Structure of p62 Filaments and Cell Details Around Calcium-Rich p62-Coated Fat Droplets

The corresponding p62 filament samples in nM concentration were also imaged by negative staining EM. In the presence of CFP-LC3b, p62 filaments exhibited a distinctive decoration in comparison with the filaments alone (Fig. 2e, f, white arrows in insets). A quantitative analysis of the filament length distribution revealed a shift towards shorter filaments for CFP-LC3b (Fig. 2g), suggesting that interaction with LC3b induced depolymerization of p62 filaments. Similarly, in the presence of CFP-LC3b, GST-4xUb-induced condensates were disintegrated into smaller substructures (Fig. 2h, i). In support of previous studies^49,50, our single-particle studies in fluorescence and electron microscopy performed in diluted nM concentration showed that poly-ubiquitinated cargoes induce condensate formation, while LC3b can promote disassembly of preformed p62 filaments as well as preformed condensates.

In order to molecularly assess the cryo-EM structure and the presented binding experiments, we generated a multimer AI-prediction of full-length p62 using AlphaFold3 (AF3)⁵¹, consisting of ten subunits that formed closed rings mediated by the well-characterized PB1 domain interactions (Supplementary Fig. 2a, Materials and Methods). In an expanded model matching the helical symmetry, the C-terminal UBA domain fits tightly to the major groove next to the PB1 domain that is consistent with the cryo-EM density (Supplementary Fig. 2b). When the model is seen in top view, the PB1 domain scaffold and UBA domain are located at the outer rim of the filament while the ZZ domain is located on the inside. Notably, parts of the IDR, including the LIR and KIR binding motifs, are positioned towards the filament's outside, supporting the observed molecular access of p62 interacting partners through the major groove of the filament. To assess the interaction accessibility within the p62 filament in more detail, we generated a second p62 multimer model in the above-described way, this time with LC3b known to bind to the LIR motif of p62²⁸. In the resulting model, LC3b is tightly fit in the filament's major groove surface between the PB1 domain and covering the UBA domain (Supplementary Fig. 2c). The LIR of p62 is positioned outward and bound in the hydrophobic pocket 1 (HP1) and 2 (HP2) of LC3 in agreement with previously determined X-ray structures⁵². A third p62 AF3-model with bound ubiquitin revealed that ubiquitin also resided in the major groove, suggesting a competing mode between LC3b and ubiquitin binding (Fig. 2j). Together, the AF3-based integrative molecular model of the p62 filament supports the observed accessibility to binding partners of ubiquitin and LC3b, while it also provides a molecular framework for the observed opposing higher-order structures of filament bundling and disassembly.

In order to verify that our observations were not limited to adherent RPE1 cells, we extended our experiments to mouse primary neurons and brain cortical tissue. First, we isolated primary neuronal cells from conditional Atg5flox:CAG-Cre^Tmx newborn mice, a model previously shown to exhibit robust and complete knockout (KO) of ATG5⁵⁵. After fixation, we labeled WT and KO astrocytes and neurons with an antibody against p62, and stained lipid droplets with Lipidspot488 (Fig. 4d–g). Second, we analyzed cortical brain sections from three-month-old WT and conditional ATG5 (cKO) mice, in which ATG5 deletion is driven by the neuronal-specific Scl32a1 promoter⁵⁶. These sections were stained with antibodies against p62 and perilipin1 (PLIN1), a protein associated with the surface of lipid droplets (Fig. 4h, i). We used confocal microscopy to image these samples and quantified the amount of lipid droplets per cell, the amount of p62 punctae per cell, the average size of the lipid droplets and the percentage of lipid droplets that co-localized with p62 signal in each cell. Both in human and mouse cells, we observed an increase in the amount of lipid droplets, while this effect was most pronounced in mouse primary KO astrocytes and cortical 3-month-old KO neurons (Fig. 4j). Additionally, as expected, we found a significant increase in the number of p62 punctae upon ATG5 KD or KO in the analyzed cells (Fig. 4k). In the absence of ATG5, the average size of the lipid droplets increased from 0.3 to 0.4 μm², from 0.3 to 0.6 μm², from 0.7 to 0.8 μm², for mCherry-p62 RPE1, mouse astrocytes and neurons, respectively (Fig. 4l). The mouse cortical neurons were harder to quantify as they, on average, contained few, very large lipid droplets. Finally, we also detected a significant increase in the number of lipid droplets that co-localized with p62 punctae in those samples lacking ATG5 (Fig. 4m). Interestingly, in our mock-transfected RPE1 cells, we also observed a moderate increase in lipid droplet sizes (Supplementary Fig. 4d), possibly due to the fact that the cationic lipid mixtures used in transfection mixtures have been observed to induce autophagy, promoting LC3 conversion and p62 degradation^57,58. Together with the autophagy-stalling effect of ATG5 siRNA, the lipids likely contribute to further enrichment of p62-positive structures. In conclusion, when transfecting RPE1 cells with ATG5 siRNA, and upon ATG5 KO in mouse primary astrocytes, mouse primary neurons and/or 3-month-old mouse cortex, we observed an accumulation of lipid droplets co-localizing with p62 punctae, as well as a size increase for lipid droplets. The characterized condition of ATG5 knockdown in human RPE1 cells may be well suited for the in situ visualization of p62 punctae adjacent to lipid droplet cargo, as autophagosome formation and trafficking to the lysosome is impaired.

Given the unexpected finding of inorganic elements located in the p62 coat, we tested whether the addition of Ca²⁺ ions had an effect on purified p62 filaments. When increasing the CaCl₂ concentrations from 100 nM to 100 mM, we observed a bundling of p62 filaments at 100 mM CaCl₂ in negative staining EM (Fig. 6f–h). Moreover, when testing other divalent cations such as Zn²⁺ and Mg²⁺, we found that the addition of ZnCl₂ had a stronger clustering effect already visible at 100 µM, while the addition of MgCl₂ did not show a significant bundling (Supplementary Fig. 10). Together, these experiments show that the observed p62-positive structures correspond to lipid droplets surrounded by a coat made of p62, including significant amounts of calcium (Fig. 6i) that also promotes the formation of p62 oligomer bundles in vitro and may explain the characteristic dense layer observed in the tomograms.

Finally, we inspected proteins that were possibly related to the observed calcium accumulation around the p62-lipid droplet structures. Intracellular calcium chelators, such as calmodulin or calcineurin, were neither enriched nor depleted upon ATG5 knockdown. However, we identified a calcium channel and a regulator of calcium homeostasis in the ATG5 knockdown but not in the endogeneous ATG5 sample: First, ITPR2, an ER-resident, inositol 1,4,5-trisphosphate-gated calcium channel and, second, MCUb, known as a negative regulator for the mitochondrial calcium uniporter. The former proteins, i.e., ITPR channels, release calcium from the ER and have been shown to affect autophagy⁶⁵. The latter protein, i.e., MCUb, has been described to block calcium overload of nearby mitochondria, and has also been associated with the early stages of autophagy and mitophagy^66–68. Together, our proximity proteomics experiments on the ATG5 knockdown induced p62-encapsulated lipid droplets revealed proteins such as SIRT1 and alpha-synuclein, linking lipid droplets to autophagy. Moreover, upon ATG5 knockdown, proteins involved in calcium transport from the ER, such as ITPR2 and MCUb, were found in close proximity to p62, supporting the notion that the observed calcium layer and nearby single phospholipid bilayer may represent calcium released from the ER.

Full-length p62 (440 residues) was cloned into a pETM43 expression vector containing an N-terminal maltose-binding protein (MBP) tag and a C-terminal His-tag. Both tags featured a recognition sequence for HRV-3C protease, facilitating their removal after purification. 4x-linear-Ubiquitin, CFP-LC3B, and LC3B with an additional C-terminal cysteine residue were cloned into the pGEX expression vector containing an N-terminal Glutathione-S-Transferase (GST) tag with an HRV-3C protease cleavage site. For protein expression, E. coli BL21 (DE3) cells were used, with bacterial cultures grown at 37 °C in Terrific Broth (TB) medium to an OD₆₀₀ of ~2. Induction with IsoPropylThioGalactoside (IPTG) followed, and cells were left to grow overnight at 23 °C. Cells were harvested, and the pellet was stored at −80 °C until further use.

After thawing on ice, cells were resuspended in lysis buffer [50 mM HEPES, pH 8.0, 300 mM NaCl, 2.5 mM MgSO₄, 10 μM ZnCl₂, 0.5 mM TCEP and 1 tablet EDTA-free protease inhibitor cocktail] and lysed using a cell disruptor at 1.85 bar. Cell debris was removed by centrifugation, and the cleared lysate was applied to a Ni-NTA affinity column. The column was washed twice with lysis buffer containing 10 mM and 20 mM imidazole, respectively, and the protein was eluted with lysis buffer containing 300 mM imidazole. Eluted protein fractions were pooled and subsequently applied to an amylose resin that binds MBP-fusion proteins with high affinity. The column was washed with lysis buffer, and the protein was eluted with lysis buffer containing 25 mM maltose. Eluted protein fractions were pooled, and buffer was exchanged to desalting buffer [50 mM HEPES, pH 8.0, 300 mM NaCl and 0.5 mM TCEP] using Zeba desalting columns.

For the purification of LC3b and GST-4xUb, the cell pellets were resuspended in lysis buffer [50 mM HEPES, pH 7.4, 300 mM NaCl, 1 mM DTT, and 1 tablet EDTA-free protease inhibitor cocktail] and lysed using a cell disruptor at 1.9 bar. The cleared lysate was applied to Glutathione Sepharose 4 Fast Flow GST-tagged protein purification resin (Cytiva, Cat No 17513201), washed with 10 column volume lysis buffer, and eluted in lysis buffer containing 10 mM reduced glutathione. Elution fractions containing the protein of interest were pooled, further purified by size exclusion chromatography via SD200 16/600 column (Cytiva) [in 50 mM HEPES, pH 7.4, 150 mM NaCl, 1 mM DTT], and frozen at −80 °C for storage.

Protein was concentrated to approximately 2 mg/mL using a 30 kDa Amicon concentrator and subsequently incubated with 1:2.5 mass ratio of GST-HRV-3C protease for 3 h at room temperature and overnight at 4 °C to remove the MBP and His-tag, allowing p62 filament formation. Filaments were isolated by centrifugation at 20,000 × g for 30 min and resuspended in desalting buffer. Isolated filaments were labeled with lysine residues by incubating with 1:1 molar ratio of Cy5-NHS for 1 h at room temperature and overnight at 4 °C. Excess dye was removed by centrifugation, labeled filaments were resuspended in storage buffer [50 mM HEPES, pH 8.0, 50 mM NaCl and 0.5 mM TCEP] and stored at 4 °C until further use.

Quantifoil Cu 200 mesh R1.2/R1.3 grids were glow-discharged, and 2–5 μL of the protein sample was applied in a Vitrobot Mark IV (ThermoFischer Scientific). The sample chamber was maintained at 70–100 % humidity and 4–10 °C, with a 2–6 s blotting time before plunging into an ethane/propane mixture. Data collection was performed using SerialEM⁹⁴ on a Talos Arctica 200 KeV G2 transmission electron microscope (ThermoFischer Scientific), equipped with a K3 direct electron detector camera and Quantum GIF filter (Gatan, Inc). Approximately 4000 micrographs were collected in counted superresolution mode at 100,000x magnification with an underfocus range of −0.5 μm to −3 μm. The micrographs were motion corrected in WARP⁹⁵, and particles were picked using the deep-learning-based object detection software crYOLO⁹⁶. Initially, the picked particle locations and micrographs were imported into RELION and SPRING^46,48, which used CTFFIND4 for CTF estimation⁹⁷. Later, CTF estimated micrographs were imported into cryoSPARC⁴⁷ and particle locations were extracted with a 90 % overlap, yielding approximately 1,000,000 segments of a box size of 503 Å. Two rounds of 2D classification were performed in order to remove heterogeneous particles. Classes with clearly defined PB1 domains were then selected, leading to a total of 600,000 segments. The helical parameters as determined by Jakobi et al.²² of truncated p62 helical filaments were taken as starting conditions for helical refinement in cryoSPARC, where the symmetry parameters converged (Supplementary Fig. 11). The final resolution was estimated by FSC using the 0.143 cutoff⁹⁸ and FDR-FSC criterion⁹⁹. Maps were rendered in UCSF ChimeraX¹⁰⁰.

To model the p62 filament structure, we used AlphaFold3 (AF3) multimer predictions⁵¹: first, we tested p62 full-length sequences arranged in dimers, trimers, tetramers, up to decamers. AF3 consistently favored closed ring-like assemblies, even for small oligomers (≥3 subunits), while for larger rings the resulting p62 rings exhibited close-to-rotational symmetry. A single p62 subunit was excised from an AF3-generated p62 closed ring-like decamer with a diameter of 145 Å, approaching the diameter of the filament. To relate these predictions to the p62 filament, we fitted the PB1 domain in the corresponding well-resolved density of one strand of the filament map while including the full-length p62 model and repeated the operation for the opposite PB1 strand. Using the determined helical symmetry parameters of 11.0 Å helical rise and a −26.1° helical rotation, we propagated the two subunits along the helical axis, generating a double-helical filament model composed of repeating p62 units. This helical assembly yielded a plausible model of the p62 filament, with subunits well fitting the experimental cryo-EM density while maintaining interaction geometries close to the AF3-predicted oligomers.

Surface imaging was performed with an inverted microscope (Olympus IX-81, Olympus Corp., Shinjuku, Japan) in total internal reflection fluorescence (TIRF) mode. Molecules were excited using lasers at 488 and 639 nm (Sapphire 488–200 and Obis 637–140, Coherent Inc.), focused off-axis into the back-focal plane of an oil-immersion objective (Olympus UApoN, 100x, NA = 1.49, Olympus Corp). The excitation power and corresponding intensity at the sample stage were 3.25 mW and 10.2 W/cm² for the blue laser, and 2.2 mW and 6.8 W/cm² for the red laser. The imaging was performed on an EMCCD camera (Andor iXon Ultra 888, Andor Technology) with 50, 100, 40 and 100 ms exposure time for Supplementary Videos 1–4, corresponding to 20, 10, 25 and 10 fps. Midway through acquiring Supplementary Video 2, the illumination was switched from blue to red laser to observe the same p62 filament with LC3b bound to it. Each video consists of 200 frames. Image analysis and video processing were carried out using Fiji (ImageJ, version 2.14.0)¹⁰¹. The emitted fluorescence was collected with the same objective (objective-based TIRF) and passed through a dual-band dichroic mirror as well as a dual-band emission filter (ZT488/640rpc and ZET488/640 m, Chroma Technology Corp). p62 filaments-Cy5 and CFP-LC3b were diluted to concentrations of a few nM, and mixed in a 1:1 molar ratio. GST-4xUb was added at a concentration of 5 µM.

For negative staining, 3.5 μL of the sample was applied onto glow-discharged carbon-formvar-coated 300-mesh copper grids and allowed to absorb for 2 min before being blotted according to the side-blotting method. The sample was subsequently stained for 30 s with 6 μL 2% uranyl acetate, blotted and allowed to dry. Negatively stained samples were examined on a Talos L120C G2 transmission electron microscope (ThermoFisher Scientific), operated at 120 kV. Micrographs at high magnification (45,000x and 92,000x) were collected on a 4k x 4k Ceta 16 M CEMOS camera using TEM Imaging & Analysis Software (TIA, ThermoFisher Scientific).

We used Origin2021b to perform the Kolmogorov-Smirnov (K-S) test for the two datasets of filament lengths (p62 filaments without and with LC3b present). We sorted the two datasets in ascending filament length order and computed the cumulative distribution function (CDF) for both datasets. CDF increases from 0 to 1 from the smallest to the largest value in the distribution. The CDFs from both datasets were overlaid in a plot and the largest absolute difference between them was computed (Supplementary Fig. 12).1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$D=\max |{F}_{1}\left(x\right)-{F}_{2}\left(x\right)|$$\end{document}D=max∣F1x−F2x∣where F₁(x) and F₂(x) are the CDFs of each dataset.

Then we computed the K–S test statistic K2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$K=D\times \sqrt{\left({n}_{1}{n}_{2}\right)/\left({n}_{1}+{n}_{2}\right)}$$\end{document}K=D×n1n2/n1+n2where n₁ and n₂ are the sizes of each dataset.

The approximate p-value was calculated as 3 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p\approx {e}^{-2{K}^{2}}$$\end{document} p ≈ e − 2 K 2

p62 filament cryo-EM grids were prepared as described above. Tilt series were recorded on a 200 kV Talos Arctica G2 transmission electron microscope (ThermoFisher Scientific), equipped with a Bioquantum GIF (Gatan, Inc), a Volta Phase Plate and a K3 direct electron detector (Gatan, Inc) using SerialEM (Mastronarde, 2005). Each tilt was recorded as a movie at 63,000x magnification with a 0.681 Å/pixel size in super resolution mode, corresponding to a dose of approximately 2.0 e/Ų at 0° tilt, 10 frames per tilt, and a total dose per tilt series of approximately 130 e/Ų. All tilt series were recorded from −54° to 54° with 3° increments, using a dose-symmetric bidirectional acquisition scheme including 'the varied intensity' option (i.e. higher tilt angles received higher doses, as 1/cosine of the tilt angle) in SerialEM and a nominal defocus of −1.2 μm. The raw data were motion corrected and CTF corrected using WARP (version 1.0.9)⁹⁵ and the tilt series were reconstructed using IMOD¹⁰².

GST-4x-linear-Ubiquitin and p62 filaments were prepared as described above. 15 µM of GST-4x-linear-Ubiquitin were mixed with 1 µM p62 filaments. After 4 min incubation time at 4 °C, the formed phase separation droplets were negatively stained as described above or plunge frozen with a Vitrobot Mark IV (ThermoFisher Scientific). 50 nm SUVs were prepared containing 5% DSPE-PEG(2000) maleimide (Avanti Polar lipids, Cat No 880126) in DOPS (Avantiv Polar lipids, Cat No 840035). The liposomes were incubated with 100 µM LC3b-Cys for 16 h at 4 °C before removing unbound LC3b-Cys by size exclusion chromatography on Sephacryl S-500 at Äkta Micro (Cytiva). A total of 2 mg/ml of liposomes was mixed with 1 µM p62 filaments and incubated for 5 min at 4 °C. The sample was plunge frozen using a Leica GM2 plunge freezer and imaged in a 300 kV Titan Krios G4 equipped with a BioContinuum energy filter and K3 camera at 64,000x using Tomo5 software (ThermoFisher Scientific). The tilt series were recorded from −60° to 60° with a 3° increment in a dose-symmetric bidirectional acquisition scheme. An electron dose of 4.0 e⁻/Ų/tilt (at 0° tilt) with 6 frames per tilt resulted in a total dose per tilt series of approximately 160 e⁻/Ų. The defocus was varied between −2.0 to −4.0 μm. The images were motion corrected in WARP⁹⁵ and tomograms reconstructed with AreTomo¹⁰³. Overall binning was 16x and resulted in a pixel size of 11.2 Å. Tomograms were denoised and missing wedge corrected with IsoNet¹⁰⁴.

Cell culture experiments were carried out in human RPE1 cells (hTERT RPE1, CRL-4000, ATCC) cultured at 37 °C and 5% CO₂ in DMEM GlutaMAX media (Gibco, Cat No 31331028) supplemented with 10% FCS (Sigma-Aldrich, Cat No F7524) and Pen/Strep (Gibco, Cat No 15140122). For confocal microscopy experiments prior to plunge-freezing, the culturing media were substituted with media of low fluorescence background: Fluorobrite DMEM (Gibco, Cat No A1896701). Additional experiments were performed on RPE1 cells stably expressing human mCherry-p62. Transient expression of mCherry-p62 was executed by transfection with FugeneHD (Promega, Cat No E2311) and a modified pDEST vector containing the human WT p62 gene fused with an N-terminal mCherry sequence (Pankiv et al., 2007). Knockdown experiments were carried out by transfection with RNAiMAX (ThermoFisher, Cat No 13778075) and ON-TARGETplus SMARTpool siRNA against ATG5 (Dharmcon Horizon Discovery, Cat No L-004374-00-0005), GAPDH (Dharmcon Horizon Discovery, Cat No D-001830-10-05) and a non-targeting control (Dharmcon Horizon Discovery, Cat No D-001810-10-05). Follow-up experiments were performed 72 h after a reverse transfection with 75 nm siRNA. Lipid droplets were stained with 1:500 Lipi-Blue (Dojindo, Cat No LD01-10) for 24 h. Lysosomes were stained with the CytoPainter Lysosomal staining kit (Abcam, Cat No ab112136).

All animal experiments were approved and performed according to the regulations of the LANUV, NRW, Germany (AZ: 81-02.04.2020.A418, 81-02.05.40.20.075, 81-02.04.2021.A132, Anzeige §4.20.021) guidelines. Mice were maintained in a pathogen-free environment in ventilated polycarbonate cages and housed in groups of five animals per cage with constant temperature and humidity at 12-h/12-h light–dark cycles. Food and water were provided ad libitum. Atg5flox:Slc32a1-Cre:tdTomato (ATG5 cKO) mice were described previously^56,74. The following genotyping primers were used to genotype the animals: ATG5 forward 1 (GAA TAT GAA GCC ACA CCC CTG AAA TG), ATG5 forward 2 (ACA ACG TCG AGC ACG CTG GCG AAG G), ATG5 reverse (GTA CTG CAT AAT GGT TTA ACT CTT GC), Cre_1(GAA CCT GAT GGA CAT GTT CAG G), Cre_2(AGT GCG TTC GAA CGC TAG AGC CTG T), Cre_3(TTA CGT CCA TCG TGG ACA) and Cre_4(TGG GCT GGG TGT TAG CC). In accordance with the SAGER guidelines, we have ensured that both male and female mice were included in all experimental groups. Cortical and hippocampal neurons from postnatal pups (P1–5) were isolated as previously described⁵⁵ and plated on PDL-coated coverslips or dishes. Neurons were grown in a MEM (Gibco) medium, containing 0.5% Glucose (Sigma), 0.02% NaHCO3 (Roth) and 0.01% Transferrin (Merck), which was further supplemented with 5% FBS (Sigma), 0.25% L-GlutMax (100×, Gibco), 2% B27 (50×, Gibco), 1% Pen/Strep (10000 Units/ml Penicillin, 10,000 µg/ml Streptomycin, Gibco). After two days in vitro 2 mM AraC was added to the culture medium to limit glial proliferation. To initiate homologous recombination, primary neurons isolated from floxed animals expressing a tamoxifen-inducible Cre recombinase were treated with 0.4 μM (Z)−4-hydroxytamoxifen (Sigma) immediately after plating. After 24 and 48 h, cells were treated with 0.2 µM and 0.4 µM of tamoxifen, respectively, during medium renewal. Ethanol was added to control neurons (WT) in an amount equal to the tamoxifen. The DIV 14–16 postnatal neurons were fixed in 4% PFA/sucrose in phosphate-buffered saline (PBS) for 15 min at RT. Cells are then washed 3x with PBS and then blocked in 5% NGS in 0.3% Triton X-100 for 1 h at RT. Primary antibodies diluted in 5% NGS and 0.3% Triton X-100. Coverslips were rinsed 3x PBS and incubated with corresponding secondary antibodies diluted in 5% NGS and 0.3% Triton X-100, and incubated for 30 min at RT. The following antibodies were used: polyclonal guinea pig anti-p62 (1:1000, Progen, cat no GP62-C), Goat anti-Guinea Pig Alexa Fluor 568 (1:1000, Invitrogen, cat no A-11075), and Donkey anti-Rabbit Alexa Fluor 647 (1:1000, Invitrogen, cat no A-31573). Cells were washed 3x in PBS and then incubated in 1:1000 Lipidspot488 (Biotium, cat no #70065-T) for 15 min at RT. Following incubation, cells were directly mounted on glass coverslips. Post-natal neurons were imaged using with Leica SP8 confocal microscope (Leica Microsystems) equipped with a 63 × /1.32 oil DIC objective and a pulsed excitation white light laser (WLL; ∼ 80-ps pulse width, 80-MHz repetition rate; NKT Photonics) at a resolution of 1024 × 1024 pixels in sequential scanning frame-by-frame mode. A single Z plane was yielded. For quantitative analysis of fluorescent puncta, each channel of interest was processed by subtracting background and the area of the neuron and/or cell body was manually selected using ImageJ (Version 1.54p) selection tools (ROI). Fluorescent puncta were counted with the ImageJ Cell Counter Plugin. Fluorescent puncta size of the lipid droplets (Lipidspot488) was determined by applying the autothreshold "Huang" algorithm in ImageJ (Version 1.54p) and analyzed using the "Analyze particles" ImageJ (Version 1.54p) plugin to get average puncta size per cell in µM².

All animal experiments were approved and performed according to the regulations of the LANUV, NRW, Germany (AZ: 81-02.04.2020.A418, 81-02.05.40.20.075, 81-02.04.2021.A132, Anzeige §4.20.021) guidelines. Mice were maintained in a pathogen-free environment in ventilated polycarbonate cages and housed in groups of five animals per cage with constant temperature and humidity at 12-h/12-h light–dark cycles. Food and water were provided ad libitum. Atg5flox:Slc32a1-Cre:tdTomato (ATG5 cKO) mice were described previously^56,74. The following genotyping primers were used to genotype the animals: ATG5 forward 1 (GAA TAT GAA GCC ACA CCC CTG AAA TG), ATG5 forward 2 (ACA ACG TCG AGC ACG CTG GCG AAG G), ATG5 reverse (GTA CTG CAT AAT GGT TTA ACT CTT GC), Cre_1(GAA CCT GAT GGA CAT GTT CAG G), Cre_2(AGT GCG TTC GAA CGC TAG AGC CTG T), Cre_3(TTA CGT CCA TCG TGG ACA) and Cre_4(TGG GCT GGG TGT TAG CC). In accordance with the SAGER guidelines, we have ensured that both male and female mice were included in all experimental groups. Mice at 12–15-week were euthanized by an overdose of 1.2% ketamine, and 0.16% xylazine in PBS (i.p., 10 μl per 10 g body weight) and transcardial perfusion was performed as previously described¹⁰⁵. Brains were carefully dissected and postfixed in 4% PFA overnight before being processed for immunohistochemistry as previously described¹⁰⁵. Corresponding horizontal 40 μm sections from WT and KO littermates were washed 3x in PBS for 5 min each. Sections were blocked with 10% NGS/NDS in 0.5% PBS-T for 1 h at RT. Primary antibodies were incubated in sections in 3% NGS/NDS and 0.3% PBS-T overnight at 4 °C. Sections were washed 3 × 10 min in 0.3% PBS-T before incubation in fluorescence-labeled secondary antibodies in 3% NGS/NDS and 0.3% PBS-T for 1 h at RT, protected from light. The following antibodies were used: polyclonal guinea pig anti-p62 (1:1000, Progen, cat no GP62-C), recombinant rabbit anti-perlipin1 (1:500, Proteintech, cat no 83905-4-RR), Donkey anti-Rabbit Alexa Fluor 647 (1:1000, Invitrogen, cat no A-31573) and Goat anti-Guinea Pig Alexa Fluor 488 (1:1000, Invitrogen, cat no A-11073). Sections were imaged at a Leica SP8 confocal microscope (Leica Microsystems) equipped with a 63 ×/ 1.32 oil DIC objective and a pulsed excitation white light laser (WLL; ∼ 80-ps pulse width, 80-MHz repetition rate; NKT Photonics) at a resolution of 1024 × 1024 pixels in sequential scanning frame-by-frame mode. Stacks of 30 optical sections were acquired, with a fixed stack size of 0.25 μm. For quantitative analysis of fluorescent puncta, each channel of interest was processed by subtracting background and the area of the neuron and/or cell body was manually selected using ImageJ (Version 1.54p) selection tools (ROI). Fluorescent puncta were counted with the ImageJ Cell Counter Plugin. Fluorescent puncta size of the lipid droplets (Perlipin1) was determined by applying the autothreshold "Huang" algorithm in ImageJ (Version 1.54p) and analyzed using the "Analyze particles" ImageJ (Version 1.54p) plugin to get the average puncta size per cell in μM².

Cells were lysed by a quick wash with ice-cold PBS (Gibco, Cat No 10010056) and a subsequent 15 min incubation in ice-cold RIPA buffer (Sigma-Aldrich, Cat No R0278-50mL) supplemented with a protease inhibitor cocktail (Roche, Cat No 11836170001). Cell lysate was collected with a cell scraper before removing the cell debris by centrifugation. Total protein content was quantified and corrected using a Pierce 660 nm assay (ThermoFisher, Cat No 22662) before boiling the sampling in a reducing sample buffer and separation on a 4–15 % mini-PROTEAN Precast SDS-PAGE (Bio-Rad, Cat No 4561085). Transfer to PVDF membrane was performed using a Trans-Blot pack (Bio-Rad, Cat No 1704156) and the membrane was blocked for 2 h in 2% BSA in TBST at 4 °C. The membrane was incubated overnight with primary antibody in 1 % BSA in TBST at 4 °C, washed 3 times in TBST before the secondary antibody was added for 2 h in 1% BSA in TBST at 4 °C. Prior to imaging on the Azure Sapphire RGB Biomolecular Imager (Azure Biosystems), the membrane was washed 3 times in TBST and incubated in Azure Radiance Q HRP (Biozym, Cat No 512101). The following antibodies were used: polyclonal rabbit anti-p62 (MBL International, Cat No PM045), monoclonal rabbit anti-ATG5 (Abcam, Cat No ab108327), monoclonal mouse anti-GAPDH (Invitrogen, Cat No MA1-16757), polyclonal rabbit anti-LC3 (Abcam, Cat No ab48394), goat anti-mouse HRP conjugate (ThermoFisher, Cat No 32230), and goat anti-rabbit HRP conjugate (ThermoFisher, Cat No 31460). Band intensity quantifications were performed using ImageLab (BioRad, version 6.1.0).

RPE1 cells were seeded on micropatterned grids before plunge freezing using recently described protocols^106,107. Briefly, 200 mesh gold Quantifoil R2/1 grids with a SiO₂ support layer (Quantifoil, Cat No N1-S15nAu20-01) were glow discharged and treated with PLL-g-PEG (SuSoS, PLL(20)-g[3.5]- PEG(5)) overnight. Next, the grids were washed in PBS and placed in drops of 14.5 mg/mL PLPP (Alvéole, Cat No PLPP), and 20 μm circles in the center of 10 × 10 grid squares were micropatterned with a PRIMO photopatterning device (Alvéole, Photopatterning Device consisting of a DMD + laser 375 nm, 75 mW module, and driving electronics). The grids were then washed thoroughly in PBS and incubated in human fibronectin immediately before use (Gibco, Cat No PHE0023). The cells were trypsinized (Gibco, Cat No 25300054), passed through a 40 μm cell strainer (PluriSelect, Cat No 43-10040-40) and counted before seeding ~200,000 cells in a 30 mm glass-bottom dish (Greiner Bio-One, Cat No 627860) containing the micropatterned grids in a small amount of DMEM. The cells were allowed to attach to the grids for 4–5 h before vitrification in liquid ethane using a Vitrobot Mark IV (ThermoFisher Scientific). The following settings were used; a chamber temperature of 37 °C, a chamber humidity of 85% humidity, backside blotting with an 8.5 s blot time and blot force of −10 and a drain time of 1 s.

RPE1 cell phenotypes were studied using a Zeiss LSM710 laser scanning confocal microscope (Carl Zeiss GmbH). The cells were grown in 30 mm glass-bottom dishes with 4 compartments (Greiner Bio-One, Cat No 627870) and imaged at 37 °C using a Zeiss 63× oil immersion objective (Plan-Apochromat, DIC, 1.4 NA). Lipi-Blue was excited with a diode laser at 405 nm, and fluorescent emission was detected after passing a suitable emission filter (410–500 nm) using a Multi Alkali photomultiplier tube. The lysosomal stain was excited using an argon ion laser at 488 nm, and emission was detected using an emission filter (499–541 nm). MCherry-p62 was detected using a 543 HeNe543 laser and an emission filter set to 579–696 nm using a GaAsP photomultiplier tube. Image analysis and quantification were done using Fiji (ImageJ, version 2.14.0)¹⁰¹ and a self-written routine in CellProfiler (version 4.2.6)¹⁰⁸. Measurements were exported as.csv files and analyzed in OriginPro 2023 (OriginLab Corporation, version 10.0.0.154). Vitrified grids and thinned lamellae were studied using a Zeiss LSM800 upright laser scanning confocal microscope with a cryostage (Linkam CMS196). Super-resolution images were acquired using a Zeiss 100x air immersion objective (EC Epiplan NeoFluar, DIC, NA = 0.75) and an AiryScan Detector. MCherry-p62 was detected using a 561 nm diode. Image processing was done using Zen Blue (Carl Zeiss GmbH, version 3.3.89).

Cryo-lamellae were made in a correlative manner using the in-chamber METEOR fluorescence microscope (Delmic, Cat No 2707-999-0014-2) equipped with a LMPLFLN 50×, NA = 0.5, WD = 10.6 mm objective lens in the Aquilos2 cryo-FIB-SEM system (ThermoFisher Scientific). In short, micropatterned grids with RPE1 cells were loaded under cryogenic conditions, taking special care to avoid contamination build-up from air humidity. Cells were located using the electron beam, and the sample was placed at eucentric height using MAPS software (ThermoFisher Scientific, version 3.19). Cell phenotype was confirmed by imaging each cell in the METEOR fluorescence microscope. Each cell was imaged using the focused ion beam, and the location of the final 150 nm lamella was marked in AutoTEM (ThermoFisher Scientific, version 2.4.1). The lamellae were generated using a step-wise process; the cell was thinned to 1 μm with a 0.5 nA milling current, then thinned to 600 nm with a 0.3 nA current, next to 300 nm with a 0.1 nA current and finally thinned to 200 nm with a 50 pA ion beam. The final polishing step, in which the lamella was thinned to 120-180 nm, took place with a 30 pA ion beam. The final lamella was once again imaged using the METEOR fluorescence microscope, and these images were used to determine where tomograms would be recorded. The lamella was coated with a thin platinum layer; sputtered for 7 s at 7 mA, to prevent sample charging as previously described (Berkamp et al., 2023). Next, the samples were transferred under cryo-conditions to a 200 kV Talos Arctica G2 transmission electron microscope (ThermoFisher Scientific), equipped with a Bioquantum GIF (Gatan, Inc) and a K3 direct electron detector (Gatan, Inc). Tilt series were recorded on p62 positive areas of the lamellae, using SerialEM (version 4.2.0)⁹⁴ as movies at 24,000x magnification with a 1.7435 Å/pixel size in super resolution mode and a dose of approximately 2.0 e/Ų at 0° tilt, 10 frames per tilt, and a total dose per tilt series of approximately 130 e/Ų. All tilt series were recorded from −54° to 54° with 3° increments starting at the pre-tilt of the lamella, using a dose-symmetric bidirectional acquisition scheme with weighted dose (i.e. higher tilt angles received higher doses, as 1/cosine of the tilt angle), tilt angle and a nominal defocus of −7 μm. The raw data were motion corrected and CTF corrected using WARP (version 1.0.9)⁹⁵, and the tilt series were reconstructed using Aretomo2 (SBGrid)^103,109. To aid the cryo tomogram visualization, they were deconvolved, denoised, and the missing wedge compensated using IsoNet¹⁰⁴ and cryoCARE¹¹⁰. To aid in segmentation a pre-trained U-Net model for non-denoised data in MemBrain-Seg (version 9b)¹¹¹ was used and the processed tomograms were then loaded into Amira 3D (ThermoFisher Scientific, version 2022.2) for semi-manual segmentation. Tomograms were visualized using the IMOD software package (version 4.11.20)¹⁰². To aid in targeting the EDX spectra, high dose, single tilt micrographs were recorded on p62 positive areas of the lamellae at 24,000x and a dose of approximately 40 e⁻/Ų/micrograph and 180 frames per tilt, with each a dose of approximately 0.24 e⁻/Ų. Motion correction for these micrographs was done in Relion (SBGrid, version 4.0.1)¹¹²

The in situ EDX experiments were carried out on a probe-corrected 200 kV FEI Titan microscope (ThermoFisher Scientific) fitted with a Super-X G2 quadrant detector EDX system (Bruker) and a Simple Origin dual grid cryo transfer holder (Simple Origin, Inc). The microscope was operated under cryo-conditions in STEM mode with a 1.5 mrad convergence angle, a 1 nm probe size at a magnification of 55,000, resulting in a 2.9 nm pixel size. EDX spectra were recorded with a 50 μs dwell time and 50 frames, resulting in a total dose of 500 e/Ų, a camera length of 3.5 m and a detector dead time of approximately 70%. We found 500 e⁻/Ų to be a good trade-off between a high signal-to-noise ratio in the EDX spectra while still maintaining the ultrastructural integrity of the sample. EDX spectra processing was done in Velox (ThermoFisher Scientific, version 3.14). Atomic map representations were rendered with a 12 px pre-filter average and a 12 px post-filter average for the p62 structures and a 25 px pre-filter average and a 9 px post-filter average for the lipid droplet and lysosome control experiments. The spectra were generated by segmenting the p62 coat in the HAADF-STEM micrograph and segmenting the same number of pixels in the lipid droplet core and a random area of the cytosol. This same process was followed for the control experiments.

Lentivirus was ordered with a vector corresponding to V5-miniTurbo-mCherry-p62-IRES-puro (Sirion Biotech GmbH). The virus was added to WT RPE1 cells, clones were puromycin selected, and transduction confirmed via fluorescence microscopy. The cells were seeded in T75 flasks, and ATG5 was knocked down using siRNAs (ON-TARGETplus SMARTpool siRNA against ATG5, Dharmacon Horizon Discovery), and the cells were incubated for 72 h. Next, 500 µM biotin was added (Sigma-Aldrich, Cat No B4501-1G,) and 5 mM ATP (AppliChem, Cat No A1348,0100) and the samples were incubated for 2 h at 37 °C. The cells were placed one ice, washed in ice-cold PBS (Gibco, Cat No 10010056) and harvested with a cell scraper and RIPA buffer (Sigma-Aldrich, Cat No R0278-50mL) supplemented with protease inhibitor cocktail (Roche, Cat No 11836170001) and 1 mM PMSF. Cell debris was removed via centrifugation. All experiments were done in quadruplicates, on mCherry-p62 RPE1 cells, miniTurbo-mCherry-p62 RPE1 cells and ATG5 KD + miniTurbo-mCherry-p62 RPE1 cells. The lysates were incubated with 200 µL washed Pierce Streptavidin Magnetic Beads (Thermo Fisher Scientific, Cat No 88817) overnight at 4 °C. The beads were then washed with 3 × 500 µL 100 mM Hepes, pH = 7.4, 1 mL 1 M KCl, 1 mL 0.1 M freshly made Na₂CO_3, 3 × 1 mL 3 M urea and finally the beads were suspended in 100 µL 3 M urea in 10 mM Tris, pH = 8. DTT was added to the final concentration of 10 mM, and the sample was incubated for 30 min at 56 °C with 1000 rpm shaking. Next, the sample was alkylated with 50 mM CAA (Sigma-Aldrich, Cat No 22790-250 G) for 30 min in the dark. The reaction was quenched with 50 mM DTT for 20 min, and the beads washed twice with 100 uL 2 M urea. On-bead digestion was performed overnight at 37 °C in 50 µL 100 mM HEPES pH = 7.4 using a 1:100 trypsin to proteome ratio (Promega, Cat no V5111). The supernatant was collected and pooled with 2 × 100 µL 2 M urea wash steps. The samples were acidified with 2% TFA (Sigma Aldrich, Cat no 1082180050), to pH < 3. StageTips (Affinisep, Cat No Tips-RPS-M.T1.200.96) were equilibrated as follows: they were washed with pure methanol, with 0.1% formic acid in 80% acetonitrile and then twice with 0.1% formic acid. Each sample was split over 6 StageTips and centrifuged at 500 × g for 2 min, and then washed twice in 30 uL 0.1% formic acid and centrifuged at 500 × g for 2 min. Peptides were eluted with 60 µL of 60% ACN, 5% NH3 and vacuum-concentrated with the concentrator plus (Eppendorf). The samples were stored at −20 °C, and right before mass spectrometry measurement, reconstituted in 0.1% TFA. The concentration was determined with a Nanodrop before injecting the peptides on a Q Exactive Orbitrap (Thermo Fisher Scientific) coupled to an UltiMate® 3000 UHPLC-System (Thermo Fisher Scientific) ran in positive mode utilizing a μPACTM C18 pillar array column (200 cm bed length, Thermo Fisher Scientific) heated to 40 °C. The 140 min gradient was applied in the following solvent B concentration: 6% to 5 min, 6–10% (5–10 min), 10–45% until 110 min, 45–99% until 118 min, 99% for 7.9 min, 99–6% in 0.1 min, and 6% for 14 min. Mobile phase A consisted of 0.1% formic acid (LC-MS grade) in water, and mobile phase B consisted of 0.1% formic acid in 86% acetonitrile (LC-MS grade). After peptide separation, electrospray ionization (ESI) was used as the ionization method. Therefore, a Nanospray Flex ion source (Thermo Fisher Scientific), liquid Junction™ PST-HV-NFU (MS Wil) and a fused silica emitter (20 µm inner diameter, 360 µm outer diameter, MicroOmics Technologies LLC) were used. A spray voltage of 1.8 kV and a capillary temperature of 250 °C were applied. Full or overview MS scans were acquired in the m/z range of 385–1050 with a resolution of 70000 (at m/z 200) and an automatic gain control (AGC) target of 3 × 10⁶ ions with a maximum injection time of 60 ms. Each DIA cycle consisted of one full MS scan followed by 50 DIA windows with 12.5 m/z isolation width overlapping by 0.25 m/z covering the mass range of 400-1000. Each DIA window had a resolution of 35000, an AGC target of 1 × 10⁶ ions, and a maximum injection time of 60 ms. The peptides were fragmented by higher-energy collision-induced dissociation (HCD) with a normalized collision energy (NCE) of 26. Raw data was first converted by MSConvertGUI (64-bit, version 3.0.21229-9668f52) to the mzML format files using PeakPicking filtering. FragPipe (version 22.0) was used for database search by matching DIA spectra to the FASTA file path according to Proteome (i.e. UniProt H. sapiens canonical and isoform protein sequences, downloaded in May 2023), added with decoys and contaminants. MSFragger settings included peptide digestion with trypsin; for a peptide length from 7 to 45, a number of missed cleavages was set to 2, a precursor mass range from 300 to 5000 and the mass tolerance for precursor and fragment ions was set to 20 ppm. Commonly, methionine oxidation (M) and N-terminal acetylation ([^) were set as variable modifications and cysteine alkylation as a fixed modification. Validation settings of DIA_speclib_Quant were used, including MSBooster for deep learning prediction Peptide Spectrum Match (PSM) Validation with Percolator to increase confident identifications, and Protein Interference with ProteinProphet to generate the protein list with a 1% false discovery rate (FDR). The report.pg_matrix.tsv file from the FragPipe output was processed with RStudio (R version 4.4.0 (2024-04-24) x 86_64-apple-darwin20) using the packages limma and dplyr. A log₂ transformation was applied to the data to stabilize variance and normalize the data distribution. Next, a threshold was defined to filter rows based on the presence of valid (non-missing) values. Only rows where at least 75% of the values in each specified condition are valid were retained. The linear model was fitted to the data using the lmFit function from the limma package. An empirical Bayes moderation was performed to obtain more stable estimates of the variance. Moderated t-statistics and p-values for the comparison between two conditions were extracted, and the p-values were also adjusted for multiple hypothesis testing using the FDR method¹¹³. Finally, the log fold change between the two conditions was calculated for each comparison. All proteins found in control samples from cells expressing mCherry-p62 (mCh-p62) were considered non-specific contaminants and removed from the dataset before comparison of protein abundance in cells expressing T-mCh-p62.

All animal experiments were approved and performed according to the regulations of the LANUV, NRW, Germany and ethics committee guidelines (AZ: 81-02.04.2020.A418, 81-02.04.2023.VG076, 81-02.04.2021.A132, Anzeige §4.24.030).

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