Ubiquitination controls selective recycling and reshaping of the cell's protein-making network

Cell 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 infection²². 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).

Human 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.

Primary and secondary antibodies are presented in Supplementary Table. 1

We 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.

Plasmids are presented in Supplementary Table. 2

To 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.

Blue-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).

Quantifications 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.

Cells 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.

Escherichia coli (BL21(DE3)) cells were transformed with constructs encoding GST-tagged FAM134B (WT, 17KR mutant, RHD_90–264 or Ub–RHD_90–264–Ub), GST–Ub and GST, followed by expression and purification as previously described^4,23. 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 OD_{600 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).

Alternatively, 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.

For the purification of His–RHD_90–264–Strept-II and His–Ub–RHD_90–264–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.

For AMFR expression in mammalian cells, 3 l cultures of suspension-adapted HEK 293T cells were grown in customized DMEM²⁴ 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).

Liposomes made from synthetic lipids were prepared as previously described⁴. 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).

For 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 RHD_90–264 or GST–Ub–RHD_90–264–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).

Alternatively, 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 described²⁵. 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 aggregates²⁶. Small aliquots (1–2 μl) of the liposome suspension were then freeze-fractured^26,27. 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).

AMFR-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–RHD_90–264–Strept-II or 1 µM His–Ub–RHD_90–264–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.

HEK 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.

Dried 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.

MS 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. ^28,29). 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 strategy³⁰. Protein quantification and data normalization relied on the MaxLFQ algorithm implemented in MaxQuant v1.6.17.0 (ref. ³¹). The MS proteomics data have been deposited to the ProteomeXchange Consortium³² via the PRIDE partner repository³³ 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.

The previously built molecular model of FAM134B RHD⁴ 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).

Coarse-grained (CG) MD simulations were prepared using the MARTINI model (v2.2)^34,35. Initial CG structures were built using martinize.py³⁶. Assignments from DSSP (Dictionary of Secondary Structures in Proteins) program were used to generate backbone restraints to preserve local secondary structure^37,38. 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.py³⁶. Each system was energy minimized and equilibrated using the Berendsen thermostat³⁹ and barostat⁴⁰ 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 thermostat⁴¹ and the semi-isotropic Parrinello–Rahman⁴² barostat, respectively. All simulations were performed using gromacs (v2019.3)^43,44. 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 method⁴⁵, as implemented in the gmx_cluster tool.

Discontinuous bicelle systems containing saturated DMPC (14:0 PC) and DHPC (7:0 PC) lipids were assembled as previously described^4,5. 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 MemCurv⁴. 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 bicelles⁴ (k_RHD = 0.0018 ns⁻¹) to compute the acceleration factors for each system (acc = k_sys/k_RHD). 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.

A CG POPC bilayer was used to tile a tessellated buckled surface using LipidWrapper⁴⁶. 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 K_G(x, y) as implemented in MemCurv⁴.

The 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 work⁴. 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 method⁴⁷, we scaled the protein−protein LJ pair interaction well depth, ϵα=ϵ0+α(ϵoriginal−ϵ0). 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 = ϵ_original. 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.

For 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 chemistry⁴⁸. 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.

For 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.

Exchange DNA-PAINT data were captured as previously described⁴⁹ 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)⁵⁰ 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.

Single-molecule localization and image reconstruction were carried out using Picasso (v0.2.8)¹⁰. 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 (NeNa⁵¹) 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)⁵². The 3D movie (Supplementary Video 7) was generated using ViSP v1.0 open-source software⁵³.

FAM134B 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)⁴⁸. 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 d=e(µ−σ2). OriginPro 2020 v9.7 (Origin Lab) was used for statistical analysis.

For the quantification of FAM134B copy numbers in nanoscale clusters, the mean dark time of binding events was analysed from intensity–time traces as previously described¹³. 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 index⁵⁴, 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.

All 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.

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

Nature thanks the anonymous reviewers for their contribution to the peer review of this work. Peer reviewer reports are available.

The proteomics data are deposited in the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifiers: ubiquitination promotes FAM134B-mediated ER-phagy (PXD032721↗), FAM134B homodimer ubiquitination (PXD032740↗), FAM134B oligomer ubiquitination and ER-phagy (PXD032741↗), binding partners of FAM134B WT and 17KR oligomers (PXD032743↗), E3 ligase regulates FAM134B ubiquitination (PXD032750↗), in vitro FAM134B ubiquitination (PXD039186↗), in vivo FAM134B ubiquitination in AMFR-knockdown cells (PXD039187↗) and in vitro FAM134B RHD and Ub–RHD–Ub ubiquitination (PXD039188↗); MD simulation trajectory files and corresponding parameter files are large and span long microsecond timescales and multiple replicates can only be shared upon specific requests. All the data analysis of this study is in the Supplementary information. Source data for gels and blots are provided as Supplementary information. Source data are provided with this paper.

The analysis codes for bicelle-to-vesicle simulations and curvature computations on simulations with buckled membranes are provided in the GitHub repository (https://github.com/bio-phys/MemCurv↗).