Short transmembrane domains target type II proteins to the Golgi apparatus and type I proteins to the endoplasmic reticulum

Transmembrane proteins destined to the secretory pathway, the cell surface or endosomal compartments are inserted in the endoplasmic reticulum (ER) during their synthesis. They are then transported between the ER, the Golgi apparatus, the plasma membrane (PM) and endosomal compartments until they reach their final destination. The intracellular transport of each individual membrane protein is determined by its sequence and, notably, by sorting motifs present in its luminal domain (e.g. KDEL motifs), in its cytosolic domain [e.g. COP1-binding ER-targeting KKxx motifs or Vps74-binding Golgi-targeting (F/L)(L/I/V)XX(R/K) motifs] or in its transmembrane domain (TMD) (Dancourt and Barlowe, 2010; Eckert et al., 2014).

Sorting motifs found in TMDs have been studied extensively, and it was shown unambiguously that they can target proteins to the ER (Cosson et al., 2013). Mutational analysis established that contrary to cytosolic and luminal sorting motifs, sorting motifs in TMDs are not composed of a short consensus sequence of amino acids (AAs). Rather, a single, highly hydrophilic AA in a TMD is sufficient to ensure localization of a transmembrane protein in the ER and to prevent its transport to the cell surface (Bonifacino et al., 1991). Similarly, irrespective of their exact sequence, short TMDs (17 AA) are retained in the ER, whereas long TMDs (21 AA) are transported to the cell surface (Ronchi et al., 2008).

Very similar observations established that TMDs also play a critical role in ensuring the localization of membrane proteins in the Golgi apparatus (Cosson et al., 2013). Like ER proteins, the TMDs of Golgi proteins are shorter and contain more hydrophilic residues than surface proteins, and often contain very hydrophilic or charged residues (Sharpe et al., 2010). Mutational analysis confirmed that TMDs of Golgi proteins target fusion proteins to the Golgi apparatus, and that lengthening these TMDs or replacing hydrophilic residues with hydrophobic residues abrogated their Golgi targeting (Munro, 1991, 1995; Quiroga et al., 2013; Sousa et al., 2003). A short synthetic TMD (17 leucine residues) was sufficient to target a protein to the Golgi apparatus, whereas a long TMDs (23 leucine residues) targeted the protein to the cell surface (Munro, 1995).

Because most studies focused either on ER targeting or on Golgi targeting, it is not clear why certain TMDs are targeted to the ER and others to the Golgi apparatus. Differential targeting to the ER and to the Golgi apparatus could, in principle, be caused by subtle differences in the composition of ER and Golgi TMDs, but, over several decades of research, no specific residues or sequences have been identified that would target preferentially to the ER or to the Golgi apparatus. Another explanation would be that only type II proteins (cytosolic N-terminus) are targeted to the Golgi apparatus, as suggested by the observation that virtually all Golgi-targeted glycosylation enzymes are type II proteins (Parsons et al., 2019). However, it has been convincingly shown that type I proteins (cytosolic C-terminus) can also be targeted to the Golgi apparatus (Munro, 1995). Probably the most credible model today proposes that very short TMDs target proteins to the ER, intermediate-length TMDs target proteins to the Golgi apparatus, and long TMDs target proteins to the cell surface. In support of this model, UBC6 (17 AA TMD; cytosolic N-terminus) is localized in the ER in yeast cells. It was retargeted to the Golgi apparatus when its TMD was lengthened to 21 residues and to the PM when its TMD was lengthened to 26 residues (Yang et al., 1997). Similarly, in plant cells, chimeric LAMP1 proteins (type I) were targeted to the ER by their 17-residue TMD, to the Golgi apparatus if the TMD was lengthened to 20 residues, and to the PM by a longer TMD (23 residues) (Brandizzi et al., 2002). The experimental evidence supporting this model remains scarce. In addition, systematic analysis of the TMDs of Golgi and ER transmembrane proteins confirmed that TMDs of ER and Golgi proteins are shorter than TMDs of surface proteins, but it did not reveal differences in length between the TMDs of ER and Golgi proteins (Sharpe et al., 2010).

The aim of our study was to analyze quantitatively the ER- and Golgi-targeting potential of different TMDs in order to define more precisely the nature of transmembrane targeting motifs. Although our results qualitatively confirm all previously published observations, quantitative analysis reveals that the topology of membrane proteins is an essential and hitherto ignored feature that determines differential targeting to the ER and the Golgi apparatus.

To evaluate the influence of TMD length on localization in the secretory pathway, we first produced a series of chimeric type I transmembrane proteins composed of the extracellular domain of CD1b attached to a hydrophobic TMD of variable length (Fig. 1A), from 15 to 23 AA (Table S1), and a very short cytosolic domain (RRRSM) devoid of any known sorting motif. We then determined their localization by immunofluorescence (Fig. 1B; Fig. S1A) and quantified the localization of each CD1b chimeric protein in the ER, in the Golgi apparatus and at the PM. As detailed in the Materials and Methods, this procedure allowed us to reliably compare the localization of several chimeric proteins in different cellular compartments. As expected, proteins with long TMDs (20 to 23 AA) were mostly found at the PM (Fig. 1D–F). Proteins with shorter TMDs (15 to 19 AA) were efficiently targeted to the ER. Proteins with TMDs of intermediate length (17 to 19 AA) were detected in the Golgi apparatus, but only at low levels (Fig. 1G).

As many Golgi-targeting TMDs contain a potentially charged residue (histidine, lysine, arginine or aspartic acid), we also expressed a series of CD1b proteins with TMDs of various lengths containing an arginine residue (Fig. 1A; Table S1). The localization of each CD1b–R chimera was determined by immunofluorescence (Fig. 1C; Fig. S1B) and quantified. Long TMDs containing an arginine (22 to 24 AA) were targeted to the cell surface (Fig. 1D–F) and shorter TMDs (21 to 19 AA) were abundantly found in the ER. TMDs of intermediate length (20 to 22 AA) were also detected in the Golgi apparatus (Fig. 1E). Targeting to the Golgi apparatus was more efficient for CD1b–R22 than that described above for CD1b–17 but remained limited. Similar observations were made when a hydrophilic but uncharged glutamine was inserted in TMDs (Table S1): the strongest Golgi targeting was observed for TMDs of intermediate length (18 or 19 AA) (Fig. S1C,D).

Overall, these observations confirm that short TMDs target type I CD1b chimeric proteins to the ER, whereas long TMDs are targeted to the cell surface. The main effect of introducing a charged residue in the TMD is that the length needed to escape from the ER to the cell surface is greater (20 AA with no charged residue, 22 AA with an arginine). A TMD of intermediate length (17–19 AA without a charge, 20–22 AA with a charge) did favor localization in the Golgi apparatus. However, none of the TMDs analyzed produced the exquisitely specific Golgi localization observed for Golgi glycosylation enzymes. When the localization of all CD1b chimeric proteins was plotted on a radar plot, it was apparent that they were prominently localized either in the ER or at the PM (Fig. 1G; Fig. S2).

We next tested whether the relatively inefficient Golgi targeting of all CD1b chimeras analyzed above was due to the CD1b extracellular domain itself. For this, we analyzed proteins in which the CD1b extracellular domain was replaced with a GFP protein (Fig. 2A; Table S1). Like CD1b chimeras, a type I GFP chimera with a long hydrophobic TMD of 22 AA (GFP–22) was present at the cell surface, whereas a type I GFP chimera with a short (17 AA) TMD (GFP–17) was mostly found in the ER (Fig. 2B–D; Fig. S3). GFP–R22 was inefficiently targeted to the Golgi, but mostly detected in the ER. Overall, GFP chimeras were less observed at the PM compared with CD1b chimeras, likely due to the fact that N-glycosylation of the CD1b extracellular domain favors surface localization compared to unglycosylated GFP (Sun et al., 2020).

Another possible explanation for the relatively inefficient Golgi targeting of chimeric CD1b and GFP proteins analyzed above is that the TMDs used in this study might be less finely tuned for Golgi targeting than the real TMDs of Golgi proteins. To test this hypothesis, we analyzed the intracellular distribution of a type I GFP chimera exhibiting the TMD and cytosolic domain of an endogenous type II Golgi protein, β-1,4-galactosyltransferase 1 (GalT, encoded by B4GALT6; Fig. 2A). GalT is targeted to the Golgi apparatus by its TMD, and its short cytosolic domain does not contain a Vps74-binding sorting motif (Teasdale et al., 1992). The type I GFP–GalT chimeric protein was designed so that each residue is found at the same position in its TMD and in the TMD of type II GalT (Table S1). The type I GFP–GalT protein was mostly localized in the ER and it was absent from the Golgi (Fig. 2B–D; Fig. S3). The TMD of GalT is 20 AA long and contains a hydrophilic histidine. When the histidine residue was replaced with a leucine (GFP–GalT-v1), or when the GalT TMD was further lengthened by the addition of four leucine residues (GFP–GalT-v2), the localization of the corresponding GFP chimeras shifted from the ER to the PM, with no prominent Golgi localization (Fig. 2B–D; Fig. S3). In summary, like the CD1b proteins analyzed above, the type I GFP proteins described here were found either in the ER or at the PM, and at most poorly targeted to the Golgi apparatus (Fig. 2E).

Another possible reason for the relatively inefficient Golgi targeting of all proteins analyzed above is that type I proteins might be intrinsically poorly targeted to the Golgi apparatus. To test this hypothesis, we analyzed the Golgi targeting of a collection of GFP chimeric proteins exhibiting a type II topology. A type II GFP–GalT with the GalT TMD and a luminal GFP (Fig. 3A; Table S1) was almost exclusively detected in the Golgi apparatus (Fig. 3B–F). Mutating the histidine residue in the GalT TMD to a leucine (type II GFP–GalT-v1) increased its concentration at the cell surface, and lengthening the TMD further (type II GFP–GalT-v2) increased it even more (Fig. 3F; Fig. S4). Remarkably, the Golgi localization of these two proteins decreased compared to that of type II GFP-GalT, but remained high (Fig. 3D–F). Overall, the type II proteins analyzed here were mostly detected in the Golgi apparatus or at the PM, and poorly in the ER (Fig. 3D). All type II proteins analyzed were targeted efficiently to the Golgi apparatus, a targeting which was further improved by a short TMD exhibiting a potentially charged histidine residue.

Numerous type II proteins are found at the cell surface and absent from the Golgi apparatus, such as the human transferrin receptor (TfR, encoded by TFRC) (Trowbridge and Omary, 1981) and human glutamate carboxypeptidase 2 (Fol, encoded by FOLH1) (Murphy et al., 1998). Recent results have suggested that targeting to the Golgi apparatus is decreased by long cytosolic domains (Sun et al., 2021), as found, for example, in the full-length TfR and Fol proteins (67 AA and 19 AA versus 11 AA for GalT). Our analysis essentially confirmed these results: GFP chimeras containing the TMD of TfR or Fol associated with a short cytosolic domain were mostly targeted to the Golgi apparatus, whereas they were targeted to the PM when associated with the long TfR and Fol cytosolic domains (Fig. S5). The situation was more complex when the histidine-containing TMD of GalT was used: addition of 20 serines to the cytosolic domain of GFP–GalT (type II GFP–GalT-cyt) increased its surface expression, but this chimera maintained a significant Golgi localization (Fig. 3A–F). Golgi localization was only abolished when the TMD of GalT was made long and hydrophobic along with the addition of 20 serines to the cytosolic domain (GFP-GalT-v2-cyt) (Fig. 3C–F). Remarkably, for all the type II proteins analyzed here, no significant ER localization was observed (Fig. 3; Fig. S4). All type II GFP proteins were distributed between the Golgi apparatus and the PM as evidenced on a radar plot (Fig. 3G).

Immunofluorescence analysis does not allow us to determine the relative concentration of a protein in different compartments, notably due to the very different morphologies of the ER, Golgi and PM, and to the limited resolution of light microscopy. We used immuno-electron microscopy to determine the relative concentrations of a few proteins in different compartments (Fig. 4). As expected, a type I protein detected mostly at the cell surface by immunofluorescence (CD1b–22) was mostly detected at the cell surface by immuno-electron microscopy and sparsely found in both the ER and Golgi apparatus (Fig. 4A,E). A protein efficiently detected in the Golgi apparatus by immunofluorescence (type II GFP–GalT) was more concentrated in the Golgi membrane than in the ER, and was absent from the cell surface (Fig. 4B,E). The type I CD1b–R22 protein was detected by immunofluorescence largely at the cell surface, and to a much lower extent in the ER and Golgi apparatus. In agreement with these observations, by electron microscopy, type I CD1b–R22 appeared most concentrated at the PM, and its concentration in the ER and Golgi apparatus appeared much lower (Fig. 4C–E). Thus, quantitative electron microscopy confirmed the key results observed by immunofluorescence analysis in this study.

In this study, we analyzed how TMDs target membrane proteins to different compartments of the secretory pathway, notably the ER, Golgi apparatus and the PM. Qualitatively, our results stand in good agreement with previously published results: long hydrophobic TMDs allow access to the cell surface, and short or hydrophilic TMDs cause retention in the early secretory pathway (ER or Golgi apparatus). Type I membrane proteins are targeted (albeit inefficiently) to the Golgi apparatus by a TMD of intermediate length and hydrophilicity. Type II Golgi enzymes are targeted to the Golgi apparatus by their short or hydrophilic TMD. A large cytosolic domain increases the access of type II proteins to the cell surface.

Quantitative analysis tells a different story and stresses the dichotomy between type I and type II membrane proteins analyzed in this study. This dichotomy is particularly evident in radar plots where all proteins analyzed in this study are superimposed (Fig. 5). Depending on the length or hydrophilicity of their TMD, type I proteins with a CD1b or a GFP extracellular domain were localized either in the ER or at the cell surface, with none of them exhibiting efficient Golgi localization. On the contrary, type II proteins were present either in the Golgi or at the cell surface, with minimal ER localization. If we consider, for example, the claim that type I proteins can be targeted to the Golgi apparatus (Munro, 1995), our work qualitatively confirms this conclusion (some type I proteins with intermediate TMD length were detected in the Golgi apparatus), but quantitatively nuances it strongly (no type I protein was efficiently targeted to the Golgi apparatus).

Our results provide for the first time a comprehensive set of rules that describe how the TMD in conjunction with the size of the cytosolic domain affects (or influences) localization of a type I or type II membrane protein within the secretory pathway, and allow prediction of whether, in the absence of other sorting determinants (see below), it is present in the ER, Golgi apparatus or at the PM. Type I proteins accumulate in the ER when their TMD is short and at the cell surface when their TMD is long or hydrophobic. Type II proteins are present almost exclusively in the Golgi apparatus if their TMD is short and their cytosolic domain small. A long or hydrophobic TMD or a bulky cytosolic domain allows a fraction of type II proteins to escape to the cell surface, but the Golgi accumulation remains prominent. In our experiments, type II proteins lose Golgi localization and accumulate at the cell surface only if they combine both a long or hydrophobic TMD and a bulky cytosolic domain. Several authors have reported that other type II proteins with short cytosolic domains efficiently escape the Golgi apparatus when their TMD is lengthened (Munro, 1991, 1995; Quiroga et al., 2013; Sousa et al., 2003). It is likely that these apparently conflicting results would be reconciled in a quantitative side-by-side analysis. Alternatively, this might reflect the fact that other factors can influence this exit from the Golgi, for example, the degree of glycosylation of the luminal domain.

It remains unclear if TMDs are sorted primarily based on their biophysical properties or by interacting with yet unidentified sorting receptors (Cosson et al., 2013). To date, no candidate receptor targeting TMDs to the Golgi apparatus has been identified. Our results suggest that a hypothetical receptor recognizing and sorting TMDs to the Golgi apparatus should be able to selectively bind type II but not type I proteins with a short TMD. Transmembrane α-helices are electric dipoles, which are oriented in opposite directions in type I and type II transmembrane proteins. Electrostatic interactions between dipole moments have been shown to strongly favor association between anti-parallel transmembrane helices (Sparr et al., 2005). We speculate that this property might be sufficient to drive separation of type I and type II proteins (e.g. by association with different lipids) or to favor preferential association with a yet to be discovered sorting receptor.

Finally, it should be emphasized that this study only considered the fate of membrane proteins exhibiting a single TMD and no other sorting motif. When trying to extrapolate these rules to the sorting of actual proteins found in the secretory pathway, many other parameters need to be taken into account. First, other sorting motifs can be found in the cytosolic domain [e.g. KKxx ER-targeting motifs, (F/L)(L/I/V)XX(R/K) Golgi-targeting motifs, or DxE ER exit signals] or in the luminal domain (e.g. C-terminal luminal KDEL ER-retrieval motifs) of membrane proteins and would be expected to modify the intracellular transport. For example, in yeast, a type I protein targeted to the ER by a short TMD was relocalized to the Golgi apparatus by the addition of a DxE ER exit signal (Herzig et al., 2012). Second, proteins with multiple TMDs are abundantly found in the ER, the Golgi apparatus and at the PM, but the rules governing the sorting of multiple TMDs remain to be established. Third, many transmembrane proteins in the secretory pathway are assembled into multi-molecular complexes. For example, the invariant core of the yeast ER translocon is composed of Sec61 (10 TMDs), Sbh1 (1 TMD, cytosolic N-terminal domain) and Sss1 (1 TMD, cytosolic N-terminal domain). The trimeric Sec61 complex can associate stably or transiently with multiple other complexes, such as the Sec62–Sec63 complex (respectively two and three TMDs) (Gemmer and Forster, 2020). This example illustrates how known and unknown interactions among membrane proteins of the secretory pathway complicate the analysis of sorting determinants. Future studies will be necessary to determine if and how the sorting rules determined for relatively simple proteins can be integrated to account for the sorting of more complex entities.

HEK293T cells (American Type Culture Collection, contamination free) were grown at 37°C and 5% CO₂ in Dulbecco's modified Eagle's medium (DMEM; Gibco, 31966) supplemented with 10% fetal bovine serum (FBS; Gibco, 10270) and 100 μg/ml of penicillin-streptomycin (Gibco, 15070-063). To express chimeric proteins, cells were plated in 2.5 ml of medium on sterile glass coverslips (Menzel-Gläser, 22×22 mm) in six-well plates (Corning, 353046) 24 h before transfection. The medium was changed to serum-free DMEM (2.5 ml) before transfection. 8 µl polyethylenimine (PEI; stock 1 g/l; Polysciences, 23966-2) (Longo et al., 2013) was mixed with 63 µl serum-free DMEM, and then plasmid DNA was added (2.5 µg in 2.5 µl H₂O mixed with 63 µl serum-free DMEM). The transfection mix was incubated for 30 min then added to the cells. After an overnight incubation, the medium was changed to DMEM with 10% FBS, and the transfected cells used 3 days post transfection. For electron microscopy, cells were grown in 100 mm Petri dishes in 10 ml DMEM with 10% FBS and transfected as described above but with an eightfold larger volume of transfection mix (20 µg plasmid DNA and 64 µl PEI in a final volume of 1 ml DMEM).

CD1b chimeric proteins were composed of a signal sequence, a single-chain CD1b extracellular domain fused with the β2-microglobulin (Mercanti et al., 2010) and different TMD and cytosolic domains. This fusion protein can leave the ER without the need to associate with endogenous β2-microglobulin. CD1b proteins were ALFA-tagged (PSRLEEELRRRLTEP) on their N-terminal for electron microscopy detection (Gotzke et al., 2019; Bian and Cosson, 2020). Chimeric GFP proteins were composed of a luminal GFP fused to different TMD and cytosolic domains. The sequences of all chimeric proteins are described in Table S1. All plasmids are available upon request.

For immunofluorescence detection, we used antibodies recognizing the extracellular domain of CD1b (ABCD_AJ521; mouse Fc) (Maxit et al., 2020), human giantin (ABCD_AA341; human Fc) (Nizak et al., 2003), the ALFA tag (ABCD_AL626; mouse or rabbit Fc), GFP (ABCD_AK652; rabbit Fc), and the ER marker BiP (ABCD_AF641; rabbit Fc). Except AJ521 (produced by hybridoma cells), all antibodies were produced by the Geneva Antibody Facility (https://www.unige.ch/medecine/antibodies/) as minibodies with the antigen-binding domain (scFv or VHH) fused with the indicated Fc portion. A plasmid encoding an ER-targeted YFP–KDEL fusion protein was used to localize the ER (Perrin et al., 2018).

Immunodetection experiments were carried out at room temperature unless otherwise specified. To detect intracellular antigens, transfected HEK cells were fixed 3 days after transfection in 2.5 ml PBS containing 4% paraformaldehyde (w/v) (Applichem, A3013) for 30 min, rinsed with PBS containing 40 mM ammonium chloride (NH₄Cl) (Applichem, A3661), permeabilized in PBS containing 0.2% saponin (w/v) (Sigma-Aldrich, S7900) for 5 min, and washed once (3 min) with PBS containing 0.2% BSA (PBS-BSA) (Sigma-Aldrich, A3912). Permeabilized cells were incubated for 30 min with primary antibodies (2 mg/l). After three washes (3 min) with PBS-BSA, cells were incubated in the dark for 30 min in PBS-BSA with fluorescent secondary antibodies against mouse, human or rabbit immunoglobulins [1:400, Life Technologies; anti-mouse: Alexa Fluor 488 (A-11029), Alexa Fluor 546 (A-11030) and Alexa Fluor 647 (A-21236); anti-human: Alexa Fluor 555 (A21433) and Alexa Fluor 647 (A-21445); anti-rabbit: Alexa Fluor 546 (A-11010) and Alexa Fluor 647 (A-21245)]. Cells were then washed three times (3 min) with PBS-BSA, once with PBS, and mounted in Möwiol (Hoechst) containing 2.5% (w/v) DABCO (Fluka, 33480) (Heimer and Taylor, 1974). Pictures were taken using a Zeiss LSM800 confocal microscope, with a Plan-Apochromat 63×/1.40 oil DIC f/ELYRA oil immersion objective.

Analysis of immunofluorescence pictures was carried out using ImageJ software (https://imagej.net/). To quantify the degree of colocalization of a protein with the ER, the mean fluorescence level in three areas of the nuclear envelope (defined as the perinuclear ER–YFP- or Bip-positive region) as well as the mean fluorescence over the entire cell were quantified. The nuclear envelope/cell fluorescence ratio was then calculated for each individual cell.

To quantify the degree of colocalization of a protein with the Golgi apparatus, the mean fluorescence level in the Golgi apparatus (defined as the giantin-positive region) as well as the mean fluorescence over the entire cell were quantified. The Golgi/cell fluorescence ratio was calculated for individual cells.

To quantify the proportion of proteins present at the cell surface, unfixed transfected cells were incubated for 30 min at 4°C with primary antibodies against the CD1b or GFP extracellular domains, rinsed three times with PBS-BSA, then incubated in the dark with Alexa Fluor 546-coupled fluorescent secondary antibodies (magenta) for 30 min at 4°C. Cells were then fixed at room temperature and permeabilized, and the same primary antibody against CD1b was used, followed by an Alexa Fluor 488-coupled secondary antibody. Following this procedure, red fluorescence indicates the presence of surface CD1b, whereas green fluorescence provides an estimate of the total (surface and internal) CD1b. Total GFP fusion proteins were detected by measuring GFP fluorescence. Pictures were taken using the Plan-APO 40×/1.4 oil DIC (UV) VIS-IR oil immersion objective at constant lasers. Surface and total fluorescence intensities were quantified and the surface/total fluorescence ratio determined for each individual cell.

For ER, Golgi or PM localization, the 100% normalization value was the value obtained for the protein most abundantly detected at each of these locations (type I GFP–R22 or CD1b–R20, type II GFP–Fol and type I GFP–GalT-v2 or CD1b–23, respectively).

For every chimeric protein analyzed, at least two independent experiments were carried out and yielded very similar results. The results of both experiments were pooled in the figures shown, with a total number of cells analyzed between 15 and 80. Mean and s.e.m. of each set of data were plotted in column graphs with GraphPad Prism 8 and radar plot were computed using OriginPro 8.5 (https://www.originlab.com) (Fig. S2).

The chosen fixation and permeabilization procedures preserve best the morphology of intracellular compartments and the reactivity of antibodies used in this study. Like every immunofluorescence protocol, this staining procedure can include potential artefacts: the internal staining might be limited by incomplete permeabilization, the surface staining might be reduced by loss of surface antigens or fixation might damage antigens differentially in different compartments. The quantification procedure also differs for each compartment. Similarly, when measuring surface versus total labeling, the same antibody is used to detect both surface and total CD1b, and this might lead to underestimate the total labeling. These inevitable biases prevent a direct comparison of the signal observed in two distinct compartments (e.g. Golgi apparatus with surface). However, the use of invariant procedures allows us to reliably compare the localization of different chimeric proteins in each compartment.

Immuno-electron microscopy was performed essentially as previously described (Cosson et al., 2002; Bian and Cosson, 2020). Briefly, transfected cells were fixed in 2% paraformaldehyde and 0.2% glutaraldehyde, rinsed, detached and pelleted. The cell pellet was embedded in 12% (w/v) gelatin, infiltrated with 2.3 M sucrose (Sigma-Aldrich, 84100) and then frozen in liquid nitrogen. Frozen sections (50-nm thickness) were cut with a Leica FCS cryotome, transferred to copper grids, and incubated 1 h with anti-ALFA AL626-mouse (2 mg/l) or anti-GFP rabbit serum (1:200, Molecular Probes, A-6455). Sections were rinsed and incubated with either secondary anti-mouse Fc antibodies coupled to 10 nm gold particles (1:5, goat anti-mouse IgG, EM.GAM10, BBi Solutions) or protein A coupled to 10 nm gold particles (1:50, PAG 10 nm/S, Cell Microscopy Core, University Medical Center Utrecht) The sections were then successively negatively stained with uranyloxalate and uranyl acetate dihydrate in methylcellulose. Pictures were taken at 36,000× using a FEI Morgagni transmission electron microscope (Pôle Facultaire de Microscopie Électronique, Faculty of Medicine, University of Geneva). Immuno-electron microscopy pictures were quantified with iTEM (Olympus). Membrane lengths for ER, Golgi cisternae and PM were measured and the gold particles associated with these membranes were counted. The ratio of gold particle per micrometer of membrane was then calculated for each compartment. Labeling of the mitochondrial outer membrane was used to evaluate the level of background staining.

The Bioimaging Core Facility at the University of Geneva Medical School provided access to confocal microscopy equipment. The Geneva Antibody Facility (https://www.unige.ch/medecine/antibodies/) produced the antibodies used in this study. The Pôle Facultaire de Microscopie Électronique at the University of Geneva Medical School provided technical support and access to ultrathin cryosection and electron microscopy equipment.

The peer review history is available online at https://journals.biologists.com/jcs/lookup/doi/10.1242/jcs.261738.reviewer-comments.pdf