Molecular determinants of ER–Golgi contacts identified through a new FRET–FLIM system

Membrane contact sites (MCSs) are sites of close apposition (10–20 nm) between two distinct organelles where the exchange of solutes or lipids occurs between the facing organelles. The ER is involved in forming most (though not all) MCSs with other organelles including mitochondria, the plasma membrane (PM), endosomes, lysosomes, and the Golgi complex (Murley and Nunnari, 2016; Phillips and Voeltz, 2016). These structures have attracted considerable attention in recent years, and several tethering complexes have been identified that mediate the coupling of the apposing organelles (Eisenberg-Bord et al., 2016). The ER–TGN contact sites (ERTGoCS) are the least well defined (De Matteis and Rega, 2015), despite the fact that sites of close apposition between the ER and the last trans cisterna/trans-Golgi network (herein collectively called TGN) have been documented by ultrastructural analysis since the early 1960s (Novikoff, 1964; Rambourg et al., 1979; Ladinsky et al., 1999). This is partly due to the position of the Golgi complex in the crowded perinuclear area that has hampered the analysis of these MCSs by optical microscopy due to the resolution limits of this approach. There is thus a dearth of knowledge regarding fundamental aspects of ERTGoCS such as their frequency, regulation, and composition. These gaps in our knowledge about ERTGoCS, mainly due to methodological limits, have hampered direct studies to address this question. We therefore developed a Förster resonance energy transfer (FRET)–based approach (Grecco and Bastiaens, 2013), which is endowed with the necessary nanoscale resolution power for the study of ERTGoCS.

We initially performed a morphometric ultrastructural analysis and found that 55–60% of Golgi stacks are engaged in MCSs (defined as areas where the distance between the TGN and the ER membranes is ≤20 nm) with the ER with a distance ranging from 5 to 20 nm (average, 12 nm), a mean length of 176 nm, and a mean occupancy of 24% of the TGN surface area (Fig. 1, A–C). A 3D view of the ERTGoCS using focused ion beam–scanning electron microscopy (FIB-SEM) showed that the ER is closely apposed to the TGN occluding a surface area of up to 0.2 µm² for each organelle (Fig. 1 D and Video 1).

We next developed a FRET-based approach that involves a donor fluorophore molecule (GFP) conjugated to a TGN membrane protein (TGN46) that, when excited, transfers energy to an acceptor fluorophore molecule (mCherry) conjugated to an ER membrane protein (the C tail of cytochrome b5, Cb5) provided that the two fluorophores are within a distance of 1–10 nm (Fig. 2, A and B; and Materials and methods; Grecco and Bastiaens, 2013). In the presence of FRET phenomenon, the lifetime of the donor becomes shorter, as a consequence of the process of energy transfer to the acceptor molecule. Thus, in our assay we used FLIM of cells expressing TGN46-GFP alone or TGN46-GFP and mCherry-Cb5 at equimolar amounts (Fig. 2 A and Materials and methods). The mean fluorescence lifetime (τ, in nanoseconds) of the TGN46-GFP donor protein was measured by FLIM in defined regions of interest (ROIs; Fig. 2 C and Materials and methods). A clear decrease of the TGN46-GFP mean τ value was detected in cells expressing TGN46-GFP/mCherry-Cb5 compared with cells expressing TGN46-GFP alone or together with cytosolic, free mCherry (Fig. 2, D and E). The FRET efficiency was calculated for each pixel of the ROI, and the spatial FRET distribution was depicted using pseudocolors (Fig. 2 F). A heterogeneous FRET efficiency map was apparent with hot spots likely corresponding to a single ERTGoCS. The FRET signal was abolished by photobleaching the acceptor fluorophore while it was increased using longer helical linkers (Várnai et al., 2007) between the fluorophores and the TGN and ER membrane proteins (Fig. S1, A and B). A FRET–FLIM signal (with an average FRET efficiency of 0.2) was also measured using a different donor fluorophore molecule (Alexa Fluor 430 conjugated to anti-HA antibodies) together with the mCherry acceptor molecule in cells expressing TGN46-HA/mCherry-Cb5 (Fig. S1, C–F).

We tested the FRET-based assay under conditions that have been hypothesized to destabilize ERTGoCS or to artificially induce them (Peretti et al., 2008; Mesmin et al., 2013). To destabilize ERTGoCS, cells were depleted of the VAP proteins (VAPA and VAPB), ER integral membrane proteins that are involved in different classes of contact sites (Murphy and Levine, 2016), by siRNA treatment or by transcription activator-like effector nuclease (TALEN; Dong et al., 2016). Ultrastructural analyses showed the disruption of almost all ERTGoCS, with <10% of the Golgi stacks in VAP-KD cells having an MCS with the ER (Fig. 3 A). Concordantly, the FLIM analysis showed a decreased FRET efficiency of the ER/TGN pair of reporter proteins (with an increase in lifetime of the donor) in VAP-KD and VAP-KO cells as compared with control cells (Fig. 3, B and C; and Fig. S1, E and F). By contrast, the FRET efficiency of the ER/TGN pair of reporter proteins increased significantly (and the lifetime value of the donor decreased) under two independent conditions that stabilize ERTGoCS (Fig. 3, B, D, and E). The first involved the expression of the recombinant protein PH-FFAT (Mesmin et al., 2013) that can bridge the TGN and the ER since it contains a Golgi targeting motif, represented by a PH domain that binds PI4P (the phosphoinositide enriched at the TGN), and an ER targeting motif represented by a FFAT motif that binds the VAP proteins in the ER (Murphy and Levine, 2016; Fig. 3, B and E). The second was the chemically induced dimerization of the TGN46 and Cb5 reporter proteins fused with FRB and FKB domains (Choi et al., 1996) by short treatment with low concentrations of rapamycin (200 nM for 2 min; Fig. 2, A and B; and Fig. 3, B and D).

Having validated our FRET–FLIM strategy and with a suitable positive control (i.e., VAP depletion), we performed an siRNA-based search for components of ERTGoCS in cells expressing TGN46-GFP and mCherry-Cb5. We explored the role of proteins that have been proposed as candidate tethering factors between the ER and the Golgi since they possess dual targeting motifs for the two organelles, i.e., a FFAT (FFAT-like motif or putative FFAT motif) and a PI4P-binding PH domain. These proteins included OSBP1 and members of the oxysterol-binding protein-related protein (ORP) family that localize at the Golgi complex (ORP9, ORP10, and ORP11); the ceramide transfer protein CERT; and the FAPP proteins, FAPP1 and FAPP2 (Hanada et al., 2003; D’Angelo et al., 2007; De Matteis et al., 2007; Ngo and Ridgway, 2009; Mikitova and Levine, 2012; Nissilä et al., 2012; Weber-Boyvat et al., 2015; Fig. 4 A). Of note, all the tested proteins, except FAPP1, also possess a lipid transfer domain with a distinct lipid specificity. To our surprise, only one (ORP10) out of the entire set of candidates tested affected the stability of the ERTGoCS when depleted. In fact, the mean lifetime of TGN46-GFP in ORP10-depleted cells expressing TGN46-GFP and mCherry-Cb5 was not significantly different from that measured in cells expressing TGN46-GFP alone, and consequently, the FRET efficiency was extremely low in ORP10-depleted cells as compared with control cells (Fig. 4, B and C). We then combined depletion of two candidates at a time, driven by the hypothesis that the depletion of one component might be compensated by another protein with similar activity (Eisenberg-Bord et al., 2016). We found that except for the depletions involving ORP10, only the combined depletion of OSBP1 and ORP9 was able to significantly affect ERTGoCS stability (Fig. 4, B and D). Importantly, the results obtained by FRET–FLIM measurements were validated by ultrastructural and stereological analyses that showed that the only conditions that affected the ERTGoCS organization were ORP10 depletion and the combined depletion of OSBP1 and ORP9 (Fig. 4, E and F; and Fig. S2, A and B).

We next investigated the molecular mechanisms underlying the requirement of the identified hits in sustaining the ERTGoCS and in particular whether this was dependent on their lipid transfer activity. ORP10 associates with microtubules but also with the Golgi complex (Fig. 5 A; Nissilä et al., 2012), and while not possessing a canonical FFAT domain, it can interact with VAPs (Weber-Boyvat et al., 2015; Murphy and Levine, 2016). ORP10 belongs to the ORP subfamily that includes members that can transfer PS (Maeda et al., 2013), such as ORP5 and ORP8 that transfer PS in exchange for PI4P between the ER and the PM (Chung et al., 2015), while ORP10 itself has been shown to bind and extract PS (Maeda et al., 2013). We observed that the residues of the oxysterol-related domain (ORD) of ORP5/ORP8 that are involved in PS/PI4P exchange (Chung et al., 2015) are very well conserved in ORP10 (Fig. 5 B). To investigate whether the requirement for ORP10 in maintaining the ERTGoCS is due to a structural role or to its lipid transfer activity, siRNA-resistant ORP10 was reexpressed in ORP10-depleted cells, both in its WT form and in forms where the ORD sites involved in PS (L418D) and PI4P (H535/536A) binding had been mutated (Fig. 5 B). We found that the WT but not the mutant forms of ORP10, which were correctly targeted to the Golgi complex (Fig. 5 C), could reestablish ERTGoCS in ORP10-depleted cells (Fig. 5 D and Fig. S3, A and B), indicating that the lipid transfer activity of ORP10 is required for the stability of these MCSs. PS is synthesized in the ER but is enriched in the PM, the TGN, and endosomes (Leventis and Grinstein, 2010; Fairn et al., 2011), where it is asymmetrically distributed between the bilayer leaflets, being confined to the cytosolic leaflet. Thus, we tested the possibility that ORP10 might be involved in transferring PS from the ER to the cytosolic leaflet of TGN membranes by studying the effect of ORP10 depletion on the subcellular distribution of PS using two distinct genetically encoded probes: the PS binding domain of lactadherin (C2 domain; Yeung et al., 2008; Fig. 5 E) and of evectin (PH domain; Uchida et al., 2011; Fig. S3 C). We found that ORP10 depletion induced a decrease in PS at the Golgi as evaluated by the distribution of the two PS probes (Fig. 5 E and Fig. S3 C), which showed a marked reduction of the Golgi pool (but not of the PM pool) in ORP10-KD cells, and by the observation that a lower amount of TGN membranes from ORP10-KD cells was retained on PS-affinity beads as compared with control cells (Fig. S3 D).

Thus, ORP10 is required to safeguard the PS content in the TGN and, at the same time, the ERTGoCS (Fig. 4, A–F; Fig. 5 E; and Fig. S3, C and D), indicating that the lipid composition of the TGN, and in particular its PS content, is of crucial importance for the establishment/maintenance of these structures. Supporting this hypothesis, exogenous supplementation of PS partially restored ERTGoCS in ORP10-depleted cells (Fig. 5 F). Similar results were obtained when the production of PS was boosted through the overexpression of a mutant (P269S) form of PS synthase 1 that is insensitive to product inhibition (Sousa et al., 2014; Sohn et al., 2016; Fig. 5 G and Fig. S3 E).

We addressed the same question (i.e., a structural versus a lipid transfer activity-dependent role in maintaining the ERTGoCS) for OSBP1, which was found to be an obligate partner of ORP9 in maintaining ERTGoCS. To this end, we studied the effect of itraconazole, an inhibitor of OSBP1 lipid exchange activity (Strating et al., 2015), on the stability of the ERTGoCS and found that it had no inhibitory effect either in control cells or in ORP9-KD cells (Fig. 5 H).

Here, we have reported the development of a FRET–FLIM approach to visualize the ERTGoCS in their native form (i.e., in the absence of contact stabilizers) based on the use of a GFP-fused TGN reporter and a cherry-fused ER reporter. While FRET measurements have been previously applied to the study of ER–mitochondria contact sites, they were always under conditions where they had been stabilized by rapamycin-induced dimerization of the organelle reporters (Csordás et al., 2010; Naon et al., 2016).

By coupling FRET–FLIM and EM approaches, we performed the first systematic study aimed at identifying molecular determinants of ERTGoCS. Each approach has complementary pros and cons. The FRET–FLIM approach enables the exploration of multiple conditions, but it provides information that is averaged over the Golgi area without reaching single stack resolution. This is achieved by EM that is endowed with a very limited throughput but provides information on the distribution of ERTGoCS across the different stacks, on their extension and “tightness.” Indeed, by applying rather strict criteria to the definition of ERTGoCS (i.e., sites where the TGN and ER membranes come closer than 20 nm) we show that the distribution of ERTGoCS across the Golgi stacks is not homogeneous, as only a fraction (55%) of Golgi stacks is engaged in ERTGoCS, and the TGN area apposed to the ER in these stacks accounts on average for 24% of the TGN surface. Since the EM provides a snapshot, we cannot presently distinguish between two possible scenarios that could account for this distribution: either all the stacks have the same likelihood to engage in tight contacts with the ER, and at any given time, only 55% exhibit ERTGoCS, or there is only a subset of stacks that has the ability to contact the ER, thus generating heterogeneity among the different Golgi stacks with regard to their closeness to the ER. Additionally, there is also heterogeneity in the width of ERTGoCS, which ranges from 5 to 20 nm, with 40% of them spanning <10 nm.

We have shown that the ERTGoCS require both structural tethers and a proper lipid composition of the TGN. We identified the ER VAP proteins as major determinants of the stability of ERTGoCS since their depletion or ablation (by siRNA or by TALEN) abolished the FRET signal and eliminate the ERTGoCS as evaluated by EM.

The profound destabilizing effect of VAP depletion/ablation can be explained by the role of VAPs as a common ER anchor for the different members of the family of lipid transfer proteins acting at the ERTGoCS. These include OSBP1, ORP9, and ORP10. We have shown that OSBP1 and ORP9, while “using” the ERTGoCS for lipid exchange, can also act as tethers. The higher cellular levels of both OSBP1 and ORP9 (244 and 48 nM, respectively; Hein et al., 2015) and their higher affinity for VAPs (Mikitova and Levine, 2012; Murphy and Levine, 2016) and PI4P (Levine and Munro, 2002) relative to other members of the family may explain why they can play a tethering role. This is also in line with previous reports showing that OSBP1 can tether liposomes in vitro and, when overexpressed (as a WT or transfer-inactive mutant form), induces the wrapping of ER membranes around the Golgi complex (Mesmin et al., 2013). However, our data show that OSBP1 is dispensable for the maintenance of ERTGoCS, since they were unaffected in OSBP1-depleted cells. Instead, the combined depletion of OSBP1 and ORP9 is required to disrupt ERTGoCS demonstrating that they play redundant roles regarding their tethering function. The situation is different for ORP10 that is required to maintain the ERTGoCS due to its ability to transfer PS from the ER to the Golgi, thus indicating that the presence/levels of PS at the TGN are involved in the establishment/maintenance of the contact through mechanisms and effectors that remain to be investigated.

The development of the FRET–FLIM approach to visualize the ERTGoCS paves the way for studies that will lead to the identification of the full repertoire of regulators and constituents of ERTGoCS and to an appraisal of the functional consequences of their destabilization or stabilization. Thanks to this approach we have identified a key role of ERTGoCS in controlling the levels of PI4P at the TGN and in acting as a gatekeeper for the TGN export of selected cargoes (see Venditti et al. in this issue).

Primary antibodies used in this study were mouse monoclonal anti-HA (Covance), rabbit polyclonal anti-ORP9 (Sigma-Aldrich), rabbit polyclonal anti-ORP10 (Pierce), rabbit polyclonal anti-β actin (Sigma-Aldrich), and sheep polyclonal anti-TGN46 (Serotec). Rabbit polyclonal antibodies against FAPP1, FAPP2 (Godi et al., 2004), VAP-A, VAP-B (Jansen et al., 2011), Golgin-97 (Medina et al., 2015), ORP11, CERT, and OSBP1 were obtained in our laboratories.

The following reagents were kind gifts: TGN46-FRB-HA-CFP and WT and mutant GFP-PTDSS1 from T. Balla (National Institutes of Health, Bethesda, MD), Venus-Cb5 from N. Borgese (Consiglio Nazionale delle Ricerche [CNR], Institute of Neuroscience, Milan, Italy), PH-FFAT-mCherry of OSBP1 from B. Antonny (Centre National de la Recherche Scientifique [CNRS], Institut de Pharmacologie Moléculaire et Cellulaire, Valbonne, France), and TGN38-HRP from D. Cutler (Medical Research Council Laboratory for Molecular Cell Biology, London, UK). The pIRES-neo2, pmCherry-C3, pmCherry-N1, pEGFP-N1, pEGFP-C3, and pEGFP-C1 were purchased from Clontech; p3XFLAG(CMV-14) was purchased from Sigma-Aldrich; pGEX-4T2 was purchased from GE Healthcare; and C2-Lact-GFP (no. 22852) was purchased from AddGene.

Media, serum, and reagents for tissue culture were purchased from Thermo Fisher Scientific. Unless otherwise stated, all chemicals were purchased from Sigma-Aldrich.

All reagents for molecular biology were purchased from New England Biolabs. All mutagenesis was performed by Site-Directed Mutagenesis (Agilent Technologies). The MCS reporter construct TGN46-FRB-HA-GFP-T2A-mCherry-FKBP-Cb5-pIRES-neo2 was generated by sequential subcloning into the pIRES-neo2 vector (see Table S1 for constructs and cloning strategies). Briefly, the ER–membrane reporter Cb5 (17 residues of the transmembrane domain of rat cytochrome b5 [Bulbarelli et al., 2002], cloned in-frame downstream of the mCherry-FKBP construct) was cloned into pIRES-neo2 followed by in vitro primer annealing of the T2A peptide coding sequence (Osborn et al., 2005) to produce T2A-mCherry-FKBP-Cb5-IRES-neo2 (note that this cloning procedure resulted in the removal of the AgeI restriction site at the 3′ end, which was reconstituted at the 5′ end of the T2A sequence to allow the sequential cloning of the TGN46 construct). The TGN membrane protein TGN46-FRB-HA-GFP (TGN46 fused with the FKBP12-rapamycin-binding domain FRB) was generated by replacing CFP with GFP in TGN46-FRB-HA-CFP. The resulting TGN46-FRB-HA-GFP construct was used as a template for upstream cloning into T2A-mCherry-FKBP-Cb5-pIRES-neo2. The TGN46-FRB-HA-mCherry-T2A-FKBP-Cb5-pIRES-neo2 plasmid (as above but without GFP) was generated similarly by amplifying TGN46-FRB-HA from TGN46-FRB-HA-GFP and cloning into T2A-mCherry-FKBP-Cb5-pIRES-neo2.

All the cloning strategies adopted for generating the other plasmids used in this study are summarized in Table S1.

HeLa cells were grown in high-glucose (4,500 mg/l) DMEM supplemented with 10% FCS. HeLa WT and HeLa VAP-KO cell lines were a generous gift from P. De Camilli (Yale University School of Medicine, New Haven, CT). VAP-KO cells were generated by a TALEN-based gene-editing approach with TALEN pairs specific to exon 2 of VAPA and VAPB and were cultured as previously described (Dong et al., 2016). HeLa cells stably expressing the MCS pIRES-neo2 reporter vector were isolated as a clone obtained by single cell sorting procedures.

For transfection of DNA plasmids, HeLa cells were transfected using JetPEI (Polyplus) as transfection reagent, and the expression was maintained for 16 h before processing.

siRNA sequences used in this study are listed in Table S2. The HeLa cells were treated for 72 h with Oligofectamine (Life Technologies) for direct transfection. The knockdown efficiency was verified by Western blot analysis and is shown in Fig. S4. Cells treated with identical concentrations of nontargeting siRNAs (D-001810-01-20 and D-001810-02-20; Dharmacon) are referred to as controls.

HeLa cells were grown on coverslips and fixed with 4% PFA for 10 min, washed three times with PBS, blocked, and permeabilized for 30 min with blocking solution (0.05% saponin, 0.5% BSA, and 50 mM NH₄Cl in PBS) and incubated with primary antibodies diluted in blocking solution for 1 h at RT. Coverslips were washed with PBS and incubated with fluorochrome-conjugated secondary antibodies (Alexa Fluor 488, Alexa Fluor 568, and Alexa Fluor 633) for 1 h at RT. Cells were mounted in Mowiol and imaged with a Plan-Apochromat 63×/1.4 oil objective on a Zeiss LSM800 or LSM880 confocal system equipped with an Electronically Switchable Illumination and Detection (ESID) module and an AiryScan module (for LSM880) and controlled by a Zen blue software. Fluorescence images presented are representative of cells imaged in at least three independent experiments and were processed with Fiji (ImageJ; National Institutes of Health) software. PS distribution (Fig. 5 E) was quantified by performing the fluorescence intensity ratio of C2-Lact-GFP measured in an ROI defining the Golgi area (delimited by the Golgi marker Golgin-97) and in an equivalent area on the PM as described in the corresponding figure legends. Quantification of ER–Golgi contact sites (Fig. 5, D and F–H; and Fig. S3 A) was performed as indicated below and in the corresponding figure legends. Differences among groups were performed using the unpaired Student’s t test calculated with the GraphPad Prism software. All data were reported as means ± SEM or SD, as indicated in the figure legends.

For FLIM measurements, HeLa cells were seeded on glass-bottom MatTeK dishes. Cells were transfected with the TGN46-FRB-HA-GFP-T2A-mCherry-FKBP-Cb5-pIRES-neo2 plasmid (Bajar et al., 2016). 16 h after transfection, cells were fixed with 1.87% paraformaldehyde in 1× PBS for 10 min at RT. To reduce autofluorescence, cells were subsequently incubated with 1× PBS supplemented with 50 mM NH₄Cl. The imaging was performed in 1× PBS, without any mounting agent. Samples were analyzed the day of preparation without storage at low temperature (i.e., 4°C).

In a parallel set of experiments, we used a different donor fluorophore, Alexa Fluor 430, since the fluorophores Alexa Fluor 430 and mCherry form a favorable Förster donor–acceptor pair with an R0 value (defined as the distance at which 50% of the excited donor molecules transfer energy by FRET) of 6.05 nm (Scalia and Scheffold, 2015). Alexa Fluor 430 was conjugated to anti-HA antibodies (see below) in HeLa cells stably expressing TGN46-FRB-HA-T2A-mCherry-FKBP-Cb5-pIRES-neo2.

Fluorescence lifetime images were acquired using a Zeiss LSM710 (two-photon) confocal microscope, equipped with a FLIM Upgrade kit (Picoquant).

For TGN-GFP detection, the sample was excited using a pulsed femtosecond Ti:Sa laser (Chameleon VISION 2 from Coherent, set at 900 nm) at a fixed repetition rate of 80 MHz. Fluorescence signal was spectrally filtered using a narrowband emission filter (BP-filter 482/35; Chroma). Photons were collected using a single-photon sensitive detector (PMA-Hybrid 40; Picoquant) and timed using a single photon counting module (TimeHarp 260; Picoquant). FLIM data were background subtracted and processed using the pixel-based fitting software SymPhoTime 64 (Picoquant) to calculate the lifetime maps.

We measured the donor lifetime in ROIs, defined as the peripheral regions of the TGN46-GFP positive area (Fig. 2 C) where, in preliminary experiments, we registered that the donor has the lower lifetime suggesting that these were the regions where the donor (i.e., TGN) came closer to the acceptor (i.e., ER). The analysis of lifetime distribution was limited to pixels with >500 total collected photons. Lifetime was determined per single thresholded pixel and used to build histograms of lifetime versus pixel frequency (pixel frequency calculated with reference to total pixels of the ROI). The acquired fluorescence decay curve was deconvoluted with the instrument response function and fitted by a Levenberg–Marquardt least-squares algorithm, using the SymphoTime64 software (PicoQuant). Amplitude-weighted average lifetime was calculated as τ = Σ(α_iτ_i)/Σα_i, where α_i is the amplitude of each lifetime τ_i. FRET–FLIM efficiency was calculated using the equation FRET eff = 1 − (τ_DA/τ_D), where τ_DA is the lifetime of the donor in the presence of the acceptor and τ_D is the lifetime of the donor without the acceptor (Llères et al., 2017). Validation of the FRET signal in cells expressing TGN46-FRB-HA-GFP-T2A-mCherry-FKBP-Cb5-pIRES-neo2 was performed by photobleaching mCherry-Cb5 in the TGN46-GFP region and then measuring FRET–FLIM.

The monoclonal anti-HA antibody was labeled with the reactive fluorescent probe Alexa Fluor 430 NHS Ester (Succinimidyl Ester) purchased from Thermo Fisher Scientific. In brief, the antibody was concentrated and dialyzed in reaction buffer (0.1 M sodium bicarbonate buffer, pH 8.3). 0.2 mg of antibody was mixed with Alexa Fluor 430 with a molar ratio of 12:1 (dye:antibody) in a final volume of 0.125 ml and incubated at RT for 1 h. To remove unreacted fluorescence probe and change buffer, the anti-HA–Alexa Fluor 430 complex was purified using a Sephadex G25 column (GE Healthcare) preequilibrated with 20 mM sodium phosphate, pH 7.4, and 0.1 M NaCl. 50-µl fractions were collected; the absorption spectra between 220 and 600 nm were measured; and the degree of labeling was 4, calculated according to the manufacturer’s protocol.

As shown in Fig. 3 D, brief treatment with low concentrations of rapamycin (2 min, 200 nM) of HeLa cells expressing TGN46-FRB-HA-GFP and mCherry-FKBP-Cb5 induced a marked decrease of TGN-GFP lifetime, indicating a stabilization of ERTGoCS.

We observed that as a consequence of ERTGoCS stabilization, this rapamycin treatment also induced a peri-Golgi clustering of the ER reporter protein Cb5 (Fig. 3 D and Fig. S3 A). Importantly, the perinuclear clustering induced by 2-min rapamycin treatment was largely prevented in ORP10-depleted cells (Fig. S3 A), indicating that under these mild treatment conditions, rapamycin was stabilizing preexisting physiological ERTGoCS and not inducing artificial tethering between the TGN and the ER. However, the perinuclear clustering of Cb5 could be restored in ORP10-depleted cells by reexpressing an siRNA-resistant form of ORP10 (Fig. 5 D and Fig. S3 A) or by elevating PS levels (Fig. 5, F and G). To follow the rapamycin-induced stabilization of ERTGoCS, HeLa cells expressing TGN46-FRB-HA-GFP-mCherry-FKBP-Cb5 and treated with the indicated siRNAs were pretreated with 50 µg/ml cycloheximide for 30 min and then treated with 200 nM rapamycin at 37°C for the indicated time points, followed by paraformaldehyde fixation. Both the cycloheximide and the rapamycin treatments were performed in complete growth medium at 37°C. Cells were then stained with the anti-HA antibody (to verify TGN46 expression) and with a Golgi marker (Golgin-97 to check the overall status of the cells). 15–20 individual fields were imaged (for a total of 150–200 cells) and analyzed to assess mCherry-Cb5 localization. P values were calculated on the basis of mean values from at least three independent experiments.

The EM samples were prepared as previously described (D’Angelo et al., 2007). Briefly, cells were fixed by adding to the culture medium the same volume of a mixture of PHEM buffer (10 mM EGTA, 2 mM MgCl₂, 60 mM Pipes, and 25 mM Hepes, pH 6.9), 4% PFA, and 2% glutaraldehyde for 2 h and then stored in storage solution (PHEM buffer and 0.5% PFA) overnight. After washing with 0.15 M glycine buffer in PBS, the cells were scraped and pelleted by centrifugation, embedded in 10% gelatin, cooled on ice, and cut into 0.5-mm blocks. The blocks were infused with 2.3 M sucrose and trimmed with an ultramicrotome (Ultracut R; Leica) at −120°C using a dry diamond knife.

For the quantification of ERTGoCS, only the Golgi stacks in which both the cis and trans elements were clearly distinguishable were considered. In one set of experiments, ERTGoCS were quantified in HeLa cells expressing TGN38 conjugated with HRP in which the TGN could be identified by the dark DAB reaction product (Fig. 1 A). The membranes of the ER were identified by the presence of associated ribosomes.

The number of ERTGoCS was quantified using the software of a Philips Tecnai-12 electron microscope, considering a contact site to be a region where ER and TGN membranes come closer than 20 nm for a tract of at least 60 nm. We measured both the contact site length, which spanned from 61 to 290 nm, and its width, expressed as the mean distance of three segments of the contact.

To measure the percentage of the TGN surface in contact with the ER, the TGN surface was estimated by using a stereological approach (grid intersection counting; Mayhew, 1979; Ferguson et al., 2017). Briefly, a grid made of arrays of 50 × 50-nm squares was superimposed on each image, and the total intersections of ERTGoCS (areas were the distance between TGN and the ER membranes is ≤20 nm) and the total intersections of TGN membranes with the test square edges were counted. The ratio between the two counts is an estimation of the percentage of TGN surface engaged in MCSs with the ER.

At the same time, we measured the ratio between the TGN surface (calculated as detailed above) and the area of the cytoplasm (expressed in square micrometers) to assess whether the total TGN area was affected in any of the analyzed conditions. We found that this ratio remained constant and similar to the control in all the analyzed conditions.

The 3D organization of ERTGoCS was analyzed in a Crossbeam 550 FIB-SEM system operating under SmartSEM and Atlas3D software (Zeiss). Epon-embedded cell pellets were first investigated by TEM to assess the quality and contrast of the specimens. Resin blocks were then mounted on SEM specimen stubs and coated with a thin layer of platinum. Trimmed surfaces of Epon blocks were used to find the cells and then to obtain tomographic datasets with the slice and view technique. Slicing was performed using the focused Ga ion beam with each step removing 15 nm of the Epon-embedded material. The SEM images were recorded with an EsB detector at 5 × 5-nm pixel size. The contrast of the images was inverted so that they appeared like a conventional brightfield TEM image. Serial FIB-SEM images were imported into open-source ImageJ software to generate QuickTime movies and annotations (Video 1).

The HeLa cells (control and ORP10-KD cells) were broken in hypotonic buffer (10 mM Tris HCl, pH 7.4, and 1.5 mM MgCl₂) containing 1 mM DTT and protease inhibitors using a cell cracker apparatus (at a clearance of 8 µm). Cell lysates were diluted 1:2 in incubation buffer (final concentration, 40 mM Tris HCl, pH 7.4, 80 mM NaCl, 70 mM KCl, 250 mM sorbitol, 1 mM EDTA, 1 mM MgCl₂, 1 mM DTT, and protease inhibitors) and clarified by centrifugation at 1,500 ×g for 10 min at 4°C. The supernatants (1.6 mg ml⁻¹) were preincubated with 0.1 µM GST-LacC2 domain or with 0.1 µM GST for 5 h at 4°C and then incubated with Glutathione Magnetic Agarose Beads (Pierce; 25 µl beads for 500 µl cell lysate) for 1 h at 4°C. After six washes with incubation buffer (without sorbitol and with 0.0005% Tween 20) the beads were treated with Laemmli sample buffer, and the eluted fractions were analyzed by SDS-PAGE and Western blot with antibodies against a TGN marker (TGN46). The ratio between the TGN46 signal in the eluted fractions and that in the input was calculated in control and ORP10-KD samples, and the results were expressed as a percentage of the control.

Fig. S1 shows the validation of the FRET–FLIM approach to visualize ERTGoCS. Fig. S2 shows the morphometric analysis of ERTGoCS in cells depleted of lipid transfer proteins found to be dispensable for ERTGoCS integrity by FRET–FLIM. Fig. S3 shows how ORP10 controls PS levels at the TGN. Fig. S4 shows the effect of siRNAs used in the study on target protein levels. Table S1 shows a list of primers and cloning strategies used in this study. Table S2 shows the list of siRNA sequences used in this study. Video 1 shows the FIB-SEM analysis of ERTGoCS.

We thank all the colleagues who kindly provided the reagents listed in Materials and methods and Francesca Zappa, Leopoldo Staiano, Sangram Raut, Zygmunt Gryczynski, Voelker Buschmann, and Cathal Wilson for helpful discussion.

M.A. De Matteis acknowledges the support of Telethon grant TGM11CB1, Associazione Italiana per la Ricerca sul Cancro grant IG2013_14761, and European Research Council Advanced Investigator grant 670881 (SYSMET). V.M. Olkkonen acknowledges the support of the Academy of Finland (grant 285223) and the Sigrid Juselius Foundation. R. Venditti acknowledges the University of Naples Federico II (Bando STAR Linea1-20187 grant).

The authors declare no competing financial interests.

Author contributions: R. Venditti, L.R. Rega, and M.A. De Matteis conceived the work. R. Venditti, L.R. Rega, and M.C. Masone planned and analyzed most of the experiments. E. Polishchuk and R. Polishchuk performed EM and FIB-SEM analyses. G. Di Tullio and M. Santoro provided technical support. D. Sarnataro and S. Paladino performed the initial FRET analysis. V.M. Olkkonen provided ORP10 reagents and discussed the ORP10 relevant data. S. D’Auria and A. Varriale contributed to the FRET studies and labeled the anti-HA antibody. M.A. De Matteis conceptualized the work and strategy and wrote the manuscript together with R. Venditti and M.C. Masone.