Summary
Nuclear pore complexes (NPCs) mediate exchange across the nuclear envelope. They consist of hundreds of proteins called nucleoporins (Nups) that assemble in multiple copies to fuse the inner and outer nuclear membranes. Elucidating the molecular function and architecture of NPCs imposes a formidable challenge and requires the convergence of in vitro and in situ approaches. How exactly NPC architecture accommodates processes such as mRNA export or NPC assembly and turnover inside of cells remains poorly understood. Here we combine integrated in situ structural biology, correlative light and electron microscopy with yeast genetics to structurally analyze NPCs within the native context of Saccharomyces cerevisiae cells under conditions of starvation and exponential growth. We find an unanticipated in situ layout of nucleoporins with respect to overall dimensions and conformation of the NPC scaffold that could not have been predicted from previous in vitro analysis. Particularly striking is the configuration of the Nup159 complex, which appears critical to spatially accommodate not only mRNA export but also NPC turnover by selective autophagy. We capture structural snapshots of NPC turnover, revealing that it occurs through nuclear envelope herniae and NPC-containing nuclear vesicles. Our study provides the basis for understanding the various membrane remodeling events that happen at the interface of the nuclear envelope with the autophagy apparatus and emphasizes the need of investigating macromolecular complexes in their cellular context.
Text
Nuclear pore complexes (NPCs) are giant macromolecular assemblies with a very intricate architecture. About 30 different genes termed nucleoporins (Nups) encode for components of NPCs. Scaffold Nups contain folded domains and form a cylindrical central channel. This channel is lined with FG-Nups, which harbor intrinsically disordered FG-rich repeats that interact with nuclear transport receptors. Nups assemble in multiple copies to form an eight-fold rotationally symmetric complex, totaling in ~550 protein building blocks in yeast and ~1000 in mammals1. The NPC consists of two outer rings, also called nuclear (NR) and cytoplasmic rings (CR), which are placed distally to the inner ring (IR) that resides at the fusion plane between the inner and outer nuclear membranes. Within the three-ringed architecture, Nups are organized into subcomplexes that form specific substructures. The Y-complex is the key scaffolding component of the outer rings, whereas the IR scaffold is built by the inner ring complex2. Further, the yeast Nup159 complex (Nup214 complex in mammals) associates asymmetrically with the Y-complex at the CR and facilitates the terminal steps of mRNA export. Its core consists of two Nup159-Nup82-Nsp1 heterotrimers that dimerize into a characteristic P-shaped structure3,4.
Several studies have highlighted that the accurate spatial positioning of the Nup159 complex with respect to the central channel at the cytoplasmic face of the NPC is critical for the spatial organization and directionality of mRNA export3,5,6. While mRNAs are chaperoned through the FG repeats within the central channel in a Mex67-dependent (but Ran independent) manner, exported mRNPs encounter the ATPase activity of the DEAD-box RNA helicase Dpb5 at the cytoplasmic face that removes Mex67 and ratchets the RNA into the cytoplasm1,6,7. Dbp5 recruitment and positioning is ensured by the N-terminal beta propeller of Nup159, which is separated by a flexible linker from the C-terminal coiled coils that anchor Nup159 to the NPC scaffold7.
However, understanding how Nup subcomplexes are positioned to each other within the context of the nuclear membranes and the overall architecture of NPCs imposes a considerable challenge to structural biologists and requires the convergence of in vitro and in situ approaches8. While several high-resolution structures of Nups have been solved by X-ray crystallography, biochemical analysis and cross-linking mass spectrometry revealed the interaction topology. Cryo-electron microscopy (cryo-EM) maps provided an overall framework for the positioning of subcomplexes using systematic fitting approaches9,10,11,12,13. Cryo-electron tomography (cryo-ET) with subsequent subtomogram averaging is the method of choice to generate such cryo-EM maps, because it allows structural analysis of NPCs within the native context of the nuclear membranes that have to be considered as an integral part of their architecture. Such in situ structural analysis is available for both mammals and algae (reviewed in1) and revealed that key features of the NPC architecture, such as the stoichiometry of Y-complexes within the outer rings, are not conserved. Intriguingly, the P-shaped outline of the yeast Nup159 complex4 was not apparent in any of the cryo-EM maps available to date, either suggesting structural diversity, or alternatively, questioning its physiological relevance.
To date, in situ structural analysis is still missing for the S. cerevisiae NPC (ScNPC), which has been extensively studied as a model organism not only for Nup structure, NPC architecture and the mechanism of mRNA export, but also NPC surveillance and turnover1,14. Thus far, only detergent extracted, purified NPC species have been analyzed13 that bear neither the nuclear membranes nor any distinct features of Y-complexes (reviewed in1). An integrative model of the entire ScNPC architecture based on extensive experimental analysis in vitro has been recently put forward13 and in principle could enable structure-to-function analyses based on the very powerful yeast genetics approaches. However, it remains unknown to what extent the native architecture of the ScNPC has been preserved in vitro. Therefore, there has been a pressing need to elucidate the spatial organization of subcomplexes of the ScNPC in situ within the cellular context.
Structural analysis of the ScNPC in the cellular context
To characterize ScNPC architecture in situ, we prepared thin cryo-FIB lamellae of exponentially growing, plunge frozen cells15. We acquired 230 cryo-electron tomograms (Fig. 1a) containing ~500 ScNPCs and determined the structures by subtomogram averaging (see Methods). The resulting cryo-EM map at ~25 Å resolution (Extended Data Fig. 1a-b) provides the most detailed overview of the native configuration and conformation of subcomplexes of actively transporting NPCs in cells of any species to date (Fig. 1b-c, Extended Data Fig. 1b-c). A visual inspection of the structure reveals striking disparities, not only compared to the previously analyzed structures of the algal NPC10 and human NPC9 that likely account for species-specific differences, but also in comparison to previous cryo-EM maps of the biochemically isolated and detergent-extracted ScNPC (Extended Data Fig. 1c-d, Source Data 1 attached for review process). The previously published integrative model of one asymmetric unit of the IR13 fits unambiguously into the observed density (p-value 1.6*10-12 Extended Data Fig. 2), however the model of the entire IR has to be dilated by ~20 nm in diameter in order to fit the in situ conformation (Extended Data Fig. 3a), thereby spatially separating the eight individual spokes (Extended Data Fig. 3a). This analysis underlines the plasticity of the NPC within cells15, which might be physiologically relevant for the transport of large cargos and inner nuclear membrane proteins.
As expected13,16, 16 nuclear and cytoplasmic rings of Y-complexes are apparent in our cryo-EM map (Fig. 1c-d, Extended Data Fig. 1c), although some additional, yet unexplained density is observed in the nuclear ring. While the crystal structures and homology models of the yeast Y-complex vertex17 (Sec13–Nup145C–Nup84–ScNup120–ScSeh1–ScNup85) or its fragments fit into the observed Y-shape density with significant p-values (Extended Data Fig. 2), the previous integrative model of the Y-complex13 does not fit due to the curvature of its tail (Extended Data Fig. 3c). We therefore applied an integrative modeling procedure (Methods) to take the observed in situ conformation into account and obtained complete structures of the Y-complexes in both CR and NR. Due to the larger circumference of the NPC, the Y-complexes occupy a more extended conformation (Extended Data Fig. 3b). These findings further emphasize that the combination of in vitro and in situ structural analysis is essential to understand NPC architecture in physiological conditions.
A prominent feature of the CR, which remains yet unassigned after localization of the Y-complexes, is the P-shaped density of the Nup159 complex that is strikingly reminiscent of the p revious in vitro analysis of the isolated, negatively stained Nup159 complex3,4 (Fig. 2). Systematic fitting of the Nup159 complex negative stain map4 into our in situ structure confirmed this fit (Fig. 2b, p-value 0.0027, Extended Data Fig. 4a, see Methods). Based on the top resulting fit, we superimposed a previously published representative integrative model of the Nup159 complex3,13 and we locally fit it into our cryo-EM map (Extended Data Fig. 4b). In comparison to previous architectural models of the entire ScNPC3,13, the P-shape is flipped around the axis that points towards the central channel (Extended Data Fig. 4b). Thereby, the Nup82 ß-propellers are positioned towards the inner ring (Extended Data Fig. 4b). The previous architectural model was in part guided by distance restraints from crosslinking mass spectrometry. In our updated Nup159 complex configuration, all four of the previously published crosslinks13 between the Nup159 and Y-complexes that could be mapped to the structure are satisfied (Extended Data Fig. 4e). The arm of the P-shape that consists of Nup159 DID tandem repeats binding multiple Dyn2 dimers4,18 is clearly apparent in situ (Fig. 2). It projects towards the cytoplasm at an angle of ~45° with respect to the nucleocytoplasmic axis (Fig. 2), while it had been previously thought to rather face the IR3 (Extended Data Fig. 4b). Instead, the protein termini preceding FG-repeats of Nsp1 face the central channel.
Yeast genetics confirms orientation of the RNA export platform
To independently confirm this arrangement, we combined yeast genetics with in situ structural analysis. An important interactor of Nup82 is Nup116, one out of three yeast homologues of the essential vertebrate Nup98. This is a key Nup for establishing the NPC permeability barrier19,20, mRNA export6,19 and pre-ribosome translocation21. It is also an architecturally critical linker Nup that connects the inner ring, Y and Nup159 complexes through short linear motifs in its intrinsically disordered domains12,22,23. We superimposed the crystal structure of Nup116(966-1111)-Nup82(1-452)-Nup159(1425-1458)24 with the respective parts of two copies of Nup159 and Nup82 contained in the P-complex (Extended Data Fig. 4c-d). This analysis predicts that the autoproteolytic domain of Nup116 is placed into two yet unassigned densities that are proximate to each other and project towards Nup188 of the IR (Fig. 2c, Extended Data Figs. 4c-d and 5a). Crosslinking analysis13 and previous biochemical data suggesting that Nup116 links to Nup188 and Nup192 of the IR22,23 agree well with this configuration (Extended Data Figs. 4e and 5a). To validate this assignment, we structurally analyzed the ScNPC in nup116Δ cells (knock-out, k. o.) cells at permissive temperature, at which NPCs have an ordinary morphological appearance in electron micrographs25. We acquired ~120 tomograms and obtained a cryo-EM map as described above (see Methods). The nup116Δ structure lacks density exactly at positions proximate to Nup188 as predicted by superposition of the Nup116-Nup82-Nup159 X-ray structure (Fig. 2c and Extended Data Figs. 4c and 5c). This finding corroborates our spatial positioning of Nup116 at the interface between the IR and Nup159-complexes and previous biochemical work that has established Nup116 as a linker nucleoporin associated with the inner ring components22,23. The positioning of the two copies of Nup116 agrees with only one of the two that were previously assigned by integrative modeling13 (Extended Data Fig. 5b). Our new model of NPC architecture spatiotemporally accommodates extensive biochemical analysis of mRNP export5,7,19. In fact, it places the FG-repeats that interact with Mex67 and are relevant earlier during the process, more towards the equatorial plane. Dbp5 and the N-terminus of Nup159 that facilitate the terminal release of Mex67 are positioned more towards the cytoplasm26 (Fig. 2d).
The Nup159 complex mediates selective autophagy through nuclear envelope herniae
At nonpermissive temperature, the nup116Δ strain forms nuclear envelope (NE) herniae (Extended Data Fig. 6a), a phenotype in which NPCs are engulfed by the NE membranes and not accessible to the cytosol25. Interestingly, this phenotype was also observed when other Nup159 complex associated Nups, such as Gle2 were genetically perturbed27, and has been linked to the surveillance of NPC assembly by ESCRT proteins28,29. Since herniae were also observed under seemingly unrelated conditions, such as genetic perturbation of inner nuclear membrane proteins30,31 in both yeast and human cells32, their exact functional relevance remained to be determined14,33. We wondered if NPC architecture under the herniae is altered and addressed this using cryo-FIB milling, cryo-ET (~40 tomograms) and subtomogram averaging of NPCs at the herniae basis in the nup116Δ strain at non-permissive temperature. We found that in these trapped NPCs not only Nup116, but the entire cytoplasmic ring, including the Nup159 and Y-complexes are missing (Extended Data Fig. 6c). Concurrently, the nuclear envelope is less curved at the fusion plane, which is typical for the neck of herniae (Extended Data Fig. 6a-c). This result is in line with models that conceptualize herniae as failed inside-out assembly intermediates of NPCs, in which the lack of fusion of the NE membranes prevented the assembly of the cytoplasmic components (reviewed in14,33). Another finding further corroborates this model. We noticed that as compared to WT cells, the NE of nup116Δ. cells at permissive temperature contains a high number of mushroom-shape evagination of the inner nuclear membrane (Extended Data Fig. 6d-e). The structures are morphologically reminiscent of interphase assembly intermediates previously characterized in human cells34. These data point to a direct or indirect contribution of Nup116 to membrane fusion during NPC assembly.
A recent study identified Nup159 as an intrinsic receptor for selective autophagy of nuclear pores (Lee et al in revision; attached for review process). The AIM (Atg8-interacting motif), located proximate to the DID-Dyn2 arm of Nup159, mediates NPC turnover under conditions of low nitrogen supply. It recruits the ubiquitin-like autophagosomal protein Atg8 that is essential for anchoring autophagic protein cargo to autophagosomal membranes. Above findings suggest that the AIM is exposed into the cytoplasm together with the DID-Dyn2 arm (Fig. 3a), thus favoring the recruitment of cytoplasmic Atg8. We hypothesized that the herniae might be cleared by selective autophagy. This would make sense if they would not only occur upon genetic perturbation as previously observed, but also upon stress conditions that trigger autophagy, which was indeed the case. Under conditions of nitrogen starvation, NE herniae were observed, although they were less abundant as compared to the nup116Δ strain at nonpermissive temperature (Fig. 3b). How exactly herniae clearance could be mediated by the AIM of Nup159 in the nup116Δ background is however not obvious because Nup159 complex should not be present at the herniae according to the NPC structure determined in nup116Δ cells at not permissive temperature (Extended Data Fig. 6c).
To visualize how the autophagy machinery is recruited to the NPC, we employed a split-Venus approach specifically targeting the Nup159-Atg8 interaction. To enhance autophagic clearance of spatially clustered NPCs, we used a nup120D background35 as described in (Lee et al) but at shorter nitrogen starvation exposure (Fig. 3c-d and Extended Data Fig. 7a-b). We identified the regions of interest using correlative light and electron microscopy (CLEM) on plastic sections36 and 3D cryo-CLEM37. We acquired tomograms at positions of NPCs engaged in Atg8 interaction. These data revealed that agglomerates of herniae were engulfed by additional double membranes (Fig. 3c-d and Supplementary Video 1). Concurrently, the membrane topology of the herniae was more complex and displayed NPCs budding out of the NE, indicating additional membrane remodeling (Fig. 3c-d, Extended Data Fig. 7a and Supplementary Video 1). Thereby, many NPCs and thus AIMs are exposed to the cytoplasm (Fig. 3c-d, Supplementary Video 1). This finding would predict that herniae initially form independently from Atg8, possibly in the context of NPC surveillance28. Indeed, herniae formed in atg8Δ cells exposed to nitrogen starvation (Fig. 3b and Extended Data Fig. 8). We further noted, that under conditions where hernia formation was triggered but not selective autophagy, as in the nup116Δ strain at non-permissive temperature without starvation, herniae accumulate (Fig. 3b, Extended Fig. 6a) and are also observed as NPC-containing nuclear vesicles in the cytoplasm (Fig. 4c, Extended Data Fig. 9c and Supplementary Video 2), which agrees with the biochemical analysis of Lee et al Instead, under conditions in which selective autophagy is triggered by starvation, NPC-containing nuclear vesicles (~340 nm diameter) are surrounded by ribosomes and double membranes typical of autophagosomes (Fig. 4e-f and Extended Data Figs 7c and 9e-f), that are en route towards the vacuole as demonstrated by Lee et al (Extended Data Fig. 9g). Taken together, we observed several snapshots of NPC turnover, although the transitions between them, such as the budding of herniae into nuclear vesicles or fusion of autophagosome-engulfed NPC-containing-nuclear-vesicles to the vacuole need to be further investigated in the future (Extended Data Fig. 9). It will be also interesting to elucidate if herniae triggered under alternative genetic conditions, also in human cells32, ultimately feed into the autophagy pathway.
In conclusion, our in situ structural analysis of ScNPC clarifies the layout of the major Nup subcomplexes under native conditions. It confirms the physiological relevance of the P-shaped configuration of the Nup159 complex but identifies an unexpected orientation with respect to the scaffold that accommodates both the terminal steps of mRNA export (Fig 2d) and exposure of the intrinsic AIM (Fig. 4d). Our findings highlight the power of in cell structural biology to provide novel insights into fundamental processes of eukaryotic life.
Legends for Supplementary items
Video Supplementary 1: 3D cryo-CLEM tomogram of Nup159-Atg8 interaction. The video shows how clustered NPCs depicted in red are surrounded by double membranes depicted in yellow after 5,5h starvation. Fig. 3b for statistics and 3d for scale bar.
Video Supplementary 2: Cryo electron tomogram of nup116Δ strain #4 after shifting 5h to 37°C shows budded herniae. NPCs at the herniae are red, membranes are pink. Scale bar: 100 nm.
Extended Data Table 1: S. cerevisiae strains used in this study.
Source Data 1: Comparison of Cryo-EM maps of the ScNPC. Zip file for reviewers containing the in cell cryo-EM map of the ScNPC (this study), the cryo-EM map (EMD-7321) of detergent extracted NPCs previously obtained by Kim et al. and a Chimera session that displays both drawn to scale at suggested contour level (in cell cryo-EM map in grey, EMD-7321 in cyan).
Methods
Yeast strains, growth conditions, fluorescence imaging
Strains used in this study are listed in Extended Data Table 1. Fresh cells were grown to mid-log phase in YPAD (1% Yeast Extract + 2% Peptone (YP), adenine (A), 2% Dextrose (D)). Nitrogen starvation was carried out by switching cells grown at OD600 ~0.5 with YPAD into SD-N medium (synthetic minimal medium lacking nitrogen; 0.17% yeast nitrogen base, without aminoacids and ammonium sulfate, supplemented with 2% glucose) for indicated times. Fluorescent images of unfixed yeast cells upon starvation (strains #2, 3) were acquired using a Zeiss Axio Observer. Colocalization figure was made in FIJI38.
Correlative fluorescence and electron tomography in plastic sections
CLEM analysis was conducted as previously described39,40. S cerevisiae (Sc) cells strain #2 after ~5h of starvation were high pressure frozen (HPM010, AbraFluid) and freeze substituted (EM-AFS2, Leica) with 0.1% uranyl acetate in acetone for 55h at −90°C. The temperature was then raised to −45°C at 3.5°C/h and samples were further incubated for 5h. After rinsing in acetone, the samples were infiltrated in Lowicryl HM20 and the resin was polymerized under UV light. 300 nm sections were cut with a microtome (EM UC7, Leica) and placed on carbon coated 200 mesh copper grids (S160, Plano).
The fluorescence microscopy (FM) imaging of the sections was carried out as previously described36,40 using a widefield fluorescence microscope (Olympus IX81) equipped with an Olympus PlanApo 100X 1.40 NA oil immersion objective and a CCD camera (Orca-ER; Hamamatsu Photonics).
After fluorescence imaging, grids were poststained with uranyl acetate and lead citrate. 15 nm protein A-coupled gold beads were also added as fiducial markers used for overlaying low-mag with high-mag tomograms. 60 ° to 60 ° tilt series of the cells of interest were acquired semi-automatically on a Tecnai F30 (Thermofisher, FEI) at 300 kV with Serial-EM41 at 20000x and at 4700x to facilitate ease of correlation.
Tomograms were reconstructed automatically with IMOD batchruntomo42. Tilts were aligned by patch tracking. 3DMOD software was used for manual segmentation of the tomograms and for making the videos. Overlays of fluorescence spots and tomograms was performed with ec-CLEM Plugin43 in ICY44 by clicking manually on corresponding pairs of notable features in the two imaging modalities.
2D electron microscopy
Strains #1, 2, 3, 5 upon starvation were high pressure frozen as described above and freeze substituted with 0.2% uranyl acetate 0.1% glutaraldehyde in acetone for 60h at −90°C. The temperature was then raised to −45°C at 3°C/h and samples were further incubated for 9h. After rinsing in acetone, the samples were infiltrated in Lowicryl HM20 resin and the resin was polymerized under UV light. Acquisition was performed with a Tecnai F30 (Thermofisher, FEI) at 300 kV. Herniation counting was performed using 2D electron microscopy of 70-100 nm sections and 3D electron microscopy using 300 nm sections. Boxplots were done with BoxPlotR (http://shiny.chemgrid.org/boxplotr/) using around 150 nuclei counted per each strain or condition.
Cryo-FIB milling and cryo-CLEM
Sc wild type strain #1 and knock-out #4 were grown in YPAD liquid medium to an OD600 of 0.2 at 30°C, OD 0.5 for correlative studies using strain #2. Per grid, 3.5ul of cell suspension were applied on 200-mesh copper grids coated with R 2/1 holey carbon or SiO2 films (Quantifoil Micro Tools GmbH) and plunge frozen in liquid ethane at ~-186°C using a Leica EM GP grid plunger. The blotting chamber conditions were adjusted to 30°C, 95% humidity and 1,5-2 seconds blotting time. For cryo-CLEM, Crimson FluoSpheres™ Carboxylate-Modified Microspheres, 1.0 μm, crimson fluorescent (625/645), were washed in 1×PBS and added to the cell suspension at 1:20 dilution, and 1 second blotting time was used.
The frozen grids were fixed in modified autogrids to allow milling at a shallow angle45 and transferred into an Aquilos (cryo-FIB-SEM dual beam, Thermofisher). For correlative studies, the clipped grids were imaged on a prototype Leica cryo-confocal microscope based on Leica TCS SP8 CFS equipped with a Cryo Stage (similar to the commercially available EM Cryo CLEM Widefield system). Imaging was performed using a 50× objective, NA 0.90, 552 nm laser excitation, and detecting simultaneously at 560–620 nm and 735–740 nm.
In the Aquilos, samples were sputter-coated with inorganic platinum and coated with an organometallic protective platinum layer using the Aquilos gas injection system (GIS)46. Lamellae were produced using the Gallium ion beam at 30 kV and stage tilt angles of 17°-19° by milling in two parallel rectangular patterns. The lamella preparation was conducted in a step-wise fashion, gradually reducing the current of the ion beam until the final polishing of a thin slab of biological material of around 150-250 nm.
For correlative studies, beads (5 to 10) were picked in the squares of interested and overlaid with fluorescence signals coming from the confocal stacks using 3DCT in both the electron and ion beam images37. After choosing the signal of interest in the confocal stacks, 3DCT provides the position to place the milling patterns in x, y and zed. Milling is then performed as described above. After polishing, the signal of interest is retained in the lamellae provided that the milling did not result in deformation of the carbon film.
Autogrids with lamellae were unloaded and placed in storage boxes. In some cases, a further inorganic Pt layer was added for reducing charging during TEM imaging.
Cryo-electron tomography data collection and processing
Grids with lamellae were loaded into the Krios cassette. Cryo-electron tomographic tilt series (TS) were acquired on a Titan Krios (Thermofisher FEI) operating at 300 kV equipped with a Gatan K2 summit direct electron detector and energy filter. The autogrids were carefully loaded with the lamella orientation perpendicular to the tilt-axis of the microscope prior to TS acquisition.
Lamellae were mapped at low mag (2 nm/pix, 30 kV slit) and for correlative studies the maps were overlaid with the electron beam images from the Aquilos using serialEM registration points. Spots of interested were chosen accordingly.
All data was collected using the K2 operating in dose-fractionation mode at 4k x 4k resolution with a nominal pixel size of 0.34 nm. TS collection was automated at using a modified version of the dose-symmetric scheme47 taking the lamella pre-tilt into account. Defocus TS were acquired over a tilt range of +65 to −45 for a positive pre-tilt with a tilt increment of 2-3°, a total dose of ~140 e-/A2 and a targeted defocus of around 2 to 4.5 μm.
All images were pre-processed and dose-filtered as described in11. Tilt-series alignment was performed using IMOD software package 4.9.2.42, by patch tracking function on bin4 image stacks. Initial tilt series alignment was manually inspected and improved by removing contours showing large deviations (i.e. a large mean residual error) from the alignment model function. The software package gCTF48 was used for CTF estimation. 3dCTF correction and tomogram reconstruction was performed with NovaCTF49.
Subtomogram alignment and averaging
Subtomogram alignment and averaging was performed with slight modifications from a previously described workflow50 using the Matlab TOM package re-implemented in C++. In brief: Particle coordinates and initial orientations were manually picked and assigned. Initial NPC alignment was performed on 8 and 4 times binned subtomograms followed by a manual inspection and curation of the initial alignment for each particle. From the whole aligned NPCs all 8 spokes were assigned according to an 8-fold rotational symmetry and the subunits were extracted from the tomograms removing subunits which are located outside of the lamellae. Subunits were further aligned. The alignment was once more manually inspected, and all misaligned subunits were removed. Final alignments were done binning two times the subtomograms using focused masks on the cytoplasmic, inner and nucleoplasmic ring of the NPC.
In the S. cerevisiae wild type data set ~500 NPC (4000 asymmetric units) were initially picked from ~230 tomograms, ~200 NPCs (1600 asymmetric units) from 120 tomos for the nup116Δ strain at permissive temperature and ~40 NPCs from 40 tomos for the nup116Δ strain at not-permissive temperature (37°C). The particles were split in two half datasets for gold-standard processing. After alignment and manual curation 2000 subunits were used in the final average for the WT NPC and ~800 for the nup116Δ at permissive temperature and 320 for the nup116Δ strain at 37°C. For the latter NPCs filtered volumes were used for alignment and weighted back projection volumes to get the final structure. Gold standard Fourier shell correlation (FSC) was calculated with Fourier Shell Correlation Server (www.ebi.ac.uk/pdbe/emdb/validation/fsc/results/) with half-maps as inputs. Local resolution was calculated with Resmap51, B-factor sharpening was estimated empirically as in9 being between 2000 Å2 for the inner ring and 3000 Å2 for the cytoplasmic ring. For counting the number of inner nuclear membrane evaginations (Extended Data Fig. 6d-e), all the WT tomograms and ~120 tomograms from nup116Δ cells at permissive temperature were used.
Systematic fitting of ScNPC components to the cryo-ET map
An unbiased systematic (global) fitting approach was performed using structural models of various ScNPC subcomplexes derived from previously published structures3,13. All structural models were low-pass filtered to 40 Å prior the fitting. The resulting model maps were then independently fitted into the ScNPC cryo-EM map using global fitting as implemented in UCSF Chimera52. The fitting of the Y complex structures and Nup159 complex was performed for the isolated maps of the cytoplasmic and nuclear rings. The inner ring model was fitted to the IR map. The nuclear envelope density was erased prior to the fitting. The regions of the nuclear envelope distant from apparent contact points between membrane and protein densities were erased prior to the fitting to eliminate fits significantly overlapping with the membrane. All fitting runs were performed using 100,000 random initial placements and the requirement of at least 30% of the model map to be covered by the ScNPC density envelope defined at low threshold. For each fitted model, this procedure yielded between 500-17,000 fits after clustering.
For fitting the filtered atomic models, the cross-correlation about the mean (cam score, equivalent to Pearson correlation) score from UCSF Chimera was used as a fitting metric as in our previous work9,10,11,12. where uave is a vector with all components equal to the average of the components of u and vave is defined analogously.
The negative stain map of the Nup159 complex was fitted using Chimera’s cross-correlation about zero (equivalent to cosine similarity).
This score was used instead of the cam score because the cam score tests the linear dependence of the two fitted maps, which is not the case when fitting negative stain maps, which contain the signal of the electron density at the surface of the complex.
To confirm this fit, the Nup159 complex was also fitted using the colors program from the Situs package54 with settings appropriate for low resolution negative stain maps (i.e. using Laplacian filtering that emphasizes on contour matching over interior volume matching).
For each fitting run, the statistical significance of the fits was assessed as a p-value calculated from the normalized cross-correlation scores. To calculate the p-values, the cross-correlation scores were first transformed to z-scores (Fisher’s z-transform) and centered, from which two-sided p-values were computed using standard deviation derived from an empirical null distribution (derived from all obtained fits and fitted using fdrtool55 R-package). All p-values were corrected for multiple testing using Benjamini-Hochberg procedure. Figures were made using UCSF Chimera52 and Xlink Analyzer56.
Integrative modeling
To build the integrative model of ScNPC, the IR model from13 was fitted as a rigid body using the procedure described above. The models of Y-complex fitted into CR and NR could not be obtained by rigid body fitting of the published models13. Therefore, we have divided the available crystal structures and homology models13 into six smaller rigid bodies (with cut points corresponding to boundaries of published crystal structures of Y-subcomplexes) and fitted them simultaneously into the cryo-EM maps of the CR and NR using the integrative modeling procedure implemented with using Integrative Modeling Platform57 version 2.9.0 as described previously by11,53. Firstly, each of the rigid bodies has been independently fitted to the EM map using UCSF Chimera52 as described above (structural models were low-pass filtered to 40 Å) to generate libraries of alternative fits for each rigid body. Then, we generated configurations of all the rigid bodies by recombining the above fits using simulated annealing Monte Carlo optimization. Each configuration was generated by an independent Monte Carlo optimization comprising 30,000 steps resulting in total 20,000 models and scored. The scoring function for the optimization was a linear combination of the normalized EM cross-correlation scores of the precalculated domain fits, domain connectivity restraint, a term preventing overlap of the Y components with the nuclear envelope and the assigned Nup159 complex density) and clash score (see ref.11 for the implementation details). The structures were simultaneously represented at two resolutions: in Cα-only representation and a coarse-grained representation, in which each 10-residue stretch was converted to a bead. The 10-residue bead representation was used for the clash score to increase computational efficiency, the Cα-only representation was used for crosslinking and domain connectivity restraints. Since the EM restraint was derived from the original EM fits generated with UCSF Chimera, it was derived from the full atom representation. The final models for visualization were selected as the top scoring model.
Author Contributions
MA, CEZ conceived the project, designed experiments, collected data, analyzed data, wrote the manuscript. VR analyzed data, performed computational analysis, wrote the manuscript. FW, PR designed experiments, collected data, analyzed data. HKHF, CWL, XZ collected data, analyzed data. WH collected data. BT analyzed data. KK performed computational analysis. CWM, JM, YS supervised the project. BP, JK, MB conceived the project, designed experiments, supervised the project, wrote the manuscript.
Depositions
The EM maps associated with this manuscript will be deposited in Electron Microscopy Data Bank upon publication. The integrative model of ScNPC will be deposited in PDB-dev (https://pdb-dev.wwpdb.org/). An in cell cryo-EM map of the ScNPC is attached for the review process (Source Data 1).
Acknowledgments
We thank members of the Beck Lab, Mahamid Lab and EMBL’s electron microscopy core facility for invaluable input and support. We thank Leica for collaboration to develop the prototype cryo-confocal. We thank Amparo-Andres Pons, David Thaller, Patrick Lusk and Ed Hurt for critical discussions and for providing relevant strains. We thank Jiameng Sun for help in tomographic segmentation. The computational part of this work has been done on the High-Performance Cluster of the EMBL, supported by the EMBL Heidelberg IT Services. This work was supported by ERC Complex Assembly (MB and MA); MA was funded by an EMBO a long-term fellowship (ALTF-1389–2016); JM received funding from the European Research Council (ERC 3DCellPhase- 760067); KHF is supported by a fellowship from the EMBL Interdisciplinary (EI3POD) programme under Marie Skłodowska-Curie Actions COFUND (664726). MB acknowledges funding by EMBL, the Max Planck Society and the European Research Council (ComplexAssembly 724349).