intracellular sorting could result in significant variations in cellular physiology.

Finally, TM9SF1 seems to behave differently from TM9SF4: it associates with glycine-rich TMDs but not with TMDs containing charged residues, and its overexpression failed to modify the intracellular localization of any of the TMDs analysed in the secretory pathway. TM9SF1 only partially colocalized with the Golgi complex, suggesting that it could play a role at other steps of intracellular transport (e.g. between the Golgi complex and endosomes). The precise role of TM9SF1, and of other TM9 proteins, remains to be established.

HeLa and HEK 293T cells (a gift from Dr M. Foti, Geneva Faculty of Medicine, Switzerland) were grown at 37°C and 5% CO₂ in Dulbecco's modified Eagle medium (Gibco) containing 10% fetal bovine serum (Gibco) and 100 μg/ml penicillin-streptomycin (Sigma-Aldrich). HCT116 cells (a gift from Dr F. Lozupone, Istituto Superiore di Sanita, Roma, Italy) were grown in Roswell Park Memorial Institute medium supplemented with 10% fetal bovine serum and penicillin-streptomycin. TM9SF4 knockout HCT116 cells were generated using the CRISPR/Cas9 method, as previously described (Perrin et al., 2015). Two individual clones were obtained with mutations leading to a frameshift in both alleles and were used in parallel in this study.

To express Tac mutant proteins, we used a previously described pCDM8-based vector containing the coding sequence of the α chain of the interleukin-2 receptor (Tac antigen) (Cosson et al., 1991), and inserted the sequence coding for the indicated TMD (Table S1) using BglII and XbaI restriction sites. Plasmids were obtained as previously described (Perrin et al., 2015).

Cells were transfected 2 days before the experiment, using polyethylenimine as previously described (Longo et al., 2013). To analyse the intracellular localization of Tac proteins, transfected cells were fixed for 30 min in 4% paraformaldehyde, permeabilized for 10 min in PBS containing 0.2% saponin, and then incubated with the mouse anti-Tac antibody (7G7, 3 µg/ml) (Rubin et al., 1985) for 30 min in PBS containing 0.2% bovine serum albumin. When indicated, a rabbit anti-Flag antibody (Sigma-Aldrich, F7425) or a human anti-giantin antibody (Nizak et al., 2003) (3 µg/ml) was also used. Finally, cells were incubated with an Alexa-Fluor-647-coupled anti-mouse-IgG antibody (Life Technologies, A11029) (1:400) and, when indicated, with an Alexa-Fluor-488-coupled anti-human-IgG antibody (Jackson ImmunoResearch, 709-545-149) or an Alexa-Fluor-555-coupled anti-rabbit-IgG antibody (Molecular Probes, A-11035), before being mounted in Möwiol. Samples were observed using a Zeiss LSM700 confocal microscope. The relative abundance of each Tac mutant protein in the ER [nuclear envelope, normalized to 100 arbitrary units (a.u.)] and in the Golgi complex was determined in individual cells. Note that this quantification was only possible for proteins mainly restricted to the ER and Golgi complex (T-R1, T-D1, T-G5) and not for proteins escaping the early secretory pathway more efficiently (Tac, T-Q1, T-T3S, T-G4).

When indicated, the ER was labelled by expressing a luminal YFP-KDEL (ER-YFP, a kind gift from Nicolas Demaurex, University of Geneva, Switzerland) or a membrane CD1b17-GFP (the 17-residue TMD of the CD1b17 chimera ensuring its ER retention) (Mercanti et al., 2010). The Golgi complex was revealed by expressing a fusion protein of GFP with the Golgi beta-1,4-galactosyltransferase (B4GALT1-GFP) (Vernay et al., 2017).

To reveal the Tac protein present at the cell surface (Vernay and Cosson, 2013), transfected cells were incubated with the 7G7 antibody for 15 min at 4°C, then fixed, and the surface Tac protein revealed with an Alexa-Fluor-647-coupled anti-mouse-IgG antibody. Cells were then permeabilized and incubated sequentially with 7G7 and with an Alexa-Fluor-488-coupled anti-mouse IgG to reveal the intracellular Tac antigen. Surface labelling was quantified with ImageJ software (http://rsb.info.nih.gov/ij/). The surface/total ratio was calculated for each individual cell using the surface fluorescence intensity and the total fluorescence intensity (in a.u., where 100 a.u. correspond to the relative surface expression of WT Tac). In each independent experiment, at least ten cells were quantified.

In order to measure the maturation of sugars coupled to the Tac luminal domain, cells were lysed in lysis buffer [PBS containing 0.5% Triton X-100 and a cocktail of protease inhibitors (20 μg/ml leupeptin, 10 μg/ml aprotinin, 18 μg/ml phenylmethane sulfonyl fluoride and 18 μg/ml iodoacetamide)], then samples were centrifuged for 15 min at 4°C (10,000 g) and the supernatants were collected. Aliquots of cleared lysate was loaded on a nonreducing 10% acrylamide SDS-PAGE gel. Proteins were separated by electrophoresis, transferred to nitrocellulose, followed by immunodetection with 7G7 antibodies and a horseradish peroxidase-coupled anti-mouse IgG (Bio-Rad 170-6516).

Interactions between transmembrane domains were measured as previously described (Perrin et al., 2015). Briefly, HeLa cells were co-transfected to express the Tac (or a Tac mutant) protein and TM9SF4 was fused to the β-galactosidase. Two days later, cells were lysed and centrifuged as described above, and an aliquot of the cleared lysate was kept to determine the total amount of β-galactosidase activity in each sample. The Tac protein was then immunoprecipitated from the lysate using protein A-agarose beads coated with 7G7 antibodies (10 µg/immunoprecipitation). The co-precipitated β-galactosidase activity, as well as the activity present in the cell lysate, was revealed by the addition of a colorimetric substrate, Chlorophenol Red-β-D-galactopyranoside, and quantified by measuring the absorbance at 600 nm. The immunoprecipitated β-galactosidase was then calculated as a percentage of the total activity in the cellular lysates.

We thank the Bioimaging Core Facility at the University of Geneva Medical School for providing access to confocal microscopy equipment.