Nanorulers
Custom-designed linear nanorulers of varying length (80, 60, 50, 30, 20 and 10 nm), carrying one Atto 647N molecule on each end, were purchased from GATTAquant.
Cell cultures
Hippocampal cultured neurons
Animals (Wistar rats, P0–P1) were treated according to the regulations of the local authority, the Lower Saxony State Office for Consumer Protection and Food Safety (Niedersächsisches Landesamt für Verbraucherschutz und Lebensmittelsicherheit), under the license Tötungsversuch T09/08. In brief, the hippocampi were dissected from the brains and washed with Hank’s balanced salt solution (14175-053, Invitrogen), before being incubated under slow rotation in a digestion solution containing 15 U per ml papain (LS003126, Worthington), 1 mM CaCl2 (A862982745, Merck), 0.5 mM EDTA and 0.5 mg ml−1 l-cysteine (30090, Merck) in DMEM. This procedure was performed for 1 h at 37 °C, before enzyme inactivation with a buffer containing 10% fetal calf serum (FCS) and 5 mg ml−1 BSA (A1391, Applichem) in DMEM. The inactivation solution was replaced after 15 min with the growth medium, containing 10% horse serum (S900-500, VWR International), 1.8 mM glutamine and 0.6 mg ml−1 glucose in MEM (51200046, Thermo Fisher Scientific), which was used to wash the hippocampi repeatedly. The neurons were then isolated by trituration using a glass pipette and sedimented by centrifugation at 80g (8 min). The cells were then resuspended in the same medium and seeded on poly(l-lysine) (PLL)-coated coverslips for several hours, before replacing the buffer with Neurobasal A culture medium (10888-022, Thermo Fisher Scientific) containing 0.2% B27 supplement (17504-044, Thermo Fisher Scientific) and 2 mM GlutaMAX (35050-038, Thermo Fisher Scientific). The neurons were then maintained in a humidified incubator (5% CO2, 37 °C) for at least 14 days before use.
Conventional cell cultures
Tubulin immunostaining was performed in the U2OS cell line, obtained from the Cell Lines Service (CLS). The cells were grown in a humidified incubator (5% CO2, 37 °C) in DMEM (D5671, Merck) with the addition of 10% FCS (S0615, Merck), 4 mM glutamine (25030-024, Thermo Fisher Scientific) and an antibiotic mixture added at 1% (penicillin–streptomycin; Thermo Fisher Scientific). For imaging purposes, cells were grown overnight on PLL-coated coverslips (P2658, Merck).
Brain slices
We dissected rat brains from P0–P1 rat pups (Wistar). The brains were then fixed with 4% PFA (30525894, Merck) in PBS for 20 h. The fixed brains were then placed in agarose (4% solution; 9012366, VWR Life Science), before cutting to the desired thickness (100–200 µm) using a vibratome.
Participants
Participants were in treatment at the Paracelsus Elena Klinik. They were diagnosed with PD according to standard criteria47,48,49. Neurological control participants were diagnosed with a variety of non-neurodegenerative disorders. A detailed presentation of participants, their ages and their diagnoses can be found in Supplementary Table 1. The informed consent of all of the participants was obtained at the Paracelsus Elena Klinik, following the principles of the Declaration of Helsinki.
CSF samples
CSF samples were collected at the Paracelsus Elena Klinik following identical standard operating procedures. CSF was obtained by lumbar puncture in the morning with the participants fasting and in sitting position. The CSF was processed by centrifugation at 2,000g for 10 min at room temperature; aliquots of supernatant were frozen within 20–30 min and stored at −80 °C until analysis. Samples with a red blood cell count > 25 µl−1 or indication for an inflammatory process were excluded.
Immunostaining procedures
Tubulin immunostaining
U2OS cells were first incubated with 0.2% saponin (47036, Sigma-Aldrich) to extract lipid membranes. This procedure was performed for 1 min in cytoskeleton buffer, consisting of 10 mM MES (M3671, Merck), 138 mM KCl (K42209636128, Merck), 3 mM MgCl2 (M8266-100G, Sigma-Aldrich), 2 mM EGTA (324626-25GM, Merck) and 320 mM sucrose at pH 6.1. The cells were then fixed using 4% PFA and 0.1% glutaraldehyde (A3166, PanReac) in the same buffer. Unreacted aldehyde groups were quenched using 0.1% NaBH4 (71320, Sigma-Aldrich now Merck) for 7 min in PBS, followed by a second quenching step with 0.1 M glycine (3187, Carl Roth) for 10 min in PBS. The samples were blocked and simultaneously permeabilized using 2% BSA and 0.1% Triton X-100 (9036-19-5, Sigma-Aldrich) in PBS (room temperature, 30 min). Primary anti-tubulin antibodies (T6199 Sigma-Aldrich; 302211, Synaptic Systems; 302203, Synaptic Systems; ab18251, Abcam) were applied for 60 min at room temperature and were then washed off with permeabilization buffer, followed by an incubation of the samples with secondary antibodies (ST635P-1001, Abberior). Five washes were performed with permeabilization buffer followed by three PBS washes (each for 10 min) before continuing with cellular expansion.
PSD95 immunostaining
Neurons were fixed with 4% PFA in PBS (D8537-500ML, Thermo Fisher Scientific) for at least 30 min before quenching with 50 mM glycine (in PBS) for 10 min and blocking and permeabilizing using 2.5% BSA (9048-46-8, Sigma-Aldrich), 2.5% normal goat serum (NGS) and 0.1% Triton X-100 (1003287133, Sigma-Aldrich) in PBS (30 min at room temperature, unless specified otherwise). The antibodies and/or primary Nbs were diluted in 2.5% BSA and 2.5% NGS in PBS and added to coverslips for 60 min at room temperature. This was followed by washing with the permeabilization buffer (30 min, three buffer exchanges) and by incubation with the primary Nb FluoTag-X2 anti-PSD95 (clone 1B2; N3702, NanoTag Biotechnologies) for 1 h at room temperature. Specimens were then washed five times with permeabilization buffer before a final wash with PBS (15–30 min, three buffer exchanges), followed by expansion procedures.
Immunostaining of CSF samples
CSF probes were obtained from persons with PD and controls at the Paracelsus Elena Klinik and stored at −80 °C before use. Then, 20 µl of CSF was placed on BSA-coated coverslips, enabling the sedimentation of multiprotein species overnight at 4 °C. Fixation with 4% PFA (10 min, room temperature) and quenching with 50 mM glycine (10 min, room temperature) were followed by the application of anti-ASYN antibodies (128211 and 128002, Synaptic Systems) or ASYN Nb2 (SynNb2 (ref. 45), custom-produced and fluorescently conjugated by NanoTag) for 1 h at room temperature in 2.5% BSA in PBS buffer. For the case of antibodies, secondary Abberior STAR 635P was applied for 1 h at room temperature. Five washes with 2.5% BSA in PBS were followed by mild postfixation with 4% PFA for 4 min and expansion procedures.
Brain slice immunostaining
The fixed brain slices were first quenched using 50 mM glycine (in PBS), followed by three washes with PBS (each for 5 min) and blocking and permeabilization in PBS containing 2.5% BSA and 0.3% Triton X-100 for 120 min at room temperature. The primary antibodies used (anti-bassoon, ADI-VAM-PS003-F, Enzo Life Sciences; anti-Homer 1, 160003, Synaptic Systems) were diluted in the same buffer (lacking Triton X-100) to 2 µg ml−1 and added to the slices overnight at 4 °C. Three washes with PBS (each for 5 min) removed the primary antibodies, enabling the addition of secondary antibodies conjugated with Abberior STAR 635P (ST635P-1001, Abberior) for Basson identification. The secondary antibodies were diluted to 1 µg ml−1 in PBS containing 2.5% BSA and incubated for 3 h at room temperature. The brain slices were finally subjected to five washes with PBS containing 2.5% BSA (each wash for 5 min), followed by two final 5-min washes in PBS.
GFP–Nb complex (TSR) generation
The monomeric (A206K) and nonfluorescent (Y66L) EGFP (mEGFP*) was modified to have an ALFA tag on its N terminus and a HaloTag on its C terminus (ALFA-EGFP-HaloTag). This construct was expressed in a NebExpress bacterial strain and it had an N-terminal His-tag, followed by a bdSUMO domain, which enabled the specific cleavage of the His-tag31 after the purification procedures. Bacteria were grown at 37 °C with shaking at 2g in Terrific Broth (TB) supplemented with kanamycin. Upon reaching an optical density (OD) of ~3, the temperature was reduced to 30 °C and bacteria were induced using 0.4 mM IPTG, with shaking for another ~16 h. Bacteria lysates were incubated with Ni+ resin (Roche, cOmplete) for 2 h at 4 °C. After several washing steps, the ALFA-mEGFP(Y66L)-HaloTag protein was eluted by enzymatic cleavage on the column using 0.1 µM SENP1 protease for 15 min. Protein concentration was determined using Nanodrop (Thermo Fisher Scientific) and purity was assessed by Coomassie gels. Complex formation was performed by mixing the following for 1 h at room temperature in a final volume of 40 µl: 25 pmol of ALFA-EGFP-HaloTag and 30 pmol of three different single-domain antibodies: FluoTag-Q anti-ALFA (N1505), FluoTag-X2 anti-GFP (clone 1H1; N0301) and FluoTag-X2 anti-GFP (clone 1B2), all from NanoTag Biotechnologies. The control experiments were performed using a similar procedure without including the target protein ALFA-EGFP-HaloTag. The expression and purification of EGFP used in Supplementary Figs. 15 and 16 were performed as previously described50. Briefly, NebExpress Escherichia coli strain (New England Biolabs) was cultured in TB at 37 °C and induced using 0.4 mM IPTG for 16 h at 30 °C. Bacteria pellets were sonicated on ice in 50 mM HEPES pH 8.0, 500 mM NaCl, 5 mM MgCl2 and 10% glycerol. After removing cell debris by centrifugation, the lysate was incubated for 1 h with cOmplete His-tag purification resin (Roche) at 4 °C. After washing the resin in batch mode with more than ten column volumes, eGFP was enzymatically eluted using 0.1 µM SUMO protease. Concentration was determined by absorbance at 280 nm using the molecular weight and extinction coefficient of eGFP. Purified protein was diluted in 50% glycerol and stored in small aliquots at −80 °C.
PAGE
A primary mouse monoclonal antibody to synaptobrevin 2 (104211, Synaptic Systems) and a secondary antibody conjugated to Abberior STAR 635P (ST635P-1002-500UG) were mixed with reducing 2× Laemmli buffer (63 mM Tris-HCl pH 6.8, 2% SDS, 100 mM DTT and 20% glycerol) and heated for 10 min at 96 °C. The denatured and reduced samples were then loaded in a self-cast Tris-glycine 12% polyacrylamide gel and 10 µg of total protein was loaded per lane. Electrophoresis was run at low voltage at room temperature. The gel was briefly rinsed using distilled water and fluorescence was read on a GE-Healthcare AI 600 imager using a far-red filter (Cy5 channel). Next, the gel was submerged for 4 h in Coomassie brilliant blue solution to stain all proteins, followed by incubation with destaining solutions, before finally being imaged using the same GE-Healthcare AI 600 gel documentation system.
Dot blot
In a stripe of nitrocellulose membrane (GE-Healthcare), 5 mg of BSA and 1 µg of ALFA-tagged EGFP-Y66L-HaloTag were spotted and left to dry at room temperature. Membranes were then blocked in PBS supplemented with 5% skim milk and 0.05% Tween-20 for 1 h with tilting and shaking. FluoTag-X2 anti-GFP Cy3 (clone 1B1), FluoTag-X2 anti-GFP Abberior STAR 635P (clone 1H1) and Fluotag-X2 anti-ALFA Abberior STAR 635P (all from NanoTag) were used at 2.5 nM final concentration in PBS with 5% milk and 0.05% Tween-20 for 1 h with gentle rocking. After 1-h incubation at room temperature while protected from light, five washing steps were performed each using 2 ml of PBS supplemented with 0.05% Tween-20 for a total of 30 min. Membranes were finally imaged using a GE-Healthcare AI 600 system.
1,6-Hexanediol treatments
1,6-Hexanediol (240117-50G, Sigma-Aldrich) was diluted in neuronal Neurobasal A culture medium at 3% for 2 min and 10% for 12 min before fixation and further processing for immunostaining.
Purified proteins
IgA and IgM were purchased from Jackson ImmunoResearch and IgG was purchased from Abberior (AffinityPure IgA 109-005-011, ChromePure IgM 009-000-012 and ST635P-1001, respectively) and all immunoglobulins were diluted in PBS before expansion procedures. Otoferlin was produced according to standard procedures51 and was diluted in 20 mM HEPES, 100 mM KCl and 0.05% DDM buffer, before being used at 0.4 mg ml−1 concentration. For GABAARs, a construct encoding the full-length human GABAAR β3 subunit (UniProt P28472) with an N-terminal TwinStrep tag was cloned into the pHR-CMV-TetO2 vector52. A lentiviral cell pool was generated in HEK293S GnTI-TetR cells as described previously53. Cells were grown in FreeStyle 293 expression medium (12338018, Gibco) supplemented with 1% FBS (11570506, Gibco), 1 mM l-glutamine (25030149, Gibco), 1% NEEA (11140050, Gibco) and 5 μg ml−1 blasticidin (ant-bl-5b, Invivogen) at 37 °C (130 r.p.m., 8% CO2) and induced as described54. Following collection by centrifugation (2,000g, 15 min), the cell pellets were resuspended in PBS pH 8 supplemented with 1% (v/v) mammalian protease inhibitor cocktail (Sigma-Aldrich). Cell membranes were solubilized with 1% (w/v) DDM (D3105GM, Anatrace) for 1 h. The insoluble material was removed by centrifugation (12,500g, 15 min) and the supernatant was incubated with 300 μl of Strep-Tactin Superflow resin (IBA Lifesciences) while rotating slowly for 2 h at 4 °C. The beads were collected by centrifugation (300g, 5 min) and washed with 150 ml of 0.04% (w/v) DDM and PBS pH 8. The sample was eluted in 2.5 mM biotin, 0.02% (w/v) DDM and PBS pH 8 and used for imaging at 1 mg ml−1 concentration. For the purification of the GABAAR in complex with the β3-specific Nb (Nb25)55, Nb25 was fluorescently labeled with STAR 635P at the N and C termini, generating Nb25-STAR 635P. Then, 20 μl of 10 μM Nb25-STAR 635P was added to the sample before the elution step and incubated for 2 h at 4 °C while rotating. The excess Nb25-STAR 635P was removed by washing the beads with six bed volumes of 0.04% (w/v) DDM and PBS pH 8, eluted with 2.5 mM biotin, 0.02% (w/v) DDM and PBS pH 8 and used for imaging at 3 mg ml−1 concentration. The same procedure was applied for the negative control anti-eGFP Nbs. To test that Nb25-STAR 635P could still bind the receptor, 2 μM Nb25-STAR 635P was added to the β3 homomeric receptor reconstituted in nanodiscs as described previously56. Next, 3.5 μl of the sample was applied to a freshly glow-discharged (PELCO easiGlow, 30 mA for 120 s) 1.2/1.3 UltrAuFoil grid (Quantifoil), which was blotted for 2.5 s and plunge-frozen using a Leica EM GP2 plunger at 14 °C and 99% humidity. Imaging was performed at the Medical Research Council (MRC) Laboratory of Molecular Biology on a Titan Krios G2 microscope equipped with an F4 detector in electron counting mode at 300 kV at a nominal magnification of 96,000×, corresponding to a calibrated pixel size of 0.824 Å. A total of 300 movies were collected using EPU (Thermo Fisher Scientific, version 2.0–2.11) with a total dose of 38 e− per Å2 and 6.43 s of exposure time. The movies were motion-corrected using MotionCor2 (ref. 57). Contrast transfer function estimation was performed with CTFFIND-4.1.13 (ref. 58). Particle picking was performed using a retrained BoxNet2D neural network in Warp59, followed by 2D classification in cryoSPARC60. Calmodulin was purified as previously described61 and was used in calcium-free buffer (150 mM KCl, 10 mM HEPES and 5 mM EGTA) or calcium-containing buffer (150 mM KCL, 10 mM HEPES and 2 mM CaCl2) at pH 7.2 before expansion procedures. Briefly, calmodulin 1 (mRNA reference sequence number NM_031969.2) was tagged with mEGFP and an ALFA tag for affinity purification purposes. The construct was transfected in HEK293 cells using Lipofectamine 2000 (11668019, Invitrogen) following the manufacturer’s protocol. After expression for ~24 h, the cells were lysed in PBS buffer containing 1% Triton X-100, 2 mM EDTA and a protease inhibitor cocktail. The debris was removed by centrifugation and the supernatant was added to an ALFA Selector PE resin (NanoTag Biotechnologies), where it was allowed to bind for 60 min (4 °C, under rotation). After two washes with lysis buffer and one wash with PBS (ice-cold), the bound proteins were eluted by adding the ALFA peptide. The purified protein was analyzed by Coomassie gel imaging as previously described61.
X10 expansion procedures
X10 expansion of cultured cells was performed using proteinase K exactly as described in the protocol article16. X10 expansion relying on autoclaving (X10ht62) was performed as follows. The samples were incubated overnight with 0.3 mg ml−1 Acryloyl-X (A-20770, Thermo Fisher Scientific) in PBS pH 7.4 at room temperature. The samples were then subjected to three PBS washes (5 min each) while preparing the gel monomer solution as previously described16. The solution was pipetted on parafilm and was covered with upside-down coverslips containing cells or with brain slices that were then also covered with fresh coverslips. Polymerization was allowed to proceed overnight at room temperature in a humidified chamber. Homogenization of proteins and single molecules was performed using 8 U per ml proteinase K (P4850, Sigma-Aldrich now Merck) in digestion buffer (800 mM guanidine HCl, 2 mM CaCl2 and 0.5% Triton X-100 in 50 mM Tris; 8382J008706, Merck) overnight at 50 °C. Homogenization of cell cultures and brain slices was performed by autoclaving for 60 min at 110 °C in disruption buffer (5% Triton X-100 and 1% SDS in 100 mM Tris pH 8.0) followed by a 90-min incubation to cool the temperature to safe levels. Before autoclaving, the gels were washed first in 1 M NaCl and then at least four times in disruption buffer for a total time of at least 120 min. Gel expansion was then performed by washing with double-distilled water (ddH2O) for several hours, with at least five solution exchanges. Expansion was performed in 22 × 22-cm square culture dishes, carrying 400–500 ml of ddH2O. When desired, the samples were labeled using a 20-fold molar excess of NHS-ester fluorescein (46409, Thermo Fisher Scientific) in NaCHO3 buffer at pH 8.3 for 1 h before the washing procedure that induced the final expansion.
ZOOM expansion procedures
Following a previously described protocol63, fixed U2OS cultured cells were incubated in anchoring solution (25 mM acrylic acid NHS-ester in 60% (v/v) DPBS and 40% (v/v) DMSO) for 60 min. Afterward, cells were moved to monomer solution (30% (w/v) acrylamide and 0.014% (w/v) N–N′-methylenbisacrylamide in PBS buffer). After 60 min, the gelation process was started by adding initiators (0.5% (w/v) TEMED and 0.5% (w/v) APS) to the monomer solution. The hydrogel–cell hybrid was homogenized in detergent solution (200 mM SDS and 50 mM boric acid in deionized water, with the pH titrated to 9.0) at 95 °C for 15 min, followed by 24 h at 80 °C. ZOOM-processed samples were then stained using the previously mentioned anti-α-tubulin antibodies (1:400 in PBST).
mCLING expansion
For mCLING gelation, we started with 2 µl of mCLING-Atto 647N (710 006AT1, Synaptic Systems), originally reconstituted to a concentration of 1.0 nmol ml−1 and mixed with 2 µl of 10 mg ml−1 Acryloyl-X, before bubbling with N2 gas for a few minutes to purge oxygen. This mixture was incubated overnight at 4 °C and then mixed with 100 µl of freshly prepared X10 polymer solution. Next, 80-µl aliquots of this gel-sample mixture were placed on parafilm in a humidified chamber and were covered with a clean 18-mm coverslip. Homogenization was carried out by X10 proteinase K digestion protocol, as previously described. Gels were then postexpansion labeled with NHS-ester fluorescein (46409, Thermo Fisher Scientific) or NHS-ester STAR 635P (07679-1MG, Sigma-Aldrich). Images were acquired using HyD X detectors on a STELLARIS 8 microscope.
mCLING structure simulation
The equilibrium structure of mCLING peptide-bonded to Atto 647N was assessed using molecular dynamics simulations with the AMBER99 force field64. The molecule was simulated in water using the TIP4P/EW model65 in a cubic system of length 6 nm with periodic boundaries. The topology for the fluorophore was generated using ACPYPE66, which interfaces with Antechamber from the AMBER suite of tools to create compatible topology files. The molecular dynamics package GROMACS64 was used with the leap-frog algorithm to integrate Newton’s equations of motion with a time step of 1 fs. Conditionally convergent long-range electrostatic interactions were calculated by the smooth particle mesh Ewald method with a cutoff distance of 1.2 nm. Lennard–Jones interactions were assessed using a single cutoff distance of 1.2 nm, supplemented by long-range dispersion corrections for both energy and pressure. After energy minimization, the system was equilibrated for 300 ns, followed by a 300-ns production run. The pressure was fixed at 1 bar by the Parrinello–Rahman barostat.
Microscope systems
For image acquisition, small gel fragments were cut and placed in the imaging chamber presented in Supplementary Fig. 7. Paper tissues were used to remove any water droplets around the gels, before enabling the gels to equilibrate for at least 30 min on the microscope stage. Epifluorescence imaging was performed using an Olympus IX83 TIRF microscope equipped with an Andor iXon Ultra 888, ×100 (1.49 numerical aperture (NA)) TIRF objective and Olympus LAS-VC four-channel laser illumination system. Confocal imaging was performed for most experiments using a TCS SP5 STED microscope (Leica Microsystems) with a ×100 (1.4 NA) HCX Plan Apochromat STED oil-immersion objective. The LAS AF imaging software (Leica) was used to operate imaging experiments. Excitation lines were 633, 561 and 488 nm and emission was tuned using an acousto-optical tunable filter. Detection was ensured by PMT and HyD detectors. Images were taken using a resonant scanner at 8-kHz frequency. The five-dimensional (5D) stacks for zONE were performed using a 12-kHz resonant scanner mounted on a Leica TCSSP8 Lightning confocal microscope. Samples were excited with a 40% white-light laser at wavelengths of 633, 561 and 488 nm and acquisitions were carried out using HyD detectors in unidirectional xyct line scans or in unidirectional and bidirectional xyczt line scans.
Image acquisition
Objectives of 1.4, 1.45 and 1.51 NA were used to acquire images with a theoretical pixel size of 98 nm. For a higher resolution, the theoretical pixel size was set to 48 nm at the cost of a slightly lower detection rate. Images acquired on the camera-based system had a predetermined pixel size of 100 nm. The acquisition speeds were 20–40 ms and 25 ms on resonant scanners of 8 and 12 kHz and on a camera, respectively, for xyct. For hyperstacks of xyczt acquisitions, images were acquired using 8-kHz and 12-kHz scanners in bidirectional mode (after the necessary alignments), allowing an achieved speed of 16 kHz and 24 kHz, respectively. Images of 8-bit depth were acquired at a line format ranging from 128 × 128 to 256 × 256. The scanning modality on a confocal was set to ‘minimize time interval’ (Leica LAS software). To maintain natural fluctuations of fluorophores, we did not use line accumulation or line averaging during scanning. A frame count from 200 up to 4,000 was acquired. We recommend a frame count of at least 1,500–2,000 for optimal computed resolution in xyct scans and 200–1,000 for xyczt scans for volume reconstructions.
Image processing
ONE image processing is enabled through a Java-written ONE Platform under ‘ONE microscopy’ in Fiji. The ONE microscopy plugin uses open-source codes from Bioformats Java library, NanoJ-Core, NanoJ-SRRF, NanoJ-eSRRF and Image Stabilizer12,13,67,68. ONE plugin supports multiple video formats of single or batch analyses in xyct. Hyperstacks with 5D xyczt format were processed with the zONE module. This module allows the user to select the optical slices and channels to resolve at ultraresolution. Upon irregularities in resolving one or more channels within one or more planes, zONE leaves a blank image and computes the remaining planes within a stack. The image processing is fully automated and requires minimal initial user input. Aside from the expansion factor, preset values and analysis modalities are automatically provided (see Supplementary Fig. 1 for more details). The ONE plugin has a preinstalled safety protocol to skip failures in computations or uncompensated drifts, without affecting the progress of batch analysis. Data analyses, parameters and irregularities are reported in log files. The ONE plugin automatically linearizes the scale on the basis of radiality magnification and expansion factor corrections. In addition, ONE offers the possibility to correct for chromatic aberration by processing multichannel bead images as a template that is applied to super-resolved images of the biological samples. The correction is performed by applying a modified Lucas–Kanade algorithm67. For the ONE microscopy plugin to store complex multidimensional images from hyperstacks, we modified the Java code of the ImageJ library and adapted it locally. The ONE Platform source code and plugin are available from GitHub (https://github.com/Rizzoli-Lab/ONE-Microscopy-Java-Plugin). For best performance, we recommend to download a preinstalled version on Fiji, available from the same link. The ONE plugin comes with predefined parameters optimized for single molecules, particularly emphasizing the highest resolution. Next to each parameter, the user will find explanations and recommendations. When the cursor hovers over the parameters, pop-up text bubbles provide further details. Users can adjust all parameters as desired. Importantly, the expansion factor should be set in accordance with the results obtained in the respective laboratories because this parameter is particularly important for obtaining the correct image scale. In addition, the temporal analysis mode should be adjusted in accordance with the type of experiment performed. For example, the temporal radiality pairwise product mean (TRPPM) analysis suits continuous and diffuse signals, while temporal radiality autocorrelation (TRAC) analysis is recommended for sparse labels and for colocalization studies requiring higher resolution. A TRAC order of 4 is preset for the analysis of single molecules because it provides the highest achievable resolution. For colocalization analysis, we recommend using the chromatic aberration correction function. The resulting images have an additional suffix of ‘_CAC’ (for chromatic aberration corrected). Additional parameters are available in the advanced options tab, which can be used to accommodate various experimental paradigms with different SNR and signal quality. When acquiring zONE images, where image quality becomes noisier and the acquisition rate slows down because of imaging in multiple axial planes, users may choose to analyze the images using a lower TRAC order of 3 or 2. However, users should note that, while zONE allows the collection of information across a volume, this comes at the cost of reducing the achieved resolution because of hardware limitations. Lastly, we recommend that the users thoroughly read Supplementary Fig. 30, in which we present the software in graphic format, and Supplementary Figs. 1 and 2, in which the imaging and analysis flowcharts are shown.
Image analysis and statistics
For single-object analyses, such as synaptic vesicle or antibody analyses, signal intensities and distances between objects were analyzed manually using ImageJ (W. Rasband and contributors, National Institutes of Health). Line scans were also performed and analyzed using ImageJ. For the analysis of PSDs (Fig. 2), spots were identified by thresholding bandpass-filtered images, relying on empiric thresholds and bandpass filters, organized in the form of semiautomated routines in Matlab (version 2017b). Spots were overlaid to determine their overall signal distributions or their center positions were determined to measure distances between spots (in the same or different channels). The same procedure was used for the averaging analysis of CSF samples (Fig. 4) and for the analysis of spot distances for the GFP–Nb assemblies (Supplementary Figs. 15 and 16). Full width at half maximum values were measured after performing line scans over small but distinguishable spots (Supplementary Fig. 16), followed by Gaussian fitting using Matlab. The averaging analysis of GABAARs is presented in detail in the main text and was performed using Matlab. In brief, receptors were detected automatically as particles with intensities above an empirically derived threshold. To remove particles with uncompensated drift, we eliminated all receptors coming from images in which a large proportion of the particles were oriented similarly. We then visually inspected all of the remaining particles to choose those that appeared to be in a ‘front view’, showing a reasonably round appearance, with Nbs placed at the edges of the receptor (visible in the second color channel). All particles were centered on the intensity maxima of the respective GABAAR channel images. The particles were subjected to an analysis of the peaks of fluorescence, using a bandpass procedure, followed by identification of maxima69; the positions of the peaks were calculated to below-pixel precision and were rounded off to a pixel size of 0.384 nm (the starting pixel size was 1 nm). These positions were then mapped into one single matrix, which represents the ‘averaged receptor’, as indicated in the main text. Averaging analyses of actin were performed similarly. In brief, actin strands were selected manually and were overlaid to generate average views. Model objects were generated as a comparison by convoluting the amino acid positions in the respective Protein Data Bank (PDB) structures with empirically derived ONE spots. All of these analyses were performed using Matlab. The SNR for single Nbs was determined by measuring the average pixel intensities within the Nb spots and away from them and then dividing the two measurements. Identically sized circular regions of interest, sufficient to capture the Nb spots completely, were used for both signal and background (noise) regions. Plots and statistics were generated using GraphPad Prism 9.3.1 (GraphPad Software), SigmaPlot 10 (Systat Software) or Matlab. Statistical details are presented in the respective figure captions. Figures were prepared with CorelDraw 23.5 (Corel Corporation).
Optimization
Overview of critical steps in ONE microscopy
The gel preparation for ONE microscopy in classical ExM cell imaging closely follows the recommendations in the X10 guide, which we published several years ago16. Here, we highlight briefly the crucial steps for ONE microscopy, which include anchoring, homogenization and oxygen purging. Proper anchoring is vital for maintaining labeled targets and fluorescence signals. Effective homogenization prevents the rupture of cell compartments and enables the proper expansion of proteins. To troubleshoot this step, one may consider tuning the strength of the homogenization process by testing both autoclave and proteinase K protocols. Milder digestion methods, including short autoclave times (<60 min) or trypsin-based digestion (instead of proteinase K), could also be considered. Improper oxygen purging results in inconsistent sticky gels, with varying expansion factors that are hard to handle. For optimal results, the user should always add the reaction initiator KPS and the catalyst TEMED to the polymer solution in a rapid fashion and then the gel amount used (typically 70–80 µl for an 18-mm coverslip) should be sealed off with a coverslip within, at most, 70 s. When preparing more than five gels simultaneously, we suggest having two people perform this step side by side to minimize oxygen exposure. In the special case of single-molecule analyses, it is crucial to work only with a thin film of fluid containing the molecules to be analyzed, to which the gel solution is quickly added. Please be aware that thin films of protein-containing buffers tend to dry very rapidly. An indicator of failure in this step is the appearance of salt and protein precipitates, looking as white clumps, which will be visible on the coverslip.
Imaging chamber optimization
All of the chamber blueprints and data are available in the Supplementary Information. For chamber usage, a gel slightly larger than the chamber should be cut, before removing excess water and fitting the gel onto the stabilizing net. Any overhanging gel should be trimmed away. The tight gel–chamber fit minimizes drift but automated drift correction in the ONE plugin is also available to address any residual drift before processing. It is automatically implemented and operates independently for each color channel. If the correction fails for one channel, it attempts to implement the drift correction coordinates from another channel. The interchannel drift correction feature is exclusive to line-by-line scanning and should not be used in frame-by-frame or stack-by-stack scan modes. Users suspecting postcorrection artifacts should sum the intensity of the entire drift-corrected raw video. Comet effects in the summed images indicate a drift correction failure, suggesting the need to discard such acquisitions. Drift correction issues often stem from dim or poorly labeled specimens or strong vibrations from an unstable imaging system.
Optimizing objective type selection
For targets in cells, which are close to the glass–gel interface, or single molecules, oil objectives with NA ≥ 1.4 should be used. For optimal imaging of single molecules, which are typically less than 1 µm in size when expanded, high-NA oil objectives should be used. Additionally, maintaining an imaging distance of ≤5 µm, by removing excess water between the gel and glass surfaces, is essential. To image cellular targets at higher depths accurately, it is crucial to address the refractive index mismatch. Using water-immersion objectives for deeper specimens is recommended to reduce artifacts.
Microscope selection
The user should consider the resolution needed and the type of specimen analyzed before settling on a particular microscope. In general, confocal microscopes are preferred. However, for general cellular imaging, epifluorescence microscopes are sufficiently accurate. Confocal microscopes offer higher resolution for single molecules and should be preferred for such uses. When using a confocal microscope, optimal results are achieved with the following detectors: HyD detectors, especially HyD X for its high quantum yield and SNR, or HyD R for near-infrared applications in photon counting mode (avoid analog and digital modes). Gallium arsenide phosphide and Avalanche photodiodes are also recommended. Classical photomultiplier tubes can be used at moderate voltage with a corrected smart offset to minimize dark counts to 1–5 per field of view.
Imaging conditions to avoid
During sample preparation, imaging single molecules from sticky gels or gels with cracks should be avoided, while ensuring that the expansion factor is corrected using known structures as rulers. For sample imaging, using noisy detectors with high dark counts should be avoided. Bidirectional scanners without manual phase shift correction should also be avoided. When processing images, users should be wary of artifactual airy disks caused by brightly labeled molecules that are partially out of focus. We suggest to opt for NHS-ester fluorescein over bright and stable modern dyes for labeling multimeric protein complexes, as bright parts of large complexes may get out of focus and lead to artifacts. The lower photon output of fluorescein reduces this problem.
Software considerations
The generated images have a 32-bit depth with negative values. These negative values represent noise and should be ignored. The users should set the dynamic display range to a zero-value minimum to exclude the noise. If gridded patterns appear in processed images, this may indicate low SNR, out-of-focus signals or incorrect bidirectional line scanning. Such images should be discarded. One can troubleshoot this by optimizing the labeling and the fluorophore selection and/or by adjusting the pixel dwell time and detector sensitivity.
3D model reconstruction
To prepare the ONE images for suitable 3D model reconstruction, we applied automated thresholding algorithms to extract dense areas of intensity, in which the expected protein should be located. The extracted areas have a window size of 200 × 200 pixels. The next step involved deconvolving the images using the Lucy–Richardson70 method with 80 iterations and a Gaussian PSF kernel of size 13 × 13 and σ = 2. Subsequently, the images were normalized to a range of 0–8 and then scaled down using bilinear interpolation to dimensions of 128 × 128 pixels. The processed images were transferred into cryoFIRE, an unsupervised ab initio autoencoder for complex shape reconstruction with amortized inference35. cryoFIRE consists of two components, the encoder \({f}_{\rm{enc}}\) and decoder \({f}_{\rm{dec}}\). The encoder contains convolutional followed by fully connected layers. It takes a processed ONE image \({Y}_{i}\) and estimates its pose \({R}_{i}\), translation \({t}_{i}\), expansion factor \({e}_{i}\) and molecule confirmation \({z}_{i}\) (that is, \({f}_{\rm{enc}}\left({Y}_{i}\right)=\left({R}_{i},{t}_{i},{e}_{i},{z}_{i}\right)\)). Here, \({e}_{i}\) was added to the original cryoFIRE approach to account for mild variations in the expansion factor between different gels. The decoder, a coordinate-based multilayer perceptron, represents the protein structure implicitly. For a given 3D coordinate, its output represents the density of the protein at this location. The decoder gets a 2D grid of coordinates, centered at the origin, which gets rotated and scaled by \(\left({R}_{i},{e}_{i}\right)\); therefore, the predicted image is \({\hat{Y}}_{{k}_{x},{k}_{y}}={f}_{\rm{dec}}({z}_{i},{e}_{i}\cdot {R}_{i}\cdot {\left({k}_{x},{k}_{y},0\right)}^{T})\) with \(\left({k}_{x},{k}_{y}\right)\in {R}^{2}\). This prediction is then shifted by \({t}_{i}\) to move it back to the original position. Because the predicted output represents a 2D central slice of the molecule in the Hartley domain, to compare the prediction \(\hat{Y}\) to the input Y, it also needs to be transformed into the Hartley domain. Because of the deconvolution in the preprocessing step, we did not need to apply a contrastive transfer function to the prediction, as proposed in cryoFIRE. With the modified (symmetric) mean squared error loss, which takes account of the handedness of the protein, the parameters are optimized using stochastic gradient descent. The 3D reconstructed images can be inspected with UCSF ChimeraX. The computation and processing were hosted by the Norddeutscher Verbund für Hoch- und Höchstleistungsrechnen servers (https://hlrn.de/).
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
