TANGO1 builds a machine for collagen export by recruiting and spatially organizing COPII, tethers and membranes

1As secretory cargoes increase in size and complexity through evolution, mechanisms for their export from the endoplasmic reticulum (ER) must adapt concomitantly. Collagens, the most abundant secretory cargo in mammals - representing nearly 25% of the dry weight of the mammalian body, are some of the most challenging of all secretory cargoes. Several requirements make collagen secretion a challenging task. First, in a complex multi-step process, collagens in the ER fold and trimerise into rigid, rod-like elements of up to 400 nm in length. The folding/assembly of collagen must be coupled to its export, to retain unassembled collagen in the ER, whilst ensuring that all rod-like fully assembled collagen is rapidly exported. Second, assembled collagens are too large to fit into generic COPII-coated vesicles that are usually less than 90 nm in diameter. Third, the rapidity with which this cargo exits the ER and passes through the secretory pathway, requires efficient transfer between compartments.

2Our identification and the subsequent characterisation of TANGO1 (8–11) has revealed a single protein, conserved through most metazoans, that stands at the crossroads of all these processes, modulating them to bring about bulky cargo export from the ER. TANGO1 is an ER exit site (ERES)-localized, transmembrane protein required for export of collagen and other bulky protein components of the extracellular matrix such as Dumpy. Figure 1 is a schematic of three TANGO1 family proteins: TANGO1, TANGO1-short and cTAGE5. A brief description of these proteins follows.

3TANGO1 is a protein of 1907 amino acids (Figure 1A) of which 709 face the cytoplasm. TANGO1 contains a full transmembrane domain and a second membrane-inserted loop, which partially inserts into the inner leaflet of the ER membrane. The lumenal part contains a coiled-coil domain and, at the N terminus, an SH3-like domain. The SH3-like domain binds collagens via HSP47. The cytoplasmic part of TANGO1 is composed of two coiled-coil domains (CC1 and CC2) followed by a C-terminal proline-rich domain (PRD). CC1 contains a domain called TEER (Tether for ERGIC at the ER) that recruits ERGIC-53-containing membranes; CC2 binds cTAGE5 (18), and PRD binds Sec23 and Sec16.

4TANGO1-short is a spliced isoform of TANGO1. It is composed of 785 amino acids that arise from the same exons that encode the cytoplasmic domains of TANGO1. The sequence of TANGO1-short differs in the membrane-inserted helix, and it contains only 15 amino acids at the N terminus, within the ER lumen. It therefore lacks any capacity to interact directly with cargoes. We expect that TANGO1-short binds the same cytoplasmic proteins as TANGO1, but this has not been directly tested.

5Evolutionarily, TANGO1 appears to have been duplicated early in metazoans, yielding a TANGO1-like protein (TALI). Like TANGO1, TALI is expressed as two isoforms. The long isoform is expressed in select tissues while the short isoform (cTAGE5) has a ubiquitous expression. cTAGE5 is composed of 804 amino acids, with a short lumenal stretch of 38 amino acids, followed by a single transmembrane domain. The organisation of cytoplasmic domains is the same as TANGO1, with two coiled-coil domains and a PRD. The first (CC1) of cTAGE5 interacts with Sec12; CC2 interacts with TANGO1 and Sec22, and the PRD, like TANGO1, interacts with Sec23.

6From the published data on these proteins, we can conclude that all three family members bind each other and Sec23. cTAGE5 binds Sec12 and Sec22. TANGO1 (and therefore TANGO1-short) does not bind Sec12. Of these proteins, only TANGO1 can bind cargo in the lumen. How different binding partners could affect the overall function of these proteins in ERES assembly and cargo export, remains untested.

7A newly discovered feature of TANGO1 is its lateral organisation into rings of up to 300 nm diameter, which corral COPII coats at the ERES. The organisation of cTAGE5 and TANGO1-short in TANGO1 rings is not known.

8Exploiting the modular composition of TANGO1, we have generated forms of TANGO1 (Figure 1—figure supplement 1A), each missing one specific domain and hence with one specific set of functions/interactions abrogated. With this set of reagents, we now address how TANGO1 assembles into a functional ring or a fence. We show that this fence of TANGO1 family proteins surrounds COPII, and through specific tethers, physically links the ER and ERGIC for collagen export.

9The role of COPII in TANGO1 ring assembly could be addressed by using a mutant form of TANGO1 that lacks the PRD (TANGO1ΔPRD), which therefore cannot interact with Sec23 (schematic of TANGO1, Figure 1A). 2H5 cells (HeLa cells with TANGO1 deleted using the CRISPR/Cas9 system) were co-transfected with collagen VII and either TANGO1 or TANGO1ΔPRD and imaged using STED microscopy. Full length TANGO1 formed distinct rings of somewhat uniform shape and size (Figure 2A). Surprisingly, TANGO1ΔPRD also assembled into rings, but with two clear differences. First, rings were smaller (Figure 2B, Figure 2—figure supplement 1A); and second, some rings appeared fused with each other to form either a planar tessellation (Figure 2C, Figure 2—figure supplement 1G) or long linear assemblies (Figure 2D, Figure 2—figure supplement 1B–F). Quantitative morphological descriptors of the size and shape of structures formed by TANGO1 constructs, were extracted using semi-automated image analysis (Figure 2—figure supplement 2, Table 1) and are described in detail in the Materials and methods section and the figure legend. Specifically, we fitted rings to an elliptical shape and measured the diameters of the ring in terms of major and minor axes of its fitted ellipse. This works well for regular-shaped ellipses, however for structures and shapes that deviate from an elliptical shape, a rectangular bounding shape is a more useful approximation. Therefore, maximum and minimum diameters (Feret’s maximum or minimum) were also extracted and all these values are plotted in Figure 2E. From this quantification, we confirmed that rings formed by TANGO1ΔPRD, are significantly smaller than rings formed by TANGO1 (Figure 2E, Table 1). We used the aspect ratio (the ratio of the major to minor axes of the fitted ellipses) as a descriptor of the shape of rings. By this measure, rings formed by TANGO1 and TANGO1ΔPRD had a similar shape (Figure 2F).

10It is important to note that these cells still contain TANGO1-short and cTAGE5 (Figure 1), both of which will recruit TANGO1ΔPRD to ERES. These data suggest that the cytoplasmic domains of the TANGO1-family of proteins act as a single unit and any one can assemble into a ring, however TANGO1 brings cargo to the exit site. This suggests that overexpressing cytoplasmic isoforms (either TANGO1-short or cTAGE5) would increase the capacity of an ERES to export cargo, however TANGO1 is the only protein with the capacity to bring cargo to ERES. Collagen secreted in the absence of TANGO1 might thus be in an unfolded or unassembled form.

11In a complementary experiment, we studied the effect of Sec23A depletion on TANGO1 ring formation in RDEB/FB/C7 fibroblasts. Depleting cells of all Sec23 could create cellular stress and affect endomembrane regulation, so we attempted to minimise such a potential stress by using siRNA that targeted exclusively Sec23A, and not Sec23B. As expected, collagen export from the ER was reduced in Sec23A-depleted cells (Figure 2—figure supplement 3).

12While TANGO1 in control cells was often visualised in rings (Figure 2G), depletion of Sec23A appeared to phenocopy our results with TANGO1ΔPRD, showing multiple seemingly fused rings of TANGO1 assembled in planar arrays (Figure 2H, Figure 2—figure supplement 4), quantified in Figure 2I. These structures/abnormal rings were almost never observed in cells expressing full length TANGO1, or cells that are not depleted of Sec23A.

13Based on our super-resolution microscopy images, we hypothesise that TANGO1 rings could be represented as a multimeric assembly of units of TANGO1 family proteins (TANGO1, TANGO1-short and cTAGE5) that assemble into a fence.

14A key feature that could provide strength to a fence of TANGO1 would be lateral interactions between components in the fence. For example, the TANGO1-interacting protein cTAGE5 (Figure 3A) should be a component of the ring and could contribute to lateral interactions in the ring. We visualised TANGO1 and cTAGE5 in RDEB/FB/C7 cells by STED microscopy. Due to the low quality of commercially available anti-cTAGE5 antibodies for immunofluorescence, we were unable to visualise the localisation of cTAGE5 as clearly as TANGO1, nonetheless cTAGE5 clearly localised along the rings delineated by TANGO1 (Figure 3B).

15To test the involvement of cTAGE5 in TANGO1 ring formation, we generated a construct of TANGO1 lacking the second cytoplasmic coiled-coil (TANGO1ΔCC2) domain (Figure 1—figure supplement 1 for a schematic) and hence, unable to interact with cTAGE5 (Figure 3A). STED microscopy revealed that, in contrast to full length TANGO1 (Figure 3C), TANGO1ΔCC2 assembled into misshapen structures (Figure 3D and Figure 3—figure supplement 1). Ring size and shape were quantified as in the previous section. Rings formed by TANGO1ΔCC2 were more variable in size (Figure 3E, Table 1) and shape (Figure 3F) than those formed by full length TANGO1.

16As a complementary approach, we characterised the effect of depleting cTAGE5, on ring formation in cells with endogenous TANGO1. As expected, in RDEB/FB/C7 fibroblasts depleted of cTAGE5 (Figure 3—figure supplement 2A), collagen secretion was blocked (Figure 3—figure supplement 2B,C). TANGO1 structures phenocopied TANGO1ΔCC2 structures in 2H5 cells: rings of TANGO1 were misassembled (Figure 3G) and formed unusual shapes, without considerably altering the number of rings observed (Figure 3H).

17Another lateral interaction that might maintain fence integrity could be an intrinsic ability of TANGO1 to self-associate. A test of this proposition would be to identify a domain in TANGO1 that mediates self-association and show that it has a role in ring formation. To identify such a domain, we tested the ability of TANGO1-FLAG to co-immunoprecipitate with TANGO1ΔPRD, TANGO1ΔCC2 or TANGO1ΔCC1 (Figure 1—figure supplement 1). We observed (Figure 4A) that TANGO1-FLAG was immunoprecipitated by TANGO1 and TANGO1ΔPRD, but not by TANGO1ΔCC2 (Figure 4A) or TANGO1ΔCC1 (Figure 4B). Reasoning that the effect of the CC2 was likely indirect, as TANGO1ΔCC2 is unable to interact with cTAGE5 (Figure 4A), we focused on the first coiled-coil domain (CC1) to identify a minimal region required for self-association. We generated two TANGO1 constructs with smaller deletions from the CC1, each of which had a deletion in a portion of the coiled-coil (TANGO1Δ1255–1295 and TANGO1Δ1296–1336). As a control, we confirmed these constructs still interacted with cTAGE5 (Figure 4B). Only TANGO1Δ1255–1295 did not immunoprecipitate TANGO1-FLAG (Figure 4B).

18With a minimal self-association domain (a.a. 1255–1295) identified, we looked for its role in TANGO1 ring formation. 2H5 cells were co-transfected with collagen VII and either TANGO1ΔCC1, TANGO1Δ1255–1295 or TANGO1Δ1296–1336 and then imaged by STED microscopy. In line with our predictions, TANGO1ΔCC1 or TANGO1Δ1255–1295 could not form rings; of the 16 and 15 cells examined respectively, there were few discernible polymeric assemblies of TANGO1 (Figure 4C,D), while TANGO1Δ1296–1336 behaved as full length TANGO1, forming distinct, readily detectable, independent rings (Figure 4E) of similar size (Figure 4—figure supplement 1A) and shape (Figure 4—figure supplement 1B) as TANGO1. These data indicate that TANGO1-TANGO1 interactions (Figure 4F), mediated by amino acids 1255–1295, are required to maintain ring integrity.

19In our coarse-grained view of this fence of TANGO1 and TANGO1 family of proteins (cTAGE5 and TANGO1-short), we would describe our data thus far in terms of two general sets of interactions. First, lateral interactions mediated by TANGO1 self-association and its interaction with cTAGE5 and TANGO1-short, and second, inward attractions of TANGO1/cTAGE5/TANGO1-short to COPII, thus affecting the ring size and its placement with respect to COPII budding machinery.

20We have shown recently that TANGO1, via its CC1, recruits ERGIC membranes that fuse at the ERES. Could TANGO1 rings concentrate membrane recruitment for mega-carrier biogenesis? What role does the TEER domain play in ring assembly? To address these questions, we first identified a minimal TEER domain within the CC1, using our previously developed approach.

21Following our previous methodology, we generated two myc-tagged, mitochondrially-targeted TEER (mit-TEER truncates) constructs of 82 and 81 amino acids, respectively. Our original construct had TANGO1 amino acids 1188 to 1396. From this, we generated two smaller constructs. In one, we deleted amino acids 1255–1295 (mit-Δ1255–1295); while in the other we deleted amino acids 1296–1336 (mit-Δ1296–1336) (Figure 5A). These corresponded exactly to the deletions in the CC1 described in the previous section.

22We expressed the constructs in HeLa cells, fixed and then stained them using an anti-myc antibody and visualised these samples using confocal microscopy (Figure 5B). We confirmed the two constructs co-localised with the mitochondrial marker HSP60 (Figure 5C). The extent of overlap of myc-epitope and HSP60 was quantified and is plotted as the Manders’ overlap coefficient (Figure 5D).

23As before, we co-stained transfected cells with anti-ERGIC-53 and anti-myc antibodies. To our surprise, mitochondria expressing mit-Δ1255–1295 showed no recruitment of ERGIC-53-containing membranes (Figure 5E). In contrast, mit-Δ1296–1336 still functioned as the TEER domain and recruited ERGIC membranes. The extent of colocalisation of ERGIC-53 and myc for the two constructs was quantified and is plotted as Manders’ overlap coefficient (Figure 5F). This tells us that the minimal TEER is exactly the same forty amino acids we identified in the previous section, as those required for the self-association of TANGO1. This implies that either a TANGO1 dimer can recruit a tether or the tether links two TANGO1 monomers. This hypothesis is tested and presented in Figure 7.

24But how does this minimal TEER domain recruit ERGIC membranes? A prime candidate for this tethering activity is the evolutionarily conserved NRZ (NBAS, RINT1, ZW10) protein tether. NRZ tether is a multi-subunit tether complex (MTC) that assembles at the surface of the ER, is required for retrograde capture of membranes, partially localises to ER exit sites and interacts with SNAREs that we have shown previously are required for collagen export from the ER. One component of the MTC (RINT1) was also identified in our screen for genes required for protein secretion. Mutations in another component NBAS, are linked to dysregulated collagen secretion in atypical osteogenesis imperfecta.

25As in previous sections, we imaged TANGO1 in RDEB/FB/C7 cells, with Sec31 and RINT1 by confocal microscopy (Figure 6—figure supplement 1) and, by STED microscopy observed the tether protein RINT1 localised to one or two puncta at rings of TANGO1, occasionally adjacent to ERGIC-53-containing membranes (Figure 6A and Figure 6—figure supplement 2).

26We transfected full-length TANGO1, TANGO1Δ1255–1295, TANGO1Δ1296–1336 or TANGO1-Lum (lumenal) in HEK293T cells and attempted to co-immunoprecipitate tether proteins. We saw that full length TANGO1 and TANGO1Δ1296–1336 immunoprecipitated all three of the proteins that form the tether (NBAS, RINT1, ZW10) (Figure 6B). This interaction was completely abrogated when we used TANGO1Δ1255–1295 (lacking the minimal TEER domain). As controls, we confirmed all constructs still interacted with cTAGE5 and TANGO1-Lum did not immunoprecipitate either tether proteins or cTAGE5 (Figure 6B).

27Depleting TANGO1, NBAS or RINT1 from RDEB/FB/C7 fibroblasts inhibited collagen VII secretion (Figure 6C–E) and arrested collagen in the ER (Figure 6C). Does TANGO1 recruit ERGIC to intracellular collagen accumulations via the NRZ tether? In cells depleted of RINT1, NBAS or TANGO1 (Figure 6F,H), we quantified ERGIC recruitment to accumulations of collagen in the ER. In all cases, ERGIC membrane recruitment was significantly reduced (Figure 6G).

28These data showed a novel function of TANGO1, to recruit ERGIC membranes via the retrograde NRZ MTC to the ERES for collagen export. But is this function built into ring assembly?

29In RDEB/FB/C7 depleted of RINT1, TANGO1 rings were completely disrupted (siCTRL vs. siRINT1 Figure 7A vs. B). We individually depleted each of the other two proteins in the MTC (NBAS or ZW10) and checked for the ability of TANGO1 to assemble into rings in RDEB/FB/C7 fibroblasts. As seen after depleting cells of RINT1, rings were observed far less frequently (quantified in Figure 7C). In all cases, ERES, as marked by TANGO1 and SEC31 are still formed (Figure 7—figure supplement 1).

30There are at least two mechanistic possibilities that could link tether binding, the TANGO1 self-association domain, and ring formation. Either (a) the tether is required to hold together TANGO1 and TANGO1-short in the fence; or (b) complexes form with TANGO1/TANGO1 short and this dimer then recruits the tether, which stitches together a higher order structure, forming a fence.

31We tested these hypotheses by performing sequential co-immunoprecipitations to look for TANGO1, TANGO1-short and cTAGE5 in a stable complex. Using lentiviral infections, we generated HEK293T cells stably expressing cTAGE5-FLAG and TANGO1-HA. We depleted these cells of individual NRZ tether proteins and then performed sequential immunoprecipitation, pulling first on cTAGE5-FLAG and then TANGO1-HA and finally probed for TANGO1-short (Figure 7D,F for schematic). We observed that the NRZ tether had no effect on the association of TANGO1 and TANGO1-short in a stable complex (Figure 7E).

32These data showed that the NRZ tether is required for TANGO1 to assemble into a ring and indicated that stable complexes of TANGO1, cTAGE5 and TANGO1-short, recruit the tether.

33Our new data describe a mechanism whereby the very processes by which TANGO1 recruits ERES machinery and cargo, also bring about its own assembly into a fence of defined size. This in turn remodels the ERES, and in the lumen, via Hsp47, binds and potentially segregates assembled bulky cargoes (Figure 8A). Such a concerted mechanism circumvents a causality dilemma (the chicken-or-the-egg problem) in this process – neither ring nor function precedes the other; they assemble together, requiring each other to do so.

34There are several broad implications of our data, addressing fundamental aspects of early secretory pathway organisation and cargo export.

35Tethers play a central role in membrane targeting and organelle biogenesis. Improved structural understanding has revealed fascinating models for the mechanisms of membrane recruitment by tethers. Our discovery of membrane recruitment by TANGO1 and its use of the NRZ tethering complex (Figures 6 and 7) has far reaching implications. A critical aspect of TANGO1 biology is that it functionally and physically couples anterograde to retrograde traffic at an ERES, coupling two successive compartments in the secretory pathway, allowing for more rapid and efficient cargo transport between the compartments. The NRZ tether would bind to, and recruit, any COPI-coated ERGIC-53-containing membranes in the vicinity of the ERES – but what of ERES closely apposed to the cis-Golgi, and what of organisms such as D. melanogaster, which have no discernible ERGIC compartment? Under such circumstances, the ‘carrier’ for collagen formed by the retrograde recruitment of COPI-coated membranes could just be the first Golgi cisterna. In other words, we could envisage a direct continuity or ‘tunnel’ between the ER and the Golgi, with a ring of TANGO1 and its associated exit site machinery holding together the two compartments, but also functionally delimiting them.

36We have not observed a complete ring of tethers with TANGO1. The tethers instead appear as one or two puncta at the ring circumference. One can envisage that an initiation point of the TANGO1 ring recruits tethers and TANGO1 continues to assemble into a ring whereas the tethers remain at the nucleation site. This would explain the images presented (Figure 6A and Figure 6—figure supplement 2). Without the tethers, the reaction is stalled and TANGO1 fails to assemble further into a ring, providing an explanation for the requirement of tethers in TANGO1 ring assembly. An alternative is that the tethers are not recruited at the site of ring nucleation but present throughout, and we are unable to capture this final assembled state.

37We had proposed that TANGO1 functioned by binding to and stabilising the inner COPII coat to delay the recruitment of the outer coat and the subsequent fission of a newly forming carrier, for as long as is required to assemble and pack the bulky cargo collagen. We would like to suggest a possible physical mechanism of how TANGO1 rings are assembled and maintained by means of protein-protein interactions and eventually regulate the formation of a collagen-containing megacarrier. First, based on our observations of TANGO1 rings by STED microscopy, (Figure 2), and our data indicating the different protein–protein interactions between the members of the TANGO1 family, we propose that a fence of TANGO1 can be described as a filament, held together by these lateral protein–protein interactions, which normally surrounds COPII patches at the ERES. Importantly, this description of the ring as a filament will remain an approximation until the molecular composition and structural alignment of individual components is known. Such a filament would be subjected to elastic strains and stresses and would hence resist bending. Second, COPII subunits polymerise into structures of growing size. COPII subunits at the periphery of a polymerising domain have free binding sites and hence higher chemical energy than fully polymerised subunits at the centre of the domain, which, in physical terms, translates into the existence of an effective line-energy of the ERES. As proteins of the TANGO1 family physically interact with Sec23, Sec16, and Sec12, we propose that upon adsorption to the ERES by binding peripheral COPII subunits, TANGO1 would effectively reduce the ERES line energy. A tug-of-war between the filament bending and the effect on COPII stabilisation created by the adsorption of TANGO1 filaments around ERES would then dictate whether and how TANGO1 rings are formed. Interestingly, it has been shown that the line tension of the polymerising protein coat can play a key role in controlling the timing and size of clathrin-coated vesicles. We thus propose that the stabilising effect of TANGO1 while adsorbing around ERES would serve as a physical mechanism to delay and enlarge the COPII vesicle, commensurate with cargo size. Furthermore, TANGO1 rings could serve as a mould to impose a cylindrical curvature at the base of a growing carrier by coupling to the first layer of the COPII coat (Figure 8A,B), as proposed by Ma and Goldberg.

38We expect that the diameter of a TANGO1 ring and associated components, will be maximal, proximal to the plane of the membrane. The more distal parts of the proteins for example the PRD (of TANGO1, cTAGE5 and TANGO1-short) will have two extreme positions:1, lying pointing radially inward like spokes of a wheel and 2, pushed aside to the ring periphery by the growing carrier. It is therefore difficult to make definitive statements about relative locations - based on antibodies that bind to distal parts of the molecule - within the ring. We have also not tested whether cTAGE5, or for that matter TANGO1 short, can assemble into a ring in cells lacking TANGO1. We have not been able to create a form of cTAGE5 and TANGO1-short with a label or an antibody to visualise the domains proximal to the membranes, which makes it difficult to discern their location precisely, even in the presence of endogenous TANGO1. However, within these limitations, based on the involvement of various parts of TANGO1 and its interactors into discrete rings for collagen export, we could now begin to address the placement of various proteins such as TFG, KLHL12 or sedlin in collagen export from the ER.

39Under these conditions, there is the possibility that a mega carrier, of the form recently reported by Schekman and colleagues, is produced. Regardless of the final form adopted by the cells to transfer collagen from the lumen of the ER to the Golgi, with the data presented herein, we have taken the first steps toward arriving at a quantitative understanding of this hypothesis. We envision that a full description and analysis of such a quantitative physical model of TANGO1 ring assembly and megacarrier formation will help us better understand this fundamental process.

40Little is understood about how client folding in the ER is coupled to export, how misfolded proteins and ER residents are excluded from an ERES, and what role the client plays in the biogenesis of its own carrier. TANGO1 recruits collagen via HSP47 – a chaperone that selectively recognises triple helical (export-competent) collagen. Can this interaction of triple helical collagen and TANGO1 help effect ring assembly? Does a ring of TANGO1 (and therefore a carrier) form in response to selection of folded collagen, excluding misfolded collagen? Does folded cargo define the site and size of a transport carrier?

41In toto, our data indicate that TANGO1, by assembling into a ring at ERES generates a semi-stable sub-domain across multiple compartments. The processes that allow this assembly also co-ordinately select, partition, and organise export machinery, and membrane for a cargo-export tubule/carrier, thus defining the minimal machinery for collagen export.

42RDEB/FB/C7, HEK293T and HeLa cells were grown at 37°C with 5% CO2 in complete DMEM with 10% FBS unless otherwise stated. Plasmids were transfected in HeLa cells with TransIT-HeLa MONSTER (Mirus Bio LLC) or Lipofectamine 3000 Transfection Reagent (Thermo Fisher Scientific) according to the manufacturer’s protocols. All cells in culture were tested every month to confirm they were clear of contamination by mycoplasma.

43C-terminally HA-tagged full-length TANGO1 was cloned into the polylinker of pHRSIN transfer plasmid using BamHI/SalI restriction enzymes. Lentiviral particles were produced by co-transfecting pHRSIN-TANGO1-HA and a packaging vector pool (pCMV 8.91 and pMDG) into HEK293T cells using TransIT-293 (Mirus Bio LLC). 48 hr post transfection, the viral supernatant was harvested, filtered, and directly added to HEK293T cells. Stably expressing HEK293T cells were selected using 500 µg/ml hygromycin.

44C-terminally FLAG-tagged full-length cTAGE5 was cloned into pJLM1 transfer plasmid using NheI/EcoRI restriction enzymes. Lentiviral particles were produced by co-transfecting pJLM1-cTAGE5-FLAG and a packaging vector pool (pPAX2 and pMD2.G) into HEK293T cells with using TransIT-293 (Mirus Bio LLC). 48 hr post transfection the viral supernatant was harvested, filtered, and directly added to TANGO1-HA expressing HEK293T cells. Cells stabling expressing Tango1-HA and cTAGE5-FLAG were selected using 500 µg/ml Hygromycin and Puromycin 4 µg/ml.

45All molecular cloning was carried out using MAX Efficiency Stbl2 Competent Cells – (Thermo Fisher Scientific) following manufacturer’s instructions.

46siRNA oligos were purchased from Eurofins Genomics (Ebersberg, Germany). The oligo sequences used were RINT1 5’-GGUUAUAACUGACAGGUAU-3’, NBAS 5’-CUGCUUCAGUAUGGAUUAA ZW10 5’-UGGACGAUGAAGAGAAUUA-3’, TANGO1 5’-GAUAAGGUCUUCCGUGCUU-3’, cTAGE5 5’-UUGAAGACUCCAAAGUACA-3’, SAR1A 5’-GAACAGAUGCAAUCAGUGATT-3’, SAR1B 5’-GCAUAACUUGAAUUCAAUATT-3’. SEC23A siRNA (Cat # L-009582–01) was purchased from GE Dharmacon (Colorado, USA).

47The following antibodies were used collagen VII (rabbit anti–human [Abcam]; mouse anti–human [Sigma-Aldrich]), ERGIC-53 (mouse anti–human; Santa Cruz Biotechnology, Inc., and Enzo Life Sciences), Sec31A (mouse anti–human; BD), TANGO1 (rabbit anti–human; Sigma-Aldrich; rabbit anti-human in-house), HSP47 and calreticulin (goat anti–human; Enzo Life Sciences), HA (mouse; BioLegend), SAR1 (mouse anti–human; Abcam), β-tubulin (mouse anti-human; SIGMA-Aldrich), β-actin (mouse anti-human; SIGMA-Aldrich), NBAS (rabbit anti-human SIGMA-Aldrich), RINT1 (rabbit anti-human; SIGMA-Aldrich and goat anti-human (Santa Cruz Biotechnology), ZW10 (rabbit anti-human; Abcam), Sec23 (rabbit anti-human/mouse/rat; Abcam), cTAGE5 (rabbit anti-human Atlas antibodies, mouse anti-human Santa Cruz Biotechnology), TGN46 (sheep polyclonal, Bio-Rad), HA (mouse monoclonal, BioLegend; rat monoclonal BioLegend), FLAG (mouse monoclonal, rabbit, SIGMA-Aldrich; goat, Novus) HSP60 (mouse anti-human SIGMA-Aldrich), c-myc (mouse monoclonal, rabbit, SIGMA-Aldrich). Mounting media used in confocal and STED microscopy were either Vectashield (Vector Laboratories) or ProLong (Thermo Fisher Scientific, Waltham, Massachusetts).

48Cells extracted with lysis buffer consisting of 50 mM Tris-HCl (pH 7.4), 150 mM NaCl, 1 mM EDTA, 2% CHAPS, and protease inhibitors were centrifuged at 20,000 × g for 30 min at 4°C. Cell lysates were immunoprecipitated with FLAG M2 (SIGMA-Aldrich) or HA (Thermo Scientific) antibodies. Beads were washed three times with Tris-buffered saline (TBS)/0.5% CHAPS and processed for sample preparation.

49For sequential immunoprecipitations, a first immunoprecipitation with FLAG would bring all proteins that interact with cTAGE5; a subsequent immunoprecipitation with HA would only yield proteins that were bound to both cTAGE5 and TANGO1-HA.

50Cells grown on coverslips were fixed with cold methanol for 8 min at −20°C or 4% formaldehyde (Ted Pella, Inc.) for 15 min at room temperature. Cells fixed with formaldehyde were permeabilised with 0.1% Triton in PBS and then incubated with blocking reagent (Roche) or 0.1% horse serum for 30 min at room temperature. Primary antibodies were diluted in blocking reagent or 0.1% horse serum and incubated overnight at 4°C or at 37°C for 1 hr. Secondary antibodies conjugated with Alexa Fluor 594, 488, or 647 were diluted in blocking reagent and incubated for 1 hr at room temperature.

51Confocal images were taken with a TCS SP5 (63×, 1.4–0.6 NA, oil, HCX PL APO), TCS SP8 (63×, 1.4 NA, oil, HC PL APO CS2), all from Leica Microsystems, using Leica acquisition software. Lasers and spectral detection bands were chosen for the optimal imaging of Alexa Fluor 488, 594, and 647 signals. Two-channel colocalisation analysis was performed using ImageJ (National Institutes of Health), and the Manders’ correlation coefficient was calculated using the plugins JaCop or Coloc 2.

52STED images were taken on a TCS SP8 STED 3 × microscope (Leica Microsystems) on a DMI8 stand using a 100 × 1.4 NA oil HCS2 PL APO objective and a pulsed supercontinuum light source (white light laser). Images were acquired and deconvolved exactly as described before.

53Three-colour STED: Due to incompatible species specificities of primary antibodies available (for RINT1, TANGO1 and ERGIC-53), we were forced to use sub-optimal secondary antibodies. We used Alexa 488, Alexa 594 and Alexa 647. This required that we set the depletion laser (775 nm) at only 3–8% intensity for the Alexa 647 channel to prevent rapid bleaching.

54Multichannel 3D stacks were acquired with a z-step size of 100 nm and subsequently deconvolved using Huygens deconvolution software (Scientific Volume Imaging) for STED modes using shift correction to account for drift during stack acquisition. Sum-Intensity Projections were then generated from a subset of the deconvolved stack slices where the rings were present. Projected images showed a large fraction of the GFP signal as random dots or big aggregates in which no particular structural organisation could be distinguished. Also, a significant amount of well-defined non-random structures, i.e. both full and incomplete (arc-shaped or dotted) rings, as well as chain-like assemblies of rings.

55To ensure a systematic and unbiased analysis, these structures are first segmented via a trainable pixel level classifier, and subsequently labelled either as rings, incomplete rings or dots, or ring aggregates, on object level. Both pixel and object classification used a machine learning based open-source software, ilastik. Afterwards, we calculated different parameters for each object to compare them quantitatively in shape and size. Specifically, we measured the diameters of the ring in terms of major and minor axes of its fitted ellipse and the maximum and minimum Feret’s diameter. Statistical testing was performed using Student’s t test (continuous data, two groups). One asterisk indicates Student’s t test value p<0.06; three asterisks p<0.006; ns indicates not significant.

56To quantify the frequency of rings after depletion of specific gene products, deconvolved STED images of each condition were manually scored for rings/clusters of TANGO1. A ring is defined as an independent structure with an internal hole. A cluster however, is at least four such conjoint rings. Statistical testing was performed using Student’s t test (continuous data, two groups). One asterisk indicates Student’s t test value p<0.02; three asterisks p<0.002; ns indicates not significant.

57The secretion assay was carried out exactly as described earlier. Briefly, RDEB/FB/C7 fibroblasts were transfected in suspension on two consecutive days with siRNA (either a pool of control, non-targeting RNA or RNA targeting a specific gene). 48 hr later, cells were washed thoroughly and incubated for 20 hr in OptiMEM supplemented with 1 mM ascorbate. Cell lysates and media were harvested and processed for Western blotting of collagen VII and tubulin/actin as loading/lysis controls.