Ubiquitination regulates ER-phagy and remodelling of endoplasmic reticulum

1FAM134B is a mammalian reticulon-like protein that shapes the ER membrane,. It is also an ER-phagy receptor, mediating the fragmentation and selective degradation of ER sheets. Structural modelling and molecular simulations have revealed that the RHD of FAM134B forms wedge-shaped membrane inclusions that induce positive membrane curvature to promote ER fragmentation, assisted by RHD clustering–. However, it is unclear how FAM134B-mediated ER-phagy is regulated in mammalian cells.

2Ubiquitination regulates a large number of cellular processes, so we used mass spectrometry (MS) to investigate its potential role in ER-phagy by mapping the ubiquitination profile of FAM134B (Fig. 1a). Proteomic analysis of FAM134B-derived diGly peptides identified residues K90, K160, K264 and K247 as the primary ubiquitination sites (Fig. 1b), which are located within the cytosolic segments of the FAM134B RHD (Fig. 1c). The induction of ER-phagy with Torin 1 increased the representation of all four diGly peptides (Fig. 1b) as well as the overall endogenous ubiquitination level of haemagglutinin (HA)-tagged FAM134B (Fig. 1d,e and Extended Data Fig. 1a, compare lanes 3 and 1, and Extended Data Fig. 1b). The accumulation of total and ubiquitinated endogenous FAM134B in cells treated with bafilomycin A1 (BafA1) indicated lysosomal degradation (Extended Data Fig. 1a, compare lane 4 to lanes 3 and 2, and Extended Data Fig. 1b). No accumulation was observed in cells treated with the proteasome inhibitor MG132 (Extended Data Fig. 1c). Accordingly, cycloheximide chase experiments showed that BafA1 rendered FAM134B more stable (Extended Data Fig. 1d,e). We also found that TAK243, a potent inhibitor of the ubiquitin (Ub)-activating enzyme, abolished endogenous FAM134B ubiquitination (Extended Data Fig. 1f) and delayed its basal turnover (Extended Data Fig. 1g,h). As an alternative method, we substituted lysine residues identified by MS (K90, K160, K247 and K264) and their neighbours (K252, K265, K278 and K291) with arginine, but overall ubiquitination levels, binding to LC3B-II and the number of FAM134B–LC3B-decorated ER fragments were not affected in this mutant (Extended Data Fig. 1i,j). Only the replacement of the entire set of 17 highly conserved lysine residues (mutant HA–FAM134B17KR) resulted in a strong decrease in FAM134B RHD ubiquitination (Fig. 1f, compare lanes 6 and 5). Of note, the lack of FAM134B RHD ubiquitination correlated with decreased binding to LC3B-II (Fig. 1f, compare lanes 2 and 1, and Fig. 1g; approximately 1.32-fold reduction). High-molecular-weight (oligomeric) species of FAM134B were less abundant in the HA–FAM134B17KR mutant (Fig. 1f, compare lanes 2 and 1, and Fig. 1g; approximately 1.5-fold reduction). Similar results were observed following the chemical crosslinking of intact membranes from cells expressing the wild-type (HA–FAM134BWT) or mutant (HA–FAM134B17KR) receptor are consistent with the high-molecular-weight species representing oligomers (Extended Data Fig. 1k).

3Several results indicated that ubiquitination of the FAM134 RHD can promote the formation and/or stabilization of FAM134B oligomers implicated in ER-phagy. First, high-molecular-weight species of FAM134B accumulate when the flux of ER-phagy is inhibited and/or following the deletion of the LC3-interacting region (LIR), indicating that FAM134B is delivered to lysosomes in its oligomeric form (Extended Data Fig. 1l,m). Second, in the absence of the FAM134B LIR, the ubiquitinated forms of FAM134B accumulate, confirming its autophagy-mediated degradation (Extended Data Fig. 1n). Third, immunofluorescence analysis following exposure to Torin 1 showed that approximately 70% of FAM134B+ autophagosomes contained Ub, suggesting Ub has a wide-engaging role in FAM134B-driven ER-phagy (Fig. 1h,i). These results show that the ubiquitination of the FAM134 RHD can promote the formation and/or stabilization of FAM134B oligomers implicated in ER-phagy.

4We studied the effect of ubiquitination on RHD structure and dynamics with coarse-grained MD simulations of mono-ubiquitinated and bi-mono-ubiquitinated FAM134B RHD molecules, namely, K160–Ub, K264–Ub and (K160 + K264)–Ub, embedded in POPC (16:0/18:1 PC) lipid bilayers. The most populated conformations of the ubiquitinated FAM134B RHD molecules are shown in runs of up to 10 µs each (Fig. 2a). The Ub moieties perturb the RHD structure locally by mediating multiple interactions with the proximal cytosolic loops of the RHD and the POPC bilayer (Extended Data Fig. 2a–d). The main hydrophobic face of K160–Ub makes substantial contact with POPC lipids, widening the wedge shape of the RHD (Extended Data Fig. 2b). In the RHD of K264–Ub, Ub interacts primarily with the cytosolic loops and is located on top of the RHD (Extended Data Fig. 2c,d). In the bi-mono-ubiquitinated variant (K160 + K264)–Ub, the two Ub moieties mediate cis-interactions with each other and are bundled on top of the RHD (Extended Data Fig. 2e,f). Increased radii of gyration of ubiquitinated RHDs result in a larger footprint on the bilayer, thus possibly perturbing it more severely (Extended Data Fig. 2g,h). These changes in intrinsic RHD structure due to ubiquitination may affect its membrane curvature induction and sensing functions.

5To test whether ubiquitination affects the RHD-mediated induction of membrane curvature, we used in silico simulations to remodel a discontinuous bicelle (DMPC + DHPC lipids) into a closed vesicle at 300 K (Fig. 2b and Supplementary Videos 1–3). The ubiquitinated and non-ubiquitinated RHDs induced a positive mean curvature (+H) leading to vesicle formation within the first 500 ns (3 × 100 replicates; 1 μs each; Extended Data Fig. 3a–d). By measuring the kinetics and rates for vesicle formation (see Methods; Extended Data Fig. 3e,f), we estimated that single K160–Ub, K264–Ub and (K160 + K264)–Ub species accelerated vesicle formation moderately by factors of 1.49, 1.37 and 1.38, respectively (Fig. 2b and Extended Data Fig. 3g). To highlight the effect of Ub on curvature induction, we performed these assays at 280 K, which increases the energy barrier to induce curvature and vesiculation. We found that bicelles containing the FAM134B RHD and K160–Ub did not effectively transit to closed vesicles (Extended Data Fig. 3h,k), whereas bicelles containing K264–Ub or (K160 + K264)–Ub formed closed vesicles, indicating that these variants can overcome the barrier even at 280 K (Extended Data Fig. 3i,j). Lateral diffusion of ubiquitinated variants in buckled bilayers showed no differences in preferred curvature, indicating that the strong curvature-sensing function of the RHD is preserved upon ubiquitination (Extended Data Fig. 4a–c).

6Next, we tested how ubiquitination affects the FAM134B RHD-mediated remodelling of liposomes in vitro. We created and purified N-terminal and C-terminal gene fusions of Ub to FAM134B RHD90–264 (described as Ub–RHD–Ub). As a control, we created a construct encompassing the entire RHD of FAM134B without Ub (RHD90–264). The incubation of liposomes with Ub–RHD–Ub led to significantly smaller structures with a narrow distribution than the liposomes incubated with RHD proteins alone (Fig. 2c), indicating a significant gain of membrane-remodelling activity for ubiquitinated RHD proteins. In cells, both constructs localized to the ER, based on the ER marker REEP5. We observed a significant increase in the number of RHD–REEP5-containing puncta in cells expressing Ub–RHD–Ub. These puncta may represent clusters of RHD-containing proteins, enhanced by the presence of ubiquitinated FAM134B RHDs (Extended Data Fig. 4d,e).

7Next, we simulated the behaviour of ubiquitinated RHDs to induce curvature-mediated protein sorting and clustering (Extended Data Fig. 4f). In the absence of ubiquitination, the two RHD molecules on the buckled membrane diffused to the top of the buckle, where they formed a loose cluster. Ubiquitination slowed down the diffusion somewhat but resulted in a tighter cluster at the top of the buckle that persisted for the entire simulation (up to 25 µs). From 19 µs onwards, Ub-mediated contacts stabilized the dimeric complex (inset in Extended Data Fig. 4f and Supplementary Video 4). Next, we simulated ten ubiquitinated FAM134B RHD molecules (five K160–Ub and five K264–Ub) embedded in a closed tubule (up to approximately 8.5 μs). The Ub moieties initiated trans-interactions among the RHDs and enabled the formation of three protein clusters (dimers and trimers) on the MD timescale (Supplementary Video 5). These clusters deformed the tubule in both principal directions (squares in Extended Data Fig. 4g).

8Motivated by these locally bud-shaped structures, we studied the effect of ubiquitination on spontaneous membrane budding by simulating nine bi-mono-ubiquitinated RHDs ((K160 + K264)-Ub) embedded in POPC bilayers under four different simulation conditions (Fig. 2d). We increased the lipid-number asymmetry of the bilayer leaflets from ΔN = 0 to 300 to increase the energetic driving force for budding, and we reduced the protein–protein interaction (PPI) strength from 100% (α = 1) to 65% (α = 0.65) to weaken Ub-mediated protein clustering. Spontaneous budding was observed in the asymmetric bilayer with 100% PPI strength (Fig. 2d and Supplementary Video 6), triggered by the formation of a cluster of ubiquitinated RHDs on top of the nascent bud. After budding, the remaining ubiquitinated RHDs sorted onto the membrane bud to form a Ub-rich protein coat in the form of a ring around its neck. Ub–Ub interactions are characteristic of these RHD clusters (blue and cyan moieties in Fig. 2d). By contrast, in all the other three simulations, RHD clusters only formed transiently (because of reduced PPI strength) and induced at most local membrane bulges (because of the high-energy penalty for budding at low asymmetry, ΔN = 0). We conclude that the stabilization of RHD clusters by Ub-mediated interactions facilitates FAM134B-induced membrane budding.

9Analysis of all possible Ub–Ub interactions (Extended Data Fig. 5a–d) indicated that intermolecular or trans-Ub–Ub interactions triggered protein clustering and oligomerization (Extended Data Fig. 5c). Although changing the membrane asymmetry and PPI strength during the simulations did not change the intramolecular or cis-Ub–Ub interactions, they affected the character of intermolecular trans-Ub–Ub interactions (Extended Data Fig. 5a–d, left versus right). We also found that the membrane asymmetry enhanced the lifetime of the Ub–Ub interactions, stabilizing them to organize as a cluster (Extended Data Fig. 5e). Furthermore, whereas the intramolecular cis-interactions were predominantly mediated by specific residues on the Ub surface, namely, those forming hairpins β12, β34 and β5 (Extended Data Fig. 6a), the intermolecular trans-Ub–Ub interactions required for cluster formation were spread all over the Ub surface, indicating that no specific residues were dominant (Extended Data Fig. 6b). When the PPI strength was reduced to 65%, the alternative interactions that emerged were also spread over the surface, confirming the absence of specific interactions. Nonspecific Ub–Ub interactions create a crowded membrane environment with multiple RHDs, causing proteins to sort and aggregate locally in the membrane. These crowded regions appear to be driven by volume-exclusion effects and curvature-mediated protein-sorting mechanisms. Thus, volume exclusion in combination with multiple low-affinity nonspecific Ub–Ub interactions result in the nucleation of RHD clusters, which then induce membrane bud formation. The high curvature of the membrane bud stabilizes the RHD clusters, increases their longevity and further favours the sorting of individual Ub–RHDs to the site of the bud.

10Given that FAM134B functions in membrane remodelling, we hypothesized that ubiquitination affects FAM134B-driven ER fragmentation in cells. Using high-content imaging, we observed that the basal number and size of FAM134B–LC3B-decorated ER fragments were significantly lower in cells expressing the FAM134B 17KR mutant than in the WT control (Extended Data Fig. 7a–c). In vitro liposome remodelling driven by recombinant full-length GST-tagged FAM134B was not affected by the 17KR mutant, indicating that the folding and intrinsic activity of FAM134B were not impaired by the mutations (Extended Data Fig. 7d,e). Compared with liposomes treated with GST control, GST–FAM134B WT and 17KR decreased the liposome diameter to a similar extent (Extended Data Fig. 7f). Next, we investigated whether ubiquitination of the FAM134B RHD also affects the flux of ER-phagy using two validated reporter assays,,. First, we generated cells expressing inducible mCherry–eGFP-tagged FAM134B 17KR or its WT counterpart (Fig. 3a). As expected, the reporter localized to the ER and the mCherry signal was concentrated in microscale puncta corresponding to autophagosomes or lysosomes (Fig. 3b). The lack of FAM134B RHD ubiquitination significantly reduced the flux of ER-phagy following Torin 1 or Earle’s Balanced Salt Solution (EBSS) treatment (Fig. 3c). Similarly, the flux of ER-phagy was reduced when FAM134B WT was exposed to the E1 inhibitor TAK243 (Extended Data Fig. 7k). Second, in cells expressing the ER-phagy flux reporter ssRFP–GFP–KDEL, the flux of ER-phagy (basal and induced by Torin 1 or EBSS) was reduced in cells co-expressing HA–FAM134B 17KR compared with the WT version (Extended Data Fig. 7g–j). Accordingly, less EBSS-induced REEP5 degradation was observed in the ssRFP–GFP–KDEL/HA–FAM134B 17KR cells (Extended Data Fig. 7j).

11We next investigated the ultrastructure of FAM134B clusters by single-molecule localization microscopy in U2OS cells expressing HA–FAM134B WT or HA–FAM134B 17KR. Using DNA point accumulation in nanoscale topography (DNA-PAINT), we identified microscale and nanoscale clusters of HA–FAM134B (Fig. 3d). The microscale clusters corresponded to the puncta observed by confocal microscopy (Extended Data Fig. 7a). Furthermore, 2D and 3D high-resolution images showed that microscale HA–FAM134B clusters colocalized with the autophagosomal membrane marker LC3B-II (Extended Data Fig. 7l and Supplementary Video 7). We determined the size of HA–FAM134B WT and HA–FAM134B 17KR microscale clusters by Voronoi tessellation, indicating the area for the heterogeneous morphologies that we observed. Torin 1 treatment resulted in larger clusters for HA–FAM134B WT and the ubiquitination-deficient mutant (Fig. 3e). However, the cluster areas were significantly larger for HA–FAM134B WT than the mutant under control conditions and following exposure to Torin 1 (Fig. 3e and Extended Data Fig. 7m). Super-resolution images (28 ± 2 nm, mean ± s.d.) revealed the existence of HA–FAM134B nanoclusters much smaller than the resolution of light microscopy (Fig. 3d, dotted dashed circle). Co-labelling with REEP5 revealed that HA–FAM134B nanoscale clusters were distributed within the ER network (Extended Data Fig. 7n). Using DNA-PAINT super-resolution imaging, we found that a subset of HA–FAM134B nanoscale clusters colocalized with the autophagosomal membrane marker LC3B-II, and may represent ER-phagy initiation sites (Fig. 3f). Quantitative analysis of these sites using the DBSCAN clustering algorithm revealed that the diameter of the clusters containing ubiquitination-deficient HA–FAM134B was significantly lower (diameter = 79 nm) than those containing FAM134B WT (diameter = 101 nm) (Fig. 3g). We inferred the number of molecules in the nanoscale clusters by applying a kinetic analysis of single-molecule DNA-PAINT data, revealing that ubiquitination increased the oligomeric state of FAM134B WT in nanoclusters on average by fivefold (n17KR = 2.5, nWT = 12.6), thus promoting the assembly of high-density clusters (Fig. 3h). These data suggest that FAM134B RHD ubiquitination increases the size of ER-phagy initiation clusters, leading to larger autophagosomal structures that tune the dynamic flux of ER-phagy.

12To investigate the composition of the ER-phagy receptor complexes in more detail, we used a bimolecular complementation affinity purification (BiCAP) assay to visualize the PPIs and to characterize the cluster-specific interactome (Fig. 4a). The clustering of FAM134B was enriched by immunoprecipitation with anti-GFP antibodies following the co-expression of V1–FAM134B and V2–FAM134B (Extended Data Fig. 8a, compare lane 3 to lanes 2 and 1). Similar results were observed following the BiCAP of FAM134C, a paralogue that acts in concert with FAM134B (Extended Data Fig. 8a, compare lane 7 to lanes 6 and 5). The analysis of diGly peptides from isolated FAM134B oligomers revealed seven ubiquitinated lysine residues within the FAM134B RHD, including the four previously detected sites (Fig. 1b and Extended Data Fig. 8b). This indicated that multiple lysine residues in the RHD can be ubiquitinated within receptor clusters. Accordingly, FAM134B clusters (observed using BiCAP) were found in ER fragments, colocalizing with LC3B and Ub (Fig. 4b,c), indicating that ubiquitinated FAM134B clusters are colocalized with LC3+ autophagosomes. We characterized the interactome of immunopurified FAM134B clusters and detected 363 significantly enriched proteins (log2 enrichment factor ≥ 2, P ≤ 0.05), including novel candidates that had not been detected in previous FAM134B interactome datasets. MAP1LC3B and GABARAP were identified among the most enriched proteins (Fig. 4d, green dots). In addition, we found that FAM134B homodimers strongly interacted with several other RHDs and RHD-containing proteins (Fig. 4d, light blue dots, and Extended Data Fig. 8g) and with proteins of the ubiquitination machinery (Fig. 4d, red dots), including E3 ligases and deubiquitinases. Autophagy receptors containing Ub-binding domains (for example, p62, OPTN and TAX1BP1) were not detected, implying that FAM134B ubiquitination is not a recruitment signal for these proteins. Accordingly, there was no significant colocalization between FAM134B clusters and p62 in cells (Extended Data Fig. 8h,i). The interactomes of FAM134 isoforms, FAM134C heterodimers and FAM134B–FAM134C dimers revealed overlapping sets of interaction partners (Extended Data Fig. 8c–f). By contrast, the clustering of FAM134B 17KR reduced or abolished interactions between FAM134B and several RHD-containing proteins, mammalian ATG8 proteins and the ubiquitination machinery (Fig. 4e), indicating that ubiquitination promotes the formation of multimeric ER-phagy clusters.

13The interaction partners within oligomeric FAM134B clusters included the endogenous ER-anchored E3 ligase AMFR (also known as gp78), which is implicated in ER-associated degradation (Fig. 4d). This colocalized with FAM134B clusters in BiCAP experiments and with LC3B (Extended Data Fig. 9a,b). The expression of WT AMFR, but not its catalytically inactive counterpart (C356G H361A), increased the ubiquitination of HA–FAM134B in cells (Extended Data Fig. 9c,d). Evaluating their functional interaction in cells using BiCAP assays (Extended Data Fig. 9e) revealed that V2–AMFR–V1–FAM134B complexes colocalized with LC3B and Ub+ structures (Extended Data Fig. 9f). This effect was reduced when AMFR was replaced with its catalytically inactive mutant (Extended Data Fig. 9f,g). MS-based interactome analysis confirmed that FAM134B–AMFR complexes also recruited RHD-containing proteins, mammalian ATG8 proteins and components of the ubiquitination machinery (Extended Data Fig. 9h,i). Of note, the E2-conjugating enzyme UBE2G2, which cooperates with AMFR, was also enriched (Extended Data Fig. 9h). The Torin 1-induced ubiquitination profile of FAM134B complexes with AMFR (WT V2–AMFR–V1–FAM134B) was significantly reduced in complexes with the catalytically inactive AMFR mutant (V2–AMFR(C356G,H361A)–V1–FAM134B) (Extended Data Fig. 9j). In addition, clustering of FAM134B with catalytically inactive AMFR reduced the interactions with other RHD-containing proteins, mammalian ATG8 proteins and the ubiquitination machinery (Extended Data Fig. 9k). Silencing the expression of AMFR caused a significant increase in total endogenous FAM134B (Extended Data Fig. 10a,b) and diminished its cellular turnover induced by Torin 1 (Fig. 5a,b). Furthermore, we observed decreased ubiquitination of S protein-FLAG-streptavidin-binding peptide (SFB)-tagged FAM134B in response to AMFR depletion (Fig. 5c), also reducing FAM134B-mediated ER fragmentation in response to Torin 1 (Extended Data Fig. 10c,d). Torin 1 treatment also reduced the levels of endogenous AMFR (Fig. 5a and Extended Data Fig. 10e). This decay was diminished in the presence of HA–FAM134B 17KR (Extended Data Fig. 10f–h) or ΔLIR (Extended Data Fig. 10i,j), indicating that active FAM134B and its ubiquitination promote efficient AMFR degradation via ER-phagy. Moreover, siRNA-mediated knockdown of AMFR significantly slowed the flux of FAM134B-mediated Torin 1-induced ER-phagy compared with control siRNA (siNT) (Fig. 5d; approximately 58% and 60% reduction with AMFR-targeting siRNA #1 (siAMFR #1) and siAMFR #2, respectively).

14Next, we tested whether AMFR-mediated ubiquitination affected the membrane-shaping functions of FAM134B in vitro. First, we ubiquitinated GST-tagged FAM134B using purified recombinant AMFR (Extended Data Fig. 10k,l) and detected K160, K278 and K299 as direct targets of AMFR-dependent ubiquitination (Extended Data Fig. 10m,n). This prompted us to repeat the in vitro liposome remodelling assay in the presence of AMFR. Compared with liposomes treated with GST, the non-ubiquitinated GST–FAM134B (in the presence of AMFR, no ATP) decreased the liposome diameter, but ubiquitinated GST–FAM134B (in the presence of AMFR + ATP) reduced the diameter even further (Fig. 5e). This significant difference between non-ubiquitinated and ubiquitinated samples suggests that ubiquitination of multiple sites on full-size FAM134B promotes the formation of smaller liposomes. This agrees with the increased membrane remodelling activity for the chimaera Ub–RHD90–264–Ub compared with non-ubiquitinated RHD90–264 (Fig. 2c). Furthermore, the ubiquitination of FAM134B RHD90–264 in vitro by AMFR promoted the formation of larger complexes detected by blue native polyacrylamide gel electrophoresis (Extended Data Fig. 10o–r).

15Our data reveal an unprecedented role for Ub in the conformational changes of FAM134B that, in turn, drive ER membrane remodelling and the flux of ER-phagy. Although FAM134B RHDs inherently curve bilayers and populate regions of high local membrane curvature, the addition of Ub boosts this effect by promoting multiple cis-interactions and trans-interactions that stabilize multimeric clusters. This enables the clusters to grow in size and to nucleate large-scale membrane remodelling events in the ER (Fig. 5f). ER-phagy clusters also include multiple RHD-containing proteins such as ARL6IP1 (a recurrent finding in the interactome of FAM134B homodimers), which is also required for efficient ER-phagy. In addition, the clusters contain Ub ligases and deubiquitinases, which can alter the dynamic ubiquitination of RHD-containing proteins and influence the formation and growth of these multivalent clusters. Indeed, the E3 ligase AMFR regulates ER-phagy by acting as a critical ER quality control mechanism. It is tempting to speculate that other E3 ligases can modify ER-phagy receptor clusters in a cell-type-specific manner. The receptor clustering phenomenon depends on several interrelated factors, such as receptor abundance, distribution and other post-translational modifications (for example, phosphorylation or UFMylation). Studying these events will shed light on the dynamics of the entire ER-phagy pathway and will pave the way for a better understanding of defects in ER dynamics influencing the pathogenesis of many diseases.

16Cell lines were cultured in DMEM supplemented with 10% heat-inactivated FBS and penicillin–streptomycin (Thermo Fisher Scientific) in a 5% CO2 atmosphere at 37 °C. U2OS TRex cells (provided by S. Blacklow, Brigham and Women’s Hospital and Harvard Medical School) and U2OS cell lines (American Type Culture Collection (ATCC)) were used to generate inducible cell lines based on lentiviral infection. The mCherry–eGFP–FAM134B WT and mCherry–eGFP–FAM134B 17K constructs were introduced into the vector pcDNA5/FRT/TO using GATEWAY technology and transfected with the recombinase vector pOG44 into Flp-In U2OS TRex cells. Following selection with 300 µg ml−1 hygromycin, the resistant cells were expanded. We generated U2OS cell lines expressing HA–FAM134B or HA–FAM134B 17KR under the control of a doxycycline-inducible promoter. Expression was induced with 0.5 μg ml−1 doxycycline for the indicated time. Lentiviruses were produced in HEK 293T cells (ATCC). In brief, HEK 293T (ATCC) cells were co-transfected with the lentiviral plasmids (1.1 µg complementary DNA (cDNA)), containing the cDNA of FAM134B WT or 17KR along with the two packaging vectors pPAX2 (2.2 µg cDNA) and pMD2.G (1 µg cDNA) using Turbofect reagent (Thermo Fisher Scientific). The lentivirus-containing medium was collected after 48 h. The medium was centrifuged to remove dead HEK 293T cells and stored at −80 °C. After 24 h, cells were selected in fresh DMEM containing 3 µg ml−1 puromycin. Cells were maintained in the presence of antibiotics for selection and were grown to subconfluence before each experiment. Cells were treated with 0.5 µg ml−1 doxycycline (Sigma-Aldrich), 200 ng ml−1 bafilomycin A1 (LC Laboratories), 50 µg ml−1 cycloheximide (AppliChem PanReac), 250 nM Torin 1 (LC Laboratories) and/or 1 µM or 10 µM TAK243 (MedChemExpress) for the indicated time periods. Transient transfection was carried out using Turbofect reagent. All cell lines were regularly tested for mycoplasma contamination using the LookOut Mycoplasma PCR Detection Kit (Sigma).

17Human AMFR-targeting siRNAs (siRNA #1 5′-GCA AGG AUC GAU UUG AAU A-3′; and siRNA #2, 5′-GUA AAU ACC GCU UGC UGU G-3′) were purchased from Dharmacon. A non-targeting siRNA was used as a control (Qiagen). Transfection with siRNA was carried out using Lipofectamine RNAiMAX (Thermo Fisher Scientific). Assays were carried out 72 h post-transfection.

18Primary and secondary antibodies are presented in Supplementary Table 1.

19We introduced the cDNAs into vector pDONR223 using the BP Clonase Reaction Kit (Invitrogen) followed by transfer to GATEWAY destination vectors using the LR Clonase Reaction Kit (Invitrogen), resulting in vectors pLTD-N-HA-PURO, pcDNA5-FRT/TO-N-mCherry-eGFP, pcDNA3.1-N-HA, pMH-SFB, pGEX6P1-DEST, pDEST 527 6×His, pDEST-V1ORF, pDEST-V2-ORF, pDEST-ORF-V1 and pDEST-ORF-V2. GST–FAM134B was generated by subcloning into the vector pGEX-6P1 using the EcoRI site. GST–FAM134B 17KR was similarly generated by targeting the EcoRV and SmaI sites of pGEX-6P2.

20Plasmids are presented in Supplementary Table 2.

21To assess the ubiquitination status of FAM134B, HEK 293T cells transfected with Myc–Ub, HA–FAM134B constructs and WT AMFR–Flag or its catalytically inactive mutant as indicated were lysed (50 mM Tris-HCl pH 7.5, 150 mM NaCl, 0.5 mM ethylenediaminetetraacetic acid (EDTA), 1% Triton X-100, 10 mM N-ethylmaleimide (NEM) and Roche EDTA-free protease inhibitors, freshly added). The lysates were incubated for 15 min on ice and centrifuged (12,000g at 4 °C for 30 min) before 20 µl of the supernatant was supplemented with Laemmli sample buffer, boiled for 5 min at 95 °C and stored at –20 °C as the input control. Ubiquitinated proteins (Myc–Ub proteins) were immunoprecipitated from soluble extracts with Myc-Trap agarose (Chromotek). Beads were washed three times in lysis buffer and proteins were denatured by heating at 95 °C for 5 min before SDS–PAGE and western blot analysis with anti-HA antibodies for the detection of HA–FAM134B constructs. For other immunoprecipitation assays, cleared lysates were incubated with GFP-Trap (Chromotek), HA-agarose beads (Sigma-Aldrich) or TUBE-2 agarose beads (Life Sensors) at 4 °C overnight. The next day, tubes were centrifuged (800 rcf at 4 °C for 5 min) to sediment the beads, and the supernatant was removed and washed with ice-cold lysis buffer. The inputs and co-precipitation fractions were analysed by SDS–PAGE and western blot.

22Blue-native PAGE was performed following the manufacturer’s instructions (#BN1001BOX, protocol pub. no. MAN0007893 Rev. A.0, Life Technology). For SDS–PAGE, proteins were denatured at 90 °C for 5 min in Laemmli buffer. Native or denatured proteins were transferred to methanol-activated polyvinylidene fluoride membranes (Amersham Hybond P, 0.45 µm). Membranes were blocked for 1 h in 5% skimmed milk in PBS containing 0.1% Tween-20 and incubated overnight at 4 °C with the specified primary antibody, and for 2 h with the corresponding secondary antibody at room temperature. We used horseradish peroxidase-conjugated anti-rabbit (1:10,000), anti-mouse (1:10,000) and anti-rat (1:10,000) secondary antibodies as appropriate, followed by signal development using Western ECL substrate (sc-2048, SantaCruz) and the Chemidoc automated detection system (Bio-Rad). Densitometric quantification of western blot bands was carried out using ImageJ (version 1.51w).

23Quantifications of western blot signals were performed using ImageJ software (version 1.51w). For each assay, protein bands were quantified from at least three independent experiments. Data analysis was performed using Microsoft Excel 2016 (Microsoft Corporation) or Prism 9.4.1 (GraphPad Software). Results were graphed as the mean ± standard deviation. Statistical significance was determined by one-tailed, paired t-test. P > 0.05 or P ≤ 0.05 were regarded as not statistically significant or statistically significant, respectively. One-way analysis of variance (ANOVA) was used in assays with two independent variables. Bonferroni’s multiple comparison test was performed.

24Cells were grown on 12-mm glass coverslips or in 96-well plates. Cells were washed with PBS before fixation in 4% paraformaldehyde (PFA) for 10 min at room temperature. After several further washes with PBS, cells were permeabilized with 0.25% saponin in PBS and blocked with 5% FBS for 1 h at room temperature. Primary antibodies were diluted in 5% FBS/0.25% saponin in PBS and incubated overnight at 4 °C. After three PBS washes, secondary antibodies were added and incubated for 1 h at room temperature. The cells were washed another three times before staining with the nuclear dye DAPI for 10 min, and washed again before mounting with Fluoromount-G. The following secondary antibodies were used in a 1:1,000 dilution: anti-rabbit Alexa 488, anti-mouse Alexa 546 and anti-rat Cy5. Slides were imaged using a Leica SP8 confocal microscope fitted with a ×63 oil-immersion lens and analysed with ImageJ software (version 1.51w). For ER fragmentation and ER-phagy flux assays, doxycycline-inducible mCherry–GFP–FAM134B U2OS cells were analysed using a Yokogawa CQ1 (vR1.08.01) confocal imaging cytometer.

25Escherichia coli (BL21(DE3)) cells were transformed with constructs encoding GST-tagged FAM134B (WT, 17KR mutant, RHD90–264 or Ub–RHD90–264–Ub), GST–Ub and GST, followed by expression and purification as previously described,. In brief, transformed cells were grown in the presence of appropriate antibiotics overnight and the primary culture was used to inoculate the main culture. When the OD600 nm reached 0.6, 0.25 mM isopropyl β-d-1-thiogalactopyranoside (IPTG) was used to induce protein expression for 16 h at 18 °C, shaking at 180 rpm. After induction, the cells were harvested and the pellets were suspended in ice-cold PBS, followed by sonication and centrifugation (10,000g for 30 min at 4 °C). The supernatant was then fractionated by ultracentrifugation (80,000g for 90 min at 4 °C). RHD fusion proteins were recovered by dissolving the pellets in PBS containing 0.05% dodecyl β-d-maltoside (DDM) and loading onto glutathione-Sepharose TM4 Fast Flow columns (GE Healthcare). The columns were washed with PBS containing 0.05% DDM and the proteins were eluted in PBS containing 15 mM reduced glutathione and 0.025% DDM. GST–Ub and GST were directly loaded from the centrifugation step after sonication and were purified on the GST column. The eluted fractions were concentrated and exchanged with storage buffer (50 mM HEPES pH 7.5, 150 mM NaCl and 0.0075% DDM).

26Alternatively, cells expressing GST–FAM134B WT and 17KR were lysed in a French press (G. Heinemann Ultraschall & Labortechnik) to avoid protein denaturation by sonication. Bacterial lysates were mixed with 2% (v/v) Triton X-100, 20 mM MgCl2, 40 µg ml−1 DNAse I and 4 mg ml−1 lysozyme for 30 min at 1 °C, and cell debris was removed by centrifugation (1,000g for 20 min at 4 °C). The supernatant was incubated with ice-cold PBS, and GST fusion proteins were affinity purified on glutathione-Sepharose resin (Genscript). After washing, bound proteins were eluted by incubation with 20 mM glutathione in 50 mM Tris-HCl pH 8.0 and 120 mM NaCl. The purified proteins were concentrated using Amicon Ultra-4-10k centrifugal filters (Millipore) and dialysed at 4 °C against HN-buffer (20 mM HEPES/KOH pH 7.4, 150 mM NaCl and 2.5 mM DTT) for liposome and freeze-fracture assays.

27For the purification of His–RHD90–264–Strept-II and His–Ub–RHD90–264–Ub–Strept-II, bacterial pellets were resuspended in ice-cold binding buffer (100 mM Tris-HCl pH 8.0, 150 mM NaCl, 1 mM EDTA and 1 mM TCEP) plus a cocktail of protease inhibitors (1 mM PMSF, 1 µM GM6001, 0.25 µM bestatin, 0.5 µM pepstatin and 1 µM E-64), DNAse I (50 µg ml−1) and 0.1% DDM. The cells were disrupted by two passes through a microfluidizer at 1,500 bar, and the debris was removed by centrifugation (12,000g for 1 h at 4 °C). The supernatant, including the membranes, was centrifuged (43,000g for 2.5 h at 4 °C) on an Ultracentrifuge Optima L-90K with a 45 Ti rotor (Beckman Coulter). The pelleted membrane was solubilized in membrane extraction buffer (100 mM Tris-HCl pH 8.0, 300 mM NaCl, 1 mM EDTA, 25% glycerol and 2% DDM) at 4 °C with gentle stirring for approximately 2 h. The insoluble fractions were removed by ultracentrifugation (55,000g for 1 h at 4 °C) and the supernatant containing the solubilized membrane proteins was passed through a 0.22-µM filter and supplemented with 40 µg ml−1 avidin before purification. The first purification step targeted the strep II tag. A Hiprep strep II 5-ml column (Cytiva) on an Äkta FPLC system was pre-equilibrated with binding buffer supplemented with 0.1% DDM. The sample was allowed to bind to the column and then washed with binding buffer containing 0.01% DDM and lacking EDTA. Fractions in elution buffer (100 mM Tris-HCl pH 8.0, 150 mM NaCl, 0.03% DDM and 3 mM d-desthiobiotin) were analysed by SDS–PAGE and western blot with antibodies specific for the strep II tag. The desired fractions were pooled and supplemented with 25 mM imidazole followed by second-step purification targeting the His6 tag. Talon beads were used and a linear gradient between buffer A (50 mM Tris-HCl pH 7.5, 500 mM NaCl, 25 mM imidazole, 5% glycerol, 0.03% DDM and 1 mM TCEP) and buffer B (buffer A with 250 mM imidazole) was used to purify the protein. The resulting protein was buffer exchanged (25 mM Tris-HCl pH 7.5, 100 mM NaCl, 5% glycerol, 1 mM TCEP and 0.03% DDM) using a centricon filter with a 10-kDa cut-off. The protein was aliquoted and flash frozen in liquid nitrogen for further experiments.

28For AMFR expression in mammalian cells, 3 l cultures of suspension-adapted HEK 293T cells were grown in customized DMEM to a density of 1 × 106 cells per millilitre. We added 15% (v/v) P3 baculovirus to each flask and incubated on a shaking platform for 24 h, after which the temperature was reduced to 30 °C and 10 mM sodium butyrate was added. Cells were harvested after shaking for 48 h and the pellets were frozen at –80 °C. For AMFR protein purification, approximately 15 g of cells was thawed on ice and resuspended in 60 ml lysis buffer (20 mM HEPES pH 7.5, 300 mM NaCl, 5 mM DTT, Roche protease inhibitor cocktail, 1 mM MgCl2 and 0.001 mg ml−1 Benzonase) and sonicated on ice. To solubilize membranes, DDM and CHS was added to the lysate at a final concentration of 1% and 0.1%, respectively, and stirred for 1 h at 4 °C. Insoluble material was removed by centrifugation (40,000g for 30 min at 4 °C). The supernatant was incubated with 2 ml StrepTactin resin (Cytiva) for 2 h at 4 °C to bind to gp78-TEV-TwinStrepII. The resin was poured into a column and washed with 24 column volumes (CV) of wash buffer (20 mM HEPES pH 7.5, 300 mM NaCl, 5 mM DTT and 0.05/0.005% DDM/CHS). Bound protein was eluted with wash buffer containing 5 mM desthiobiotin. To remove the C-terminal affinity tag, eluted protein was incubated for 2 h with 0.1 mg ml−1 TEV protease at room temperature. Finally, pooled elution fractions were concentrated to 0.5 ml using 100-kDa cut-off centrifugal concentrators (Amicon) and purified by gel filtration on a Superose 6 10/300 column (Cytiva) running with SEC buffer (20 mM HEPES pH 7.5, 100 mM NaCl, 5 mM DTT and 0.05/0.005% DDM/CHS).

29Liposomes made from synthetic lipids were prepared as previously described. In brief, 2-dioleoyl-sn-glycero-3-phosphocholine (DOPC) and 1,2-dioleoyl-sn-glycero-3 phosphoethanolamine (DOPE), both from Avanti Polar Lipids, were dissolved in a mixture of chloroform and methanol (4:1) in a round-bottom flask at a molar ratio of 0.8:0.2 (DOPC:DOPE). The organic solvent was removed by rotary evaporation to obtain a uniform dry lipid film, which was then hydrated for 2 h at room temperature with liposome buffer A (50 mM HEPES pH 7.4 and 150 mM NaCl) to obtain a final 15 mg ml−1 solution. Liposomes were dissolved by vortexing followed by sonication in an ultrasound bath. Liposomes were equilibrated to 25 °C and extruded using a lipid extruder with 200-nm polycarbonate membranes (Avanti Polar Lipids).

30For negative staining assays, 2.5 µM FAM134B WT and FAM134B 17KR mutant or GST were incubated with 1 mg ml−1 liposomes in liposome buffer B (50 mM HEPES pH 7.4, 150 mM NaCl and 0.001% DDM) for 18 h at 25 °C on a table-top shaker at 600 rpm. GST–FAM134B RHD90–264 or GST–Ub–RHD90–264–Ub chimaera and Ub–GST were incubated at 0.5 µM with 1 mg ml−1 liposomes in liposome buffer B for 8 h. We then added 5 μl of each sample to the carbon-coated copper grids (SPI Supplies) without glow discharge. After 1 min, the grids were washed twice with water and stained with 1% uranyl formate for 1 min at room temperature. Excess solution was removed by blotting with filter paper. Approximately 20 micrographs were recorded for each sample using a 120 kV Tecnai Spirit Biotwin electron microscope (FEI) equipped with a 4k × 4k CCD detector (US4000-1, Gatan).

31Alternatively, membrane shaping by full-length GST–FAM134B ubiquitinated with AMFR was investigated by transmission electron microscopy using freeze-fractured liposomes prepared from Folch-fraction type I lipids (Sigma-Aldrich) as previously described. We incubated 1 mg of liposomes with 2.5 μM protein in HN-buffer (20 mM HEPES/KOH pH 7.4, 150 mM NaCl and 2.5 mM DTT) containing 0.3 M sucrose for 15 min at 37 °C. Subsequently, 15 μg proteinase K was added and incubated for 40 min at 45 °C to avoid liposomal aggregates. Small aliquots (1–2 μl) of the liposome suspension were then freeze-fractured,. The samples were examined by systematic grid exploration using an EM 900 electron microscope (Zeiss) at 80 kV. Images were acquired using a wide-angle dual-speed 2K CCD camera (Tröndle). Diameters of liposomes were determined using ImageJ software (version 1.51w).

32AMFR-mediated ubiquitination assays were based on a modified in vitro ubiquitination assay (Abcam). In brief, purified substrate (5 µM full-length GST–FAM134B, 1 µM His–RHD90–264–Strept-II or 1 µM His–Ub–RHD90–264–Ub–Strept-II) was mixed with 10 µM Ub, 10 mM ATP and 10 mM MgCl2 in 50 mM Tris-HCl pH 7.5, 150 mM NaCl, 0.8 µM E3-ligase AMFR, 100 nM E1 UBA1 and 0.8 µM E2 of AMFR UBE2G2 (Biotechne) for 2 h at 37 °C. The reaction mixture was analysed by SDS–PAGE or western blotting with antibodies against GST, His6 or Ub(P4D1). Alternatively, the samples were prepared for MS. Samples were incubated with SDC buffer (1% sodium deoxycholate, 0.5 mM TCEP, 2 mM chloroacetamide and 50 mM Tris-HCl pH 8.5) and heated to 60 °C for 30 min. We then added 500 ng trypsin to each sample and incubated overnight at 37 °C. The reaction was stopped with 1% TFA in isopropanol. Peptides were cleaned up using SDB-RPS stage tips (Sigma-Aldrich). After one wash with 1% TFA in isopropanol and one wash with 0.2% TFA in water, peptides were eluted in 80% acetonitrile plus 1.25% ammonia. Eluted peptides were dried and processed for LC–MS.

33HEK 293T cells were transiently co-transfected with the constructs V1–FAM134B WT and V2–FAM134B WT, V1–FAM134C WT and V2–FAM134C WT, V1–FAM134B WT and V2–FAM134C or V1–FAM134B 17KR and V2–FAM134B 17KR. After 16 h, cells were lysed with 1% Triton X-100 in lysis buffer (50 mM Tris-HCl pH 7.5, 150 mM NaCl, 0.5 mM EDTA, Roche EDTA-free protease inhibitor cocktail and NEM), followed by incubation with GFP-trap beads (Chromotek) overnight at 4 °C on a rotating platform. Protein-bound beads were washed three times with lysis buffer and three times in the same buffer without detergents before on-bead trypsin digestion. Samples were incubated with 25 µl SDC buffer (2% sodium deoxycholate, 1 mM TCEP, 4 mM chloroacetamide and 50 mM Tris-HCl pH 8.5) and heated to 60 °C for 30 min. We then added 500 ng trypsin in 25 µl 50 mM Tris-HCl (pH 8.5) to each sample and incubated overnight at 37 °C. The reaction was stopped with 150 µl of 1% TFA in isopropanol. Peptides were cleaned up using SDB-RPS stage tips (Sigma-Aldrich). After one wash with 1% TFA in isopropanol and one wash with 0.2% TFA in water, peptides were eluted in 80% acetonitrile plus 1.25% ammonia. Eluted peptides were dried and processed for LC–MS.

34Dried peptides were reconstituted in 2% acetonitrile containing 0.1% TFA and analysed on a QExactive HF mass spectrometer coupled to an easy nLC 1200 (Thermo Fisher Scientific) fitted with a 35-cm, 75-µm ID fused-silica column packed in house with 1.9-µm C18 particles (Reprosil pur, Dr. Maisch). The column was maintained at 50 °C using an integrated column oven (Sonation). Peptides were eluted in a non-linear gradient of 4–28% acetonitrile over 45 min and directly sprayed into the mass spectrometer equipped with a nanoFlex ion source (Thermo Fisher Scientific). Full-scan MS spectra (300–1,650 m/z) were acquired in profile mode at a resolution of 60,000 at m/z 200, a maximum injection time of 20 ms and an AGC (automatic gain control) target value of 3 × 106. Up to 15 of the most intense peptides per full scan were isolated using a 1.4-Th window for fragmentation by higher energy collisional dissociation (normalized collision energy of 28). MS/MS spectra were acquired in centroid mode with a resolution of 30,000, a maximum injection time of 45 ms and an AGC target value of 1 × 105. Single charged ions, ions with a charge state of more than four and ions with unassigned charge states were not considered for fragmentation, and dynamic exclusion was set to 20 s to minimize the acquisition of fragment spectra representing already acquired precursors.

35MS raw data were processed using MaxQuant v1.6.17.0 with default parameters. Acquired spectra were searched against the human ‘one sequence per gene’ database (Taxonomy ID 9606) downloaded from UniProt (12 March 2020; 20,531 sequences), and a collection of 244 common contaminants (‘contaminants.fasta’ provided with MaxQuant v1.6.17.0) using the Andromeda search engine integrated into MaxQuant v1.6.17.0 (ref.,). Identifications were filtered to obtain false discovery rates below 1% for both peptide spectrum matches (minimum length of seven amino acids) and proteins using a target–decoy strategy. Protein quantification and data normalization relied on the MaxLFQ algorithm implemented in MaxQuant v1.6.17.0 (ref. ). The MS proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifiers PXD032721 (Fig. 1a,b), PXD032740 (Extended Data Fig. 8b), PXD032741 (Fig. 4d and Extended Data Fig. 8c–e), PXD032743 (Fig. 4e), PXD032750 (Extended Data Fig. 9h–k), PXD039186 (Extended Data Fig. 10m,n), PXD039187 (Fig. 5c) and PXD039188 (Extended Data Fig. 10p,q). For protein assignment, spectra were correlated with the UniProt human database v2019 including a list of common contaminants. Searches were performed with tryptic specifications and default settings for mass tolerances in MS and MS/MS spectra. Carbamidomethyl cysteine, methionine oxidation and N-terminal acetylation were defined as fixed modifications. The match-between-run feature was used with a time window of 1 min. For further analysis, Perseus v2.0.7.0 was used and first filtered for contaminants and reverse entries as well as proteins that were only identified by a modified peptide.

36The previously built molecular model of FAM134B RHD was extended to include an additional ten residues at the C terminus of the RHD (residues 261–270). Isopeptide bonds between the lysine (K160 and K264) and the terminal glycine of Ub (G76) were modelled by modifying the side-chain lysine bead (SC2/+1) into the neutral backbone bead (BB/0) and restraining its distance to the terminal bead of Ub to 0.35 nm with a force constant k = 1,250 kJ mol−1. Two mono-ubiquitinated and one bi-mono-ubiquitinated RHD structures were modelled (K160–Ub, K264–Ub and (K160 + K264)-Ub, respectively).

37Coarse-grained (CG) MD simulations were prepared using the MARTINI model (v2.2),. Initial CG structures were built using martinize.py. Assignments from DSSP (Dictionary of Secondary Structures in Proteins) program were used to generate backbone restraints to preserve local secondary structure,. CG models were embedded into POPC (16:0–18:1 phosphatidylcholine (PC)) bilayers spanning the periodic simulation box in the x–y plane. Initial configurations for each system were assembled and then solvated with CG water containing 150 mM NaCl using insane.py. Each system was energy minimized and equilibrated using the Berendsen thermostat and barostat along with position restraints on protein backbone beads, followed by production runs with a 20-fs time step. System temperature and pressure during the production phase were maintained at 310 K (unless otherwise stated) and 1 atm with the velocity rescaling thermostat and the semi-isotropic Parrinello–Rahman barostat, respectively. All simulations were performed using gromacs (v2019.3),. Long-lived, highly populated RHD conformations were obtained after clustering evenly sampled conformations (n = 10,000) from each trajectory. Clusters were obtained using backbone root-mean-square deviation (cut-off = 0.8 nm) by using the gromos method, as implemented in the gmx_cluster tool.

38Discontinuous bicelle systems containing saturated DMPC (14:0 PC) and DHPC (7:0 PC) lipids were assembled as previously described,. The equilibrated ubiquitinated and non-ubiquitinated RHD molecules obtained from simulations in the POPC bilayers (after 5 μs) were then embedded in the bicelle and solvated. One hundred replicates for each system were simulated with different initial velocities at 300 K to obtain statistics on the transition times to vesicles. Shape transformations from flat bicelles (H = 0 nm−1) to curved vesicles (H = 0.15 nm−1) were monitored by measuring the signed membrane curvature (H(t)). Lipid coordinates were fitted to spherical surfaces using least squares optimization to compute membrane curvature along simulations using MemCurv. Curvature away from and towards the upper/cytoplasmic leaflet are reported as positive and negative values, respectively. The statistics of waiting times (t) for the formation of vesicles (bilayer curvature, |H| > 0.15 nm−1) for the three systems were determined from individual replicates. We modelled the kinetics of the bilayer-to-vesicle transition using a Poisson process with a lag time (t = t′ + τ). The time t′ = 1/k′ describes the Poisson process with rate k′. The constant lag time τ captures the time required for vesicle closure from the curved bilayer disc. The distributions of waiting times are thus p(t) = k′ e−k′(t − τ) for t > τ. We determined the rate of vesicle formation (k = 1/(t′ + τ)) for different systems, from fitting the cumulative distribution function for the probability density, P(t – τ) = ke−k(t – τ) corresponding to p(t), to the observed waiting time distributions estimated from replicates. We used the previously computed maximum likelihood estimates of the vesiculation rate for non-ubiquitinated RHD bicelles (kRHD = 0.0018 ns−1) to compute the acceleration factors for each system (acc = ksys/kRHD). Furthermore, to show the effect of temperature on the energy barrier to form closed vesicles from flat bicelles, we also simulated 20 replicates of each system at 280 K and estimated the number of successful vesicle closure events within the simulation timescale.

39A CG POPC bilayer was used to tile a tessellated buckled surface using LipidWrapper. The buckled membrane was solvated with CG water and ions and equilibrated in a periodic box with a fixed x–y plane (57 × 28 nm2) and excess membrane area (approximately 17 nm2). This preserved the buckled shape of the bilayer, offering a range of curvature values to be sampled by embedded proteins (H(x,y) = −0.05 ≤ 0 ≤ 0.05 nm−1). Ubiquitinated and non-ubiquitinated RHDs (in separate simulations) were initially embedded in regions with small local curvature (H(x, y) ≃ 0). Following another equilibration phase with proteins, the systems were simulated for more than 20 μs at 310 K. The membrane profiles of the buckled surface with and without embedded proteins were analysed by using the Monge representation to compute local principal curvatures k1(x, y) and k2(x, y), the mean curvature H(x, y), and the Gaussian curvature KG(x, y) as implemented in MemCurv.

40The clustering of ubiquitinated RHD molecules was simulated under periodic boundary conditions using buckled, tubular and flat membrane structures. MD simulations were initiated after embedding one K160–Ub RHD and one K264–Ub RHD into a single membrane buckle. Initially, the two proteins were placed more than 20 nm apart. The two proteins were tracked by measuring their minimum distance to identify initial contact and cluster formation. Once the initial contacts were made and dimers were formed, residue pairwise interactions were mapped across the tethered Ub moieties (K160-Ub and K264-Ub) to identify specific Ub–Ub trans-contacts. We then obtained the configuration of a closed membrane tubule (length of approximately 97–100 nm; diameter of approximately 12–15 nm) used in our previous work. Ten ubiquitinated RHD molecules were embedded along the length of the tubule such that the individual proteins were spaced maximally away from each other. The RHD-containing tubule was equilibrated in explicit solvent (approximately 3.6 × 106 beads) under NVT (constant volume, constant temperature ensemble) and NPT (constant pressure, constant temperature ensemble) conditions along with position restraints on protein backbone beads. A production run of 10 μs was carried out, after releasing position restraints to observe curvature-mediated protein sorting and the formation of RHD clusters. We also generated initial configurations of nine (K160 + K264)–Ub–RHDs on a 3 × 3 square grid embedded in POPC bilayers such that each protein was separated by its nearest neighbour by 10 nm. We changed the lipid-number bilayer asymmetry from ∆N = 0 to ∆N = 300 using the insane.py script. Following a previously implemented method, we scaled the protein−protein LJ pair interaction well depth,. A value of α = 0.65 corresponds to a reduction in PPI strength in the MARTINI model and a value of α = 1.0 recovers the full interaction in the MARTINI forcefield, ϵ1 = ϵoriginal. The resulting 2 × 2 = 4 bilayer configurations were then solvated with CG water containing 150 mM NaCl. We embedded the ubiquitinated and non-ubiquitinated RHD proteins in the asymmetric membranes (30 × 30 × 20 nm3) in a square grid and energy minimized the systems using a soft-core potential and steepest-decent algorithm to remove steric clashes with lipids. Production runs for 5 μs at NPT conditions were carried out for each simulation condition, in which position restraints were released to observe curvature-mediated protein sorting effects and Ub–Ub interactions, leading to the formation of RHD clusters. We monitored the box dimensions over time, which acted as proxies to indicate spontaneous budding. In addition, we tracked the z-coordinates of the centre-of-mass (COM) of all nine proteins along with the lowest and the highest points of the PO4 beads comprising the POPC bilayer. We also computed the distance matrix specifying all inter-RHD distances computed from COM positions and used hierarchical clustering with single linkage and a cut-off of 10 nm to obtain the distinct protein clusters in each frame of the simulation. We quantified the number of clusters and their sizes to identify the largest cluster for each simulation frame.

41For exchange DNA-PAINT experiments, donkey anti-rabbit antibodies (#711-005-152, AffiniPure) and goat anti-mouse antibodies (#115-005-003, AffiniPure) were covalently labelled with the short DNA-docking strands anti-R2 (5′-ACCACCACCACCACCACCA-3′) and anti-R1 (5′-TCCTCCTCCTCCTCCTCCT-3′), respectively, using DBCO-sulfo-NHS ester chemistry. In brief, concentrated secondary antibodies were incubated with a 20-fold molar excess of DBCO-sulfo-NHS ester (Jena Bioscience). After removing excess reagent, azide-functionalized DNA-docking strands were added at a tenfold molar excess and incubated overnight at 4 °C. Unbound DNA was removed from the samples concomitant with storage buffer exchange (PBS) using Amicon centrifugal filters (100 kDa cut-off). Antibody–DNA complexes were concentrated to 5 mg ml−1 and stored at 4 °C.

42For the exchange DNA-PAINT experiments, U2OS cells were seeded in Ibidi µ-slide VI chambers at 70% confluency. The cells were fixed for 30 min with pre-warmed (37 °C) 4% methanol-free formaldehyde (Sigma-Aldrich) in PBS followed by three washes with PBS. Fixed cells were then incubated in permeabilization/blocking buffer (10% FBS and 0.1% saponin) containing primary antibodies for 60 min at room temperature: rabbit anti-FAM134B (Genscript), mouse anti-LC3B or mouse anti-REEP5, each diluted 1:200. Excess primary antibodies were removed from the chambers by three washes with PBS. Cells were then incubated with custom DNA-labelled secondary antibodies in the permeabilization/blocking buffer for 60 min, followed by the removal of free antibodies by three washes with PBS. Finally, samples were post-fixed with 4% methanol-free formaldehyde in PBS for 10 min at room temperature, followed by three washes with PBS. For the exchange DNA-PAINT experiments, 125-nm gold beads (Nanopartz) were used as fiducial markers. The gold beads were sonicated for 10 min, diluted 1:30 in PBS and sonicated again for 10 min. We added 100 µl of the gold bead solution to each Ibidi µ-slide VI chamber, settled for 10 min and washed three times with PBS. The cell chambers were finally connected to a microfluidic device (Bruker). Before imaging, ATTO-655-labelled imager strands R1 or R2 (50 pM, 0.5 M NaCl and PBS, pH 8.3) were injected into the flow chamber at a flow rate of 200 µl per minute. In the exchange DNA-PAINT experiments, imager strands were exchanged between sequential imaging cycles by washing the samples with PBS and injecting new imager strands under equal conditions.

43Exchange DNA-PAINT data were captured as previously described using the N-STORM super-resolution microscopy system (Nikon) equipped with a ×100 oil immersion objective (Apo, NA 1.49) and an EMCCD camera (Andor Technology). ATTO-655-conjugated oligonucleotides were excited with a collimated 647-nm laser beam (at an intensity of 1.1 kW cm−2 measured at the objective) in highly inclined and laminated optical sheet mode. We acquired 20,000 consecutive frames at 10 Hz in active frame transfer mode with an EMCCD gain of 200, a pre-amp gain of 1 and at an effective pixel size of 158 nm. For astigmatism-based 3D exchange DNA-PAINT experiments, a customized cylindrical lens (RCX-39.0.38.0-5000.0-C-425-675, 10-m focal length, CVI Laser Optics) was inserted into the emission light path. NIS Elements (Nikon), LCControl (Agilent) and Micro-Manager (v1.4.22) were used for optical setup control and data acquisition. FAM134B, LC3B-II and REEP5 were imaged sequentially following the microfluidic-assisted exchange of R2 and R1 imager strands.

44Single-molecule localization and image reconstruction were carried out using Picasso (v0.2.8). Single-molecule localization was achieved by integrated Gaussian maximum likelihood estimation with the following parameters: minimum net gradient = 40,000, baseline = 205, sensitivity = 4.78 and quantum efficiency = 0.95. Fiducial gold bead markers were used for post-imaging drift correction and alignment of FAM134B, LC3B-II and REEP5 channels. Single-molecule point-spread functions were filtered based on the ATTO-655 single-molecule footprint (point-spread function (PSF) symmetry 0.7 < full width at half maximum (FWHM)(x)/FWHM(y) < 1.4, intensity threshold and localization precision < 50 nm). Signals from the same origin were linked within a radius of five times the nearest neighbour based analysis (NeNa) localization precision, and with a maximum dark time of eight consecutive frames. Signals arising from the fiducial marker that passed the filtering process were removed by excluding traces from the same origin with a length exceeding 20 consecutive frames. For exchange 3D DNA-PAINT experiments, a 3D calibration curve was recorded using z-stacks (step size = 50 nm) of surface-immobilized 0.1-µm TetraSpeck Microspheres (Thermo Fisher Scientific). The axial section was limited to 800–1,000 nm, depending on the robustness of the fit. Multi-channel single-molecule 3D localizations from exchange DNA-PAINT experiments were aligned in Picasso and visualized in ViSP (v1.0). The 3D movie (Supplementary Video 7) was generated using ViSP v1.0 open-source software.

45FAM134B and LC3B-II nanoclusters were identified in DNA-PAINT images using the density-based spatial clustering and application with noise (DBSCAN) algorithm with a radius of 31 nm and a minimum density of ten localizations. Cluster diameters were calculated for each condition from cluster areas. Microscale FAM134B clusters were segmented from nanoscale ER-phagy initiation sites using SR-Tesseler (v1.0.0.1). Voronoi tessellation was computed by calculating the first rank order local density map from single-molecule localizations. Thresholds for cluster segmentation were determined previously for each cell using Picasso (density factor 2 > δ < 6, localizations n = 800 and minimum diameter = 100 nm). HA–FAM134B nanocluster diameters and microcluster areas were tested for normal distribution using a Shapiro–Wilk normality test, and statistical significance was determined using a non-parametric Mann–Whitney U-test. The relative frequency distribution of nanocluster diameters was fitted with a log-normal distribution. The mode of the log-normal distribution is a measure for the cluster diameter (d) and was calculated as. OriginPro 2020 v9.7 (Origin Lab) was used for statistical analysis.

46For the quantification of FAM134B copy numbers in nanoscale clusters, the mean dark time of binding events was analysed from intensity–time traces as previously described. In brief, FAM134B clusters that colocalized with LC3B-II were selected manually from DNA-PAINT images and the mean dark time of binding events was determined from the plot of cumulative histograms using Picasso built-in functions. As a calibration, we used extracellular primary–secondary antibody complexes adhering to the sample coating. The inverse of dark times, also known as the qPAINT index, is proportional to the number of docking strands in clusters and was used for protein copy number determination. We fitted the relative frequency distribution of inverse dark times with a log-normal distribution and determined the mode of distribution.

47All experiments were performed in at least three independent experiments if not indicated otherwise. Data are presented as mean ± standard error of mean if not indicated otherwise. For statistical analysis, raw data were analysed for normal distribution with the Kolmogorov–Smirnov test or with graphical analysis using the Q–Q plot. If appropriate, we either used one-way ANOVA (with Bonferroni post-hoc test if not indicated otherwise), repeated-measures two-way ANOVA, Kruskal–Wallis H-test, Student’s t-test (one-tailed or two-tailed) or the Mann–Whitney U-test. P < 0.05 were considered significant.

48Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

49Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41586-023-06089-2.