Structure of cytoplasmic ring of nuclear pore complex by integrative cryo-EM and AlphaFold.

Sample preparation for cryo-EM

X. laevis has played a key role in revealing the NPC structure because each oocyte has a large number of NPC particles ( 11 , 14 , 15 , 18 , 56 ). Freshly isolated stage VI oocytes of X. laevis in the modified Barth’s saline (MBS, 10 mM HEPES at pH 7.5, 88 mM NaCl, 1 mM KCl, 0.82mM MgSO 4 , 0.33mM Ca(NO 3 ) 2 and 0.41 mM CaCl 2 ) were purchased and shipped overnight from Ecocyte Bioscience US LLC. To optimize the homogeneity of the NPC sample, we incubated these oocytes with 100 mg/ml Actinomycin D (ActD) at 4°C overnight to inhibit RNA synthesis and thus RNA export for synchronization of the transport cycles ( 18 ). Each oocyte was poked at the animal pole using a sharp tweezer to result in the ejection of the nucleus, and transferred into a low salt buffer containing ActD (LSB, 10 mM HEPES at pH 7.5, 83 mM KCl, 17 mM NaCl and 7.5 μg/ml ActD). The nucleus was further washed to reduce the contaminating yolk in a new LSB solution. Two or three washed nuclei were then transferred to the surface of a freshly glow-discharged grid. The NE was poked open, spread using glass needles, incubated for 10 min in 10 μl of LSB supplemented with Benzonase Nuclease (Sigma Aldrich, E8263) to remove the contaminating chromatin, and subsequently washed twice with 10 μl of LSB. 3 ml LSB was added to the grid before blotting it for 3 to 5 s under 100% humidity at 4°C and plunged into liquid ethane using a Mark IV Vitrobot (ThermoFisher). Negative staining EM Nuclear membranes were applied on a freshly glow-discharged grid, using a Pelco EasyGlow, as described for cryo-EM sample preparation. Excess buffer was blotted on filter paper, and 6 μl of a 1% uranyl formate solution was applied for 30 s and blotted again on filter paper. Negatively stained samples were imaged on a Joel JEM1400 Transmission Electron Microscope at 120 keV.

Cryo-EM data collection

Screening and collection were performed at Stanford-SLAC Cryo-EM center (S2C2) with a Titan Krios electron microscope (Thermo Fisher Scientific) operating at 300 keV equipped with a K3 detector and a BioQuantum energy filter (Gatan, slit width 20 eV). Movies were collected in counting mode at a 1.4 Å pixel size ( table S1 ). Because of the way the grids were made, most NPC particles would have a similar orientation with their eightfold axis perpendicular to a grid, and we were expected to use a series of stage tilt angles to alleviate this orientation bias for 3D reconstruction. Given the known knowledge that gold grids can minimize beam-induced movement ( 57 ), we tested a number of gold grid types with the goal of identifying one with smallest beam-induced movement that is often exaggerated at high tilt angles. These grids include Lacey carbon films on gold support, 300 mesh (Ted Pella), Quantifoil holey carbon films on gold support, R 1.2/1.3, 300 mesh (Quantifoil Micro Tools), UltrAuFoil holy gold films on gold support, R 1.2/1.3, 300 mesh (Quantifoil Micro Tools) and UltrAuFoil holy gold films on gold support overlaid with graphene (made by Wei Li Wang in the Wu lab). Lacey carbon films on gold support were shown to be the most stable and thus used for all data collection. To alleviate the orientation bias, we initially collected datasets at stage tilts of 0°, 35°, and 45° with a total dose of 54 e/Å 2 over 40 frames for 0° and 35°, and a total dose of 79.8 e/Å 2 over 60 frames for 45°. An ideal tilt angle of 42° was then calculated using cryoEF ( 58 ) from a preliminary 3D reconstruction, and was used for the subsequent data collection with a total dose of 80 to 140 e/Å 2 over 80 to 120 frames. SerialEM was used for fully automated data collection, with a defocus range between −1 and −3 μm.

Cryo-EM data processing

Data processing leveraged computer support from the SBgrid Consortium ( 59 ). Movies were corrected by gain reference and beam-induced motion, and summed into motion-corrected and dose weighted images using the Relion 3.08 implementation of the MotionCor2 algorithm ( 60 , 61 ). The distribution of average motions per frame for each grid type at a given tilt angle was plotted using OriginLab (OriginPro 2017 Suite, OriginLab Corporation, Northampton, MA, USA) to evaluate grid-dependent drift performance. The initial contrast transfer function (CTF) estimation of motion-corrected micrographs without dose-weighting was calculated by CTFFIND4 ( 62 ). All micrographs were manually inspected and selected based on particle uniformity and contrast, and particles were picked manually. Gctf ( 63 ) was then used to determine the per-particle defocus values ( 63 ), from which 3D plots composed of the X and Y coordinates and the CTF (Z) of the particles for selected tilt images were generated using OriginLab (OriginPro 2017 Suite, OriginLab Corporation, Northampton, MA, USA). A plane was then fit to each 3D plot of a given image ( fig. S1B ). A total of 204,551 particles were manually picked, local CTF-corrected and extracted from 30,987 dose-weighted micrographs using a box size of 330 by 300 pixels at a 4× binned pixel size of 5.6 Å in RELION 3.08 ( 61 ). These particles were imported into cryoSPARC ( 64 ) to perform 2D classification, from which 124,532 good particles were selected and merged for homogeneous refinement. The published cryo-EM map of the human NPC (EMD-3103) ( 16 ) was low-pass filtered to 60 Å and used as the initial model. The homogeneous refinement with C8 symmetry resulted in a reconstruction at 22.1 Å. These reconstructed 124,532 particles were exported to RELION, 3.08 extracted again with a box size of 660 by 660 pixels and a binned pixel size of 2.8 Å, and imported back into cryoSPARC to re-perform 2D classification. 101,366 particles were selected for homogeneous refinement using the 22.1 Å map low-pass filtered to 40 Å as the initial model. The homogeneous refinement with C8 symmetry resulted in a 19.8 Å map. Particle density subtraction with the aligned 101,366 particles for separate processing of the CR or the NR was done in cryoSPARC. The new local refinement in cryoSPARC using the subtracted particles and a NR or a CR mask led to NR and CR maps at 14.7 and 14.6 Å resolutions, respectively. The aligned 101,366 particles for the whole NPC were also exported to RELION 3.08 and ran auto-refine with local search and C8 symmetry, with the 19.8 map low-pass filtered to 30 Å as the initial model. The resolution of the auto-refined map was 19.5 Å. We then performed C8 symmetry expansion and density subtraction using a CR protomer mask, and these subtracted particles were recentered and box size re-windowed to 300 by 300 pixels, all in RELION 3.1. 3D classification using a CR protomer mask, local search with 50 iterations and K = 6 was done on these subtracted particles. A class with 333,214 particles was selected for auto-refine with a mask and local search, reaching an 11.1 Å resolution. CTF refinement accounting beam-tilt estimation, anisotropic magnification estimation and per-particles defocus estimation and the subsequent auto-refine resulted in an improved map at 9.9 Å resolution. Additional reconstructions using a tight CR protomer mask or a tight core region mask led to maps at 8.8 and 8.4 Å resolutions. These aligned 333,214 subtracted particles were also imported into cryoSPARC to perform local CTF refinement and local refinement. The final resolutions for the CR protomer and the core region were 6.9 Å and 6.7 Å, respectively. All reported resolutions were estimated based on the gold-standard FSC = 0.143 criterion ( fig. S2 ). All final maps were corrected and sharpened by applying a negative B factor using automated procedures in RELION 3.1. Local resolution variations of cryo-EM maps were estimated using Phenix. Prediction of NPC subunit structures by AlphaFold The AlphaFold structures in this study were mainly generated from the AlphaFold2 implementation in the ColabFold notebooks ( 49 ) running on Google Colaboratory ( 21 , 22 ), using the default settings with Amber relaxation (msa_method=mmseqs2, homooligomer=1, pair_mode=unpaired, max_msa=512:1024, subsample_msa=True, num_relax=5, use_turbo=True, use_ptm=True, rank_by=pLDDT, num_models=5, num_samples=1, num_ensemble=1, max_recycles=3, tol=0, is_training=False, use_templates=False). The major difference of ColabFold from the native AlphaFold2 implementation is that ColabFold uses mmseqs2 ( 65 ), which the ColabFold authors suggest give equivalent results ( 22 ). For complex prediction, sequences were entered in tandem and separated by a semicolon. For coiled coil prediction, we used homooligomer=6. Due to computing memory constraints on Google Colaboratory, we sometimes split up large proteins at disordered junctions to predict each segment separately. AlphaFold was run once with each of the 5 trained models; the five models generated were checked for consistency, and unless specified otherwise, the top-ranked model was taken in each case for density fitting. AlphaFold computes pLDDT score and pTM score to indicate the accuracy of a prediction ( 23 ). We used pLDDT for ranking single protein models and pTM for ranking protein-protein complexes, as recommended by ColabFold ( 22 ). A predicted alignment error map between pairs of residues was also calculated for each prediction, which represents confidence in domain positioning. Confidence metrics (global and per-residue pLDDT, pTM, and PAE maps) of predictions made in this work can be found in tables S2 to S4 . A few larger proteins or complexes (more than 1400 residues in total length) were run on a Boston Children’s Hospital GPU cluster, by using default AlphaFold settings. To color ribbon diagrams based on per-residue pLDDT scores (range 0 to 100, with higher being better), these scores stored at the B-factor column of the .pdb files were changed to 100-pLDDT; thus, when colored as pseudo-B-factors in Pymol ( 66 ), a light spectrum from blue to red corresponds to highest to lowest pLDDT scores. Model fitting and building Prior to beginning modeling, we used AlphaFold ( 21 , 22 ) to generate all models of known components of the CR using the specific X. laevis sequences. An initial model of the Y-complex (PDB ID: 6LK8) ( 14 ) was fitted into the cryo-EM density using ChimeraX ( 67 ), and used as a reference for manual positioning of AlphaFold-generated subunit or complex structures into the density followed by executing the “fit in map” command to refine the replacement. Flexible loops were removed to avoid steric clash. After building the two Y-complexes, we began to model the other densities. Nup205 cryo-EM density was easily recognized behind the Y-complexes due to the large size and overall shape. Inner and outer Nup205 assume a different position due to the presence of the Nup214-Nup88-Nup62 complex in the inner Y-complex. Nup358 density was easily recognized in the presence of the generated AlphaFold model with a prominent S shape, and allowed for identification of 5 copies for each CR protomer. Nup88 density was recognized due to the β-propeller and the long a-helix. The additional density which belongs to the Nup214 β-propeller was recognized upon generation of its AlphaFold model. Building of the Nup88-Nup214-Nup62 complex was assisted by predicting the hetero-trimeric coiled coil stricture in AlphaFold, from which a composite model of the Nup88-Nup214-Nup62 complex was obtained. The final model was compared with the previous atomic model (PDB ID: 6LK8) ( 14 ). The model fitting quality was estimated for each subunit by the correlation coefficient in ChimeraX ( 67 ) and in Phenix ( 68 ). A value of correlation coefficient ranges from −1 to 1, with 1 as the perfect fit, and 0.5 to 1.0 as good fit. This modeling process using AlphaFold is reminiscent of the use of stereochemical information of amino acids and nucleic acids in the current practice of structural modeling ( 53 ) that increases model accuracy.

Nup358 expression and purification

X. laevis Nup358 constructs (residues 1–800 and 1–900) were cloned into pET21a with a C-terminal His tag. Expression was carried out in E.coli BL21 DE3. Briefly, cells were grown in terrific broth media, supplemented with 100 μg/ml of Ampicillin and 30 μg/ml of Chloramphenicol, until OD 600 reached 0.6. Cells were then transferred at 4°C for 30 min before the addition of 1 mM IPTG and incubation overnight at 18°C. Cells were pelleted at 3,000 g for 20 min and resuspended in lysis buffer (50 mM Tris-HCl pH 8.0, 150 mM NaCl, 1 mM TCEP, 10 mM Imidazole) supplemented with a protease inhibitor cocktail. Lysis was performed by sonication and the soluble fraction was separated by centrifugation at 40,000 g for 1 hour at 4°C. The supernatant was incubated with Ni-NTA beads pre-equilibrated with lysis buffer, and purification was performed per manufacturer’s recommendation. Eluted fractions were further separated by gel filtration chromatography with a Superdex 200 Increase 10/300 GL in gel filtration buffer (20 mM Hepes pH 7.4, 150 mM NaCl, 0.5 mM TCEP). Fractions were analyzed by Western blotting using an Anti-His antibody (Takara 631210). The Superdex 200 Increase 10/300 GL column was previously calibrated in gel filtration buffer using a high molecular weight kit from MW of 43 kDa to 669 kDa (Cytiva 28-4038-42).

Supplementary Material Supplementary Data

Data and materials availability: All data and materials reported in the main text and supplementary materials are available upon reasonable request. The electron density maps have been deposited in the Electron Microscopy Data Bank (EMDB) with accession numbers EMD-25817 and EMD-25818 for a CR protomer and a full CR ring built from the CR protomer map, respectively, and the atomic coordinates have been deposited in the Protein Data Bank with the accession number 7TDZ.