Systematic Protein–Protein Interaction Analysis Reveals Intersubcomplex Contacts in the Nuclear Pore Complex*

The nuclear pore complex (NPC) enables transport across the nuclear envelope. It is one of the largest multiprotein assemblies in the cell, built from about 30 proteins called nucleoporins (Nups), organized into distinct subcomplexes. Structure determination of the NPC is a major research goal. The assembled ∼40–112 MDa NPC can be visualized by cryoelectron tomography (cryo-ET), while Nup subcomplexes are studied crystallographically. Docking the crystal structures into the cryo-ET maps is difficult because of limited resolution. Further, intersubcomplex contacts are not well characterized. Here, we systematically investigated direct interactions between Nups. In a comprehensive, structure-based, yeast two-hybrid interaction matrix screen, we mapped protein–protein interactions in yeast and human. Benchmarking against crystallographic and coaffinity purification data from the literature demonstrated the high coverage and accuracy of the data set. Novel intersubcomplex interactions were validated biophysically in microscale thermophoresis experiments and in intact cells through protein fragment complementation. These intersubcomplex interaction data provide direct experimental evidence toward possible structural arrangements of architectural elements within the assembled NPC, or they may point to assembly intermediates. Our data favors an assembly model in which major architectural elements of the NPC, notably the Y-complex, exist in different structural contexts within the scaffold.

The nuclear pore complex (NPC) enables transport across the nuclear envelope. It is one of the largest multiprotein assemblies in the cell, built from about 30 proteins called nucleoporins (Nups), organized into distinct subcomplexes. Structure determination of the NPC is a major research goal. The assembled ϳ40 -112 MDa NPC can be visualized by cryoelectron tomography (cryo-ET), while Nup subcomplexes are studied crystallographically. Docking the crystal structures into the cryo-ET maps is difficult because of limited resolution. Further, intersubcomplex contacts are not well characterized. Here, we systematically investigated direct interactions between Nups. In a comprehensive, structurebased, yeast two-hybrid interaction matrix screen, we mapped protein-protein interactions in yeast and human. Benchmarking against crystallographic and coaffinity purification data from the literature demonstrated the high coverage and accuracy of the data set. Novel intersubcomplex interactions were validated biophysically in microscale thermophoresis experiments and in intact cells through protein fragment complementation. These intersubcomplex interaction data provide direct experimental evidence toward possible structural arrangements of architectural elements within the assembled NPC, or they may point to assembly intermediates. Our data favors an assembly model in which major architectural elements of the NPC, notably the Y-complex, exist in different structural contexts within the scaffold. Nucleocytoplasmic transport of molecules is universally mediated by nuclear pore complexes (NPCs) 1 , ϳ40 -112 MDa assemblies composed of ϳ 30 different proteins (nucleoporins, or Nups). NPCs perforate the double-layered nuclear envelope. Because of an eightfold rotational symmetry around the central transport channel, each Nup is assumed to be present in multiples of eight copies, totaling ϳ500 individual proteins per NPC (1)(2)(3). Nups can be functionally classified: those with extended and presumed unstructured phenylalanine-glycine repeat elements that line the principal transport channel, those with membrane-interacting elements that anchor the NPC, and architectural Nups that form the principal NPC scaffold. Crystallographic studies of scaffold Nups have revealed that they are largely composed of ␤-propellers and stacked ␣-helical domain elements (4,5). At least three architectural subcomplexes can be biochemically defined: the heteromeric Y-complex (6 -10), the heteromeric (yeast) Nic96/ (human) hsNup93 complex (11,12) and the trimeric Nsp1/ hsNup62 complexes (13)(14)(15)(16)(17).
The supramolecular structure of the NPC is intensively studied by cryoelectron tomography (cryo-ET), revealing distinct features characteristic for the cytoplasmic and nuclear side of the pore. Details of the scaffold ring structure are becoming apparent at ϳ2-6 nm resolution, in particular a distinction between three main elements-a central ring sandwiched between a cytoplasmic and a nucleoplasmic ring. Cumulative evidence from many experiments and laboratories suggests that the central ring is likely built by the Nic96/ hsNup93 and the Nsp1/hsNup62 complexes, while both the cytoplasmic and nucleoplasmic rings of the scaffold are predominantly made up of Y-complexes.
Because of the resolution gap between the crystal structures and the tomographic reconstructions and an incomplete set of available crystal structures, obtained from a variety of evolutionarily distant species, docking approaches are still incomplete (9,10,18). This is further complicated by the paucity of intersubcomplex interaction data (12, 18 -21).
These interactions are presumably weak, highly dynamic, or both, thereby evading detection by typical approaches like copurification from fractionated cell extracts. Comprehensive binary protein-protein interaction information between Nups may thus be a critical piece of information in the determination of a high-resolution structural model of NPC.
Several technologies have been introduced for highthroughput (HTP) mapping of protein-protein interactions (PPIs), including affinity purification coupled to mass spectrometry and yeast two-hybrid (Y2H) assays. Using these HTP-PPI mapping approaches, relatively large proteomewide data sets have been generated (22). Also, more focused approaches were successful in systematically mapping specific cellular processes, selected functional groups of proteins, multiprotein complexes, or specific types of interactions (23,24). Y2H approaches result in binary interaction information that is complementary to copurification-based approaches and thus often provide hints as to how complexes assemble or dynamically remodel in the course of cellular processes (25). Importantly, different HTP-Y2H setups, including ours, have been empirically assessed, demonstrating that PPI data sets are obtained with high precision (26).
In order to systematically study protein-protein interactions between all Nups, we performed a pairwise HTP-Y2H interaction screen involving yeast and human Nups, respectively. Our NPC interaction map recapitulated the intrasubcomplexes interactions with high coverage and additionally revealed several connections between Nups of distinct subcomplexes, i.e. intersubcomplex PPIs. The protein interaction map contains decisive information for structural model building of the NPC.

Comprehensive High-Resolution Y2H PPI Screening for
Yeast and Human Nups-To better understand how Nups interconnect to build the NPC requires systematic analysis of binary PPIs between all Nups. Therefore, we set out to clone domain constructs as well as full-length open reading frames (ORFs) representing all Nups from yeast and human to perform a pairwise Y2H interaction analysis for the two species (Fig. 1). Because most Nups are soluble proteins, a stringent, well-controlled, high-quality Y2H approach is an optimal choice. Importantly, we wanted to obtain binary interaction information, including weak, more-transient association, which is best obtainable by Y2H analysis at large scale (26 -28). Since all proteins used in the Y2H screen are bona fide NPC members, the search space to be examined is well defined and relatively small, thus the fraction of false positive interactions is expected to be very low.
In addition to the full length ORFs, we generated structurebased Y2H clones for yeast Nups (Supplemental Table S1), specifically removing phenylalanine-glycine repeat amino acid stretches that are thought to be aggregation prone and potentially more promiscuous in binding. The domain structure, FIG. 1. Nuclear pore proteins of S. cerevisiae and H. sapiens. Yeast (Sc) and human (Hs) homologous proteins are subdivided according to different substructures in the NPC (4) and color coded. Number of different constructs for a protein used in the study is given after the pipe (͉). based on crystallographic data and structure prediction, was considered in the design of protein fragment constructs. For instance, ␤-propeller domains were kept intact and ␣-helical domains designed matching existing structures. If crystal structures or homology models were unavailable, fragments were tested in vitro for acceptable chromatographic behavior. This approach should substantially decrease the number of false negatives and give rise to high-resolution information with regard to interacting protein domains. To study the binary interactions of human Nups, we subcloned full length cDNAs into Y2H vectors. In addition to covering 100 and 83% of the bona fide yeast and human Nups, respectively (29,30), our clone sets also included accessory factors, like nuclear transport receptors and Ran. Including full-length proteins and structure-based fragments, the yeast set included a total of 66 constructs. 24 human Nups were covered with full-length clones ( Fig. 1). 173 nonautoactive bait constructs (each protein as N-terminal and C-terminal DNA-binding domain fusion) and 128 prey constructs were examined six times in all possible pairwise combinations (ϳ 130,000 tests) ( Fig. 2A). Considering only colonies that grew two or more times, the Y2H approach resulted in 79 interactions involving 14 human and 38 yeast constructs (Supplemental Table S1). While the various constructs provide more detailed information about binding regions, PPIs detected with multiple combinations of constructs and bait-prey configurations were combined into a data set that comprised 39 interactions between 25 yeast Nups, 15 interactions between 14 human Nups, and four cross-species interactions (Fig. 2B).
We next compared our data with two studies published by other labs that report binary interaction information between Nups. Leducq et al. (36) used a protein fragment complementation assay (PCA), based on the reconstitution of the enzyme dihydrofolate reductase, to assess interactions in two different yeast species. At the author's index value cut off, the resulting data set comprises 44 PPIs between 14 Nups for S. cerevisiae in the common interaction space and 12 interactions were found in the overlap (Fig. 2B). Amlacher et al. (37) investigated structural components of the NPC, namely members of the Y-and the Nic96-complexes, and performed a Y2H analysis on the respective proteins reporting 19 binary interactions, six of which are common to our data set ( Fig.  2B). Thus, the overlap of our data was 13-16% with respect to the two other sets.
In a comprehensive study, Alber et al. (38) integrated ultracentrifugation, quantitative immunoblotting, coaffinity purification (co-AP), overlay assay, electron microscopy, immunoelectron microscopy, and fractionation data to produce an integrated structural model of the NPC. An interaction network defined through contact frequencies between Nups was calculated from the ensemble of optimized structures. Contact frequencies reflect the likelihood that a protein interaction is formed in the NPC, with values from 0 -0.25, 0.25-0.75, and 0.75-1 binned as low, medium, and high frequencies (38). In agreement, a low fraction of our interactions was found to have low contact frequencies (15 of 296), and a high fraction of PPIs had high contact frequencies (8 of 26). Protein pairs with medium contact frequencies in the Alber data were also frequently found to interact (13 of 84) (Fig. 3B). However, the set of interactions with medium contact frequencies also contains five of the standard reference interactions from crystallization studies, suggesting that the importance of the medium frequency interactions may have been underestimated during the model building process.
Besides the integrative modeling approach taken, the co-AP approach by Alber et al. provided the most systematic PPI data set to date, reporting 82 affinity purifications of individually Protein A-tagged scNups (38). The protein composition of the purified complexes was resolved via SDS-PAGE/mass spectrometry analysis and composites covering all scNups were reported. These cocomplex data do represent completely independent PPI information but cannot be directly compared with our binary interaction data as such. For a comparison of our Y2H data with the co-AP data set, we transformed the composite matrix into a pairwise co-AP score weighting the number of co-occurrences of two proteins in a composite (25). Higher co-AP scores are given to protein pairs involving proteins that are found more rarely (i.e. in fewer complexes) and in more-stringent purification preparations FIG. 2. Comprehensive Y2H screening of yeast and human Nups. (A) Ten selective agar plates from the Y2H screens for the indicated sc-bait proteins (N-or C-terminal DNA binding domain fusion proteins, respectively), tested against a 384-prey plate with three independent colonies of every prey. Colony growth indicated an interacting bait-prey pair (colored arrow heads for sc-prey identity). Pairs that grew at least twice were considered for assembly of the final data set. (B) Summary of the Y2H interaction. Rows: Interacting yeast and human protein pairs (protein level) are listed. Sc-interacting pairs are separated from Hs-interacting pairs and four cross species sc-hs pairs. Columns: Y2H, Y2H (i.e. in complexes with fewer proteins). Co-AP scores for all possible protein pairs were obtained ranging from 0 (proteins never purify together) to 1 (proteins exclusively purify as heterodimers, e.g. scNup42-Gle1). We then compared the co-AP scores with our Y2H data. The average co-AP score of the interacting pairs in our data set is much higher than in randomized networks with the same size and degree distribution, keeping the number of interactions for each protein constant (Fig. 3C, z-score ϭ 6.6). For comparison, the average co-AP score for the data sets of Leducq and of Amlacher differed from their randomized networks too, albeit with smaller zscores (z-score ϭ 2.7 and 3.3, respectively, Supplemental Fig. S2). This indicates very good agreement of our binary Y2H data with the systematic co-AP study performed by Alber et al. (38), providing unbiased comprehensive validation of our data set.
The NPC PPI Network: Intra-and Intersubcomplex PPIs-Nups are largely conserved between human and yeast ( Fig.  1A), and, though details will be species specific, it is assumed that the human NPC is similar to the yeast NPC both in its structural arrangement and in its overall shape (3,39). Therefore, we investigated interactions of human and yeast Nups in parallel creating data that complement each other. The Homo sapiens Y2H network consists of 15 PPIs between 14 hsNups and is thus sparser than the yeast data set that comprises 39 PPIs between 25 scNups (Fig. 4). This was expected as only full-length human Nups were used in the analysis, which tend to be large and are therefore prone to yield false negative results. As proof in point, we did not obtain interactions for hsNup160 (162 kDa) and hsNup155 (155 kDa). On the other hand, three well-studied interactions were found in the human and the yeast screens: (i) the Y-complex interaction between hsNup133-Nup107 (7,40,41), (ii) the hsNup54 -Nup62 and scNup57-Nsp1 interactions in the hsNup62/scNsp1-complex (15)(16)(17)42), and (iii) the hsNup93-Nup35 interaction interolog to the scNic96 -Nup53 interaction (37,43). Similar to the hsNup133-Nup107 interaction that complemented the yeast network, additional human Nup interactions were found. Either the homologous interactions were not detected with the yeast proteins or the interactions involved proteins absent in S. cerevisiae such as the facultative Y-complex protein Nup37. For example, interactions between the ␤-propeller proteins hsSec13, hsSeh1, and hsNup37 were detected (Fig. 4).
Except for the scNup157/scNup170 orthologs, all yeast scaffolding Nups had at least one interaction in our network. In agreement with current interaction knowledge, more "intra" interactions, i.e. between Nups within one subcomplex, were found than "inter" interactions, i.e. between Nups of different subcomplexes. For example, in the yeast network three to four intra-Y-complex interactions are expected at random while the Y2H network contained nine PPIs (z-score ϭ 5.4). Given the high (eightfold) symmetry of the NPC, homomeric interactions may be expected. While homomeric interactions systemically remain undetected in the published co-AP studies using haploid yeast strains (38), the Y2H approach has no such limitations. However, we still only found four homomeric interactions in total, scNup145N-Nup145N (C-term, aa 457-605), scNup120 -Nup120 (N-term, aa 1-757), hsSec13-Sec13, and hsNup54 -Nup54. The latter coiled-coiled interaction may well be detected only in the absence of hsNup58 and hsNup62, which together form a stoichiometric heterotrimeric complex (15)(16)(17).
The cytoplasmic and nucleoplasmic ring of the NPC are each considered to be primarily built by Y-complexes (1,10,18,19), and they may serve as anchor points for complex assembly. Intersubcomplex interaction is critical to determine how the NPC assembles. We found connections from the scNup120 arm of the Y-complex to the Nic96/hsNup93 complex member scNup192 ( Fig. 2A) and from scNup85 to sc-Nup82, a component of the cytoplasmic filament network. The C terminus of scNup82 (aa452-713) was involved in an interaction with scNsp1 ( Fig. 2A) (44), while the N-terminal ␤-propeller domain interacted with six other Nups, including the known interaction partners scNup159 and scNup116 ( Fig. 2A) (35,45,46). The scNsp1 complex (Nsp1/Nup49/ Nup57) ( Fig. 2A) (47) connected to scNic96 (42) and sc-Nup59. The three structurally similar phenylalanine-glycine -Nups (scNup100, scNup116, scNup145N) had interactions with members of the Y-, the Nic96, and the Nsp1 complex, as well as the cytoplasmic filament network. We did not find interactions between Y-complex members and the Nsp1 complex in yeast.
In an eightfold symmetrical NPC, every subcomplex is likely represented in multiples of eight (29,48). Thus, in principle, some of the intracomplex interactions may contribute as intercomplex interaction in the assembled pore, but it is difficult to make this distinction from the Y2H interaction data alone. With the elucidation of the Nup120 -Nup145C-Sec13-Nup85 Y-complex hub structure, it was possible to assemble a composite 3D atomic model of this subcomplex (9). Because our Y2H data perfectly reflect all interactions in the subcomplex at domain resolution (Fig. 3A), we can formally separate intra-Y-complex interactions from interactions that may occur in the pore between neighboring Y-complexes. Thus, in addition to the seven intra-Y-complex interactions, score (1, 3) reflecting the success rate of how often a unique protein pair was found to interact in different replicas using different clones and different Y2H configurations. Str, X indicates that a crystal structure of the interacting pair has been determined (standard reference interaction set). Aml, X indicates an interaction reported by Amlacher et al. (37). Led, X indicates an interaction reported by Leducq et al. (36). CtcFreq, contact frequencies (medium shaded blue, high shaded green) reported by Alber et al. (38). co-AP, co-AP score (0. 00,1. 00] (25) derived from coaffinity purification (co-AP) experiments by Alber et al. (38). our Y2H analysis revealed two putative inter-Y-complex interactions in yeast: scNup84 -Nup120 and scNup120 -Nup120. Additionally, the cross human-yeast hsNup133-scNup85 C-terminal tail (aa 533-744) pair showed up most strongly in the Y2H analysis. The human data further revealed ␤-propeller-␤-propeller interactions of hsSec13  (9). All subunit interactions were recapitulated in the Y2H screen. Using several domain based constructs, minimal binding fragments were inferred from the Y2H data for yeast proteins and agree with the protein interfaces in the crystal structures. (B) Distribution of the fraction of interactions found with low (gray), medium (blue) and high (green) contact frequencies in Alber et al. (38). Fraction of interactions which were crystallized (standard reference interaction set) are indicated with blue circles. (C) Distribution of the number of randomized networks over their average co-AP score from 1000 randomizations. The value for the real network is 0.19 (blue line). with hsSec13, hsSeh1, and hsNup37 that also likely contribute inter-Y-complex interactions.
Biochemical Validation of Yeast Inter-Y-Complex PPIs-Many yeast Nups can be purified recombinantly from Escherichia coli. In particular, interacting members of the Y-complex can be copurified and used for in vitro binding measurements to validate novel inter-Y-complex interactions. We purified scNup120(1-757), scNup85/Seh1, scNup133(521-1157)/Nup84, and the scNup145C(34 -712)/Nup120/Sec13 complex (31) via affinity chromatography followed by size exclusion chromatography to verify the three putative inter-Y-complex PPIs revealed in the Y2H analysis. We turned to microscale thermophoresis (MST), an approach that can monitor protein complex formation, as a binding partner typically alters the directed movement of a fluorescent protein through a microscopic temperature gradient in solution (49,50). One protein was fluorescently labeled, kept at constant nM concentration, and the potential interaction partner was varied in concentration to measure apparent equilibrium binding constants (K D ). This approach does not require immobilization of one protein partner and thus eased problems arising from matrix-related background binding.
We could recapitulate the formation of the Y-complex hub with this setup. scNup145C(34 -712)/Nup120/Sec13 and sc-Nup85/Seh1 formed a complex with an apparent dissociation constant of K D ϳ 0.15 M (Fig. 5A). We next tested scNup120 dimerization that was not detected during the purification procedure e.g. in the size exclusion chromatography step. However, scNup120 (aa 1-757) homodimer formation can be observed in the MST experiments and resulted in a dissociation constant in the lower micromolar range (Fig. 5B). Similarly, binding of labeled scNup120 (aa 1-757) with the sc-Nup133(521-1157)/Nup84 complex was observed (Fig. 5C), matching the scNup120 -Nup84 inter-Y-complex interaction proposed from the Y2H data. Finally, binding behavior was monitored when assaying labeled scNup85/Seh1 with increasing concentration of the scNup133(521-1157)/Nup84 complex. The concentration of the scNup133(521-1157)/ Nup84 complex was limiting so that we did not reach full binding saturation in these experiments. However, a ⌬ Fnorm value range of ϳ20 AU (normalized fluorescence change ϭ F hot /F cold ) indicated binding (Fig. 5D). This experiment corroborated the scNup85-hsNup133 interolog. Dissociation constants in the low micrometer range are consistent with the idea that these interactions contribute to building the nuclear pore assembly from subcomplexes. Multiple contacts of proteins or subcomplexes within the NPC likely have an additive effect compared with the affinities of isolated components FIG. 4. Network representation of the yeast and the human Y2H network, respectively. The yeast PPI network is presented on the left and the human Y2H data on the right. Nodes indicate proteins, colored and grouped according to substructures. Edges indicate Y2H protein interactions, red: overlap with the standard reference set from x-ray structures, green: interactions that were also found cross species between human and yeast proteins. measured in vitro. They may provide specificity while remaining more dynamic and susceptible to competition than more tight monovalent interactions (51).
Confirmation of Human ␤-Propeller and Inter-Y-Complex PPIs in Intact Human Cells-Six conserved proteins, scNup120/hsNup160, Nup133, scNup84/hsNup107, Nup85, scNup145C/hsNup96, and Sec13 together form two short arms and one long stalk, connected at a central hub, to build the Y-shaped complex (9). In addition to the conserved core proteins, the Y-complex can contain additional proteins, namely Seh1, Nup37, Nup43, or ELYS, depending on the species (52). The function of these accessory proteins is not yet fully understood (53). In the Y2H analysis of the human proteins, we found four interactions with hsSec13, the conserved ␤-propeller with a central position in the hub of the Y-complex. hsSec13 interacted with hsNup62, and with three other ␤-propeller proteins: hsSeh1, hsNup37, and hsSec13 itself.
In order to test these interactions in intact U2OS cells, we employed a YFP-based protein complementation assay (PCA) (54 -56). In this assay, a YFP-fluorescence signal is reconstituted from two YFP fragments that fold through interaction of the proteins fused to the F1-YFP (aa 1-158 of Venus YFP) and F2-YFP (aa 159 -239 of Venus YFP) fragments, respectively. We demonstrated the intra-Y-complex interaction between hsNup133 and hsNup107 (7,40,41) in this system (Fig. 6A). Characteristic for proteins in the NPC, we observed YFP fluorescence at the nuclear rim in confocal images 36 h after transfection, recapitulating that hsNup133 and hsNup107 interact within the NPC. Complementation of hsNup133 or hsNup107 with either hsSec13, hsSeh1, or hsNup62 did not result any YFP signal, confirming the specificity of the test (Supplemental Fig. S3). HsSec13 interaction with hsNup62, hsSeh1, and hsSec13, as observed in the Y2H screen, were confirmed through protein fragment complementation localized mainly to the nuclear rim in intact human U2OS cells (Figs. 6B-6D). Besides the general assumption that the NPC is structurally similar between yeast and human, testing the hsNup133-scNup85 interaction provided further evidence that this interaction should be relevant in the human NPC. Assaying human Nup85 and Nup133 in our PCA approach, we observed a fluorescence signal at the nuclear rim (Fig. 6E). In addition to the biochemical interaction measurements with the yeast homologs (Fig. 5D), validation of this interaction in intact human cells seemed of particular interest providing a substantial constraint toward the arrangement of the Y-complexes with respect to each other within the nuclear pore complex.

DISCUSSION
The primary motivation for this study was to elucidate protein-protein interactions within the NPC that would provide a new basis for constructing composite assembly structures at high resolution. This is important because of the existing resolution gap between cryo-ET structures of the assembled NPC of different species. The supramolecular structure of the NPC of several species has been determined by cryo-ET at resolutions between ϳ7.6 and ϳ2 nm (3,18,19,39,(57)(58)(59)(60) and the growing, but still incomplete, list of available nucleoporin crystal structures does not yet allow for an unambiguous placement of all the components into a fully assembled NPC structure (20,21).
Certain subcomplexes that organize the 30 nucleoporins are well understood. Because the components bind with high affinity, they can be more easily copurified and, for example, cocrystallized. The 7-10 membered Y-complex is the best example for a stable subcomplex (1, 8 -10, 19, 61-63). Other interactions have been characterized as well but especially intersubcomplex interactions have been notoriously difficult to establish. Apparently, such contacts are either weak or are systematically missed by the respective assays. Here, we exploited structural knowledge of nucleoporins to design protein constructs for a comprehensive Y2H-based PPI analysis for the NPC. Specifically, unstructured domains were removed and constructs reflected structured, potential binding domains, such as ␤-propeller, stacked helical, coil-coil, and ␣/␤ domain arrangements. We assessed in a stringent pairwise Y2H matrix approach the NPC interactome for yeast and human and also examined cross-species interactions. Our data set describes 39 interactions between 25 yeast Nups and 15 interactions connecting human Nups. The substantially higher number of interactions we found between yeast Nups, using structured domains, relative to the human PPI, using full-length proteins, is explained by large proteins having a propensity to generate false negative results in Y2H, presumably for steric reasons. Benchmarking against other literature-derived data demonstrated the high quality of our data set. For example, pairs of proteins frequently copurifying in independent AP-MS analyses (1,38) are highly enriched in our data (Fig. 3C). This suggests that overall, highly reliable interaction information was obtained in the approach providing useful leads for further investigation into how the NPC is assembled.
assembly have been put forward over the years (1,19,64). The current benchmark, since based on a large body of experimental data and a ϳ3.2 nm cryo-ET map, is an assembly model first proposed by the Beck group (19). Because of the large size and distinct shape, the Y-complex is the prime substructure to be placed in the cryo-ET density map. Bui et al. posit that the Y-complex forms two eight-membered rings on the cytoplasmic as well as the nucleoplasmic side of the NPC, for a total of 32 copies per pore assembly. They suggest two reticulate rings in a head-to-tail orientation, where the Nup133 N-terminal ␤-propeller is the tail, while the hub area of the Y-complex defines the head. A central ring sandwiched between these two rings and assumed to primarily contain the hsNup93 subcomplex occupying the midplane of the NPC (20,21).
One interesting question we asked is whether our PPI data are compatible with the two tandem ring model. For that, we focused on the inter-Y-complex interactions, which are Nup120 (aa 1-757)-Nup120 (aa 1-757), Nup120 (aa 1-757)-Nup84, Nup120 (aa 822-1037)-Nup84, Nup85 (aa 533-744)-hsNup133, hsSec13-hsSec13, hsSec13-hsSeh1, and, finally, hsSec13-hsNup37 (Fig. 7). Assuming all observed inter-Y interactions occur simultaneously in the assembled NPC, we have to conclude that several different Y-complex interfaces must exist, as all inter-Y contacts cannot coexist simultaneously in a single Y-complex pair conformation. For example, the Nup120 -Nup120 and the Nup85-Nup133 are sterically prohibited from coexisting between two interacting Y-complexes, even when considering the considerable flexibility of the structure (9,63,65). This observation is in line with the tandem two-ring model that postulates at least four distinct microenvironments for Y-Y-complex contacts, i.e. each microenvironment has its specific set of interactions. That said, the interactions we see are only partially compatible with the two-ring model on a molecular level, assuming that the ring structure is generally conserved between human and yeast. While the Nup133-Nup85 contact is consistent with a possible inner ring Y-outer ring Y interaction, the Nup120 -Nup120 interaction is in conflict with the tandem two-ring model because of an offset of ϳ12 nm of the inner ring/outer ring Y-complex pair. Equally, the hsSec13-hsSeh1 may contribute to conformational flexibility of the Y-complex hub (65), and the hsNup37-hsSec13 interaction may be involved in contacts by the neighboring inner-and outer-Y-complex, but coexistence with the hsSec13-hsSec13 contact is unlikely. However, to simply compare and contrast our putative inter-Y interactions with the assembly model for the NPC put forward by the Beck group would be overly simplistic. For instance, FIG. 7. Interaction overview of the yeast Y2H interaction data emphasizing the modular structure of the NPC, showing intrasubcomplex and intersubcomplex interactions. Edges indicate Y2H protein interactions, red: overlap with the standard reference set from x-ray structures, green: interactions that were also found cross species between human and yeast proteins.
hsSec13-hsSec13 and hsSec13-hsSeh1 may also be reflective of interactions within the GATOR2 (GAP towards Rags) complex where these proteins have an additional role (66). Also, major differences between the cytoplasmic and the nucleoplasmic rings cannot be ruled out at our current structural knowledge, as specifically discussed in the report of the latest cryo-ET structure of the Xenopus NPC (60).
Interactions between scaffold nucleoporins extending beyond the Y-complex have also been found. Many of them confirm previously established interactions (Fig. 7). Nup53 and Nup59 interacting both with Nic96 and Nup192 is consistent with a previously targeted Y2H analysis (37). The interaction between Nup82 and Nup159 as well as Nup82 and Nup100 or Nup116 is expected based on existing structural data (35). For the trimeric Nsp1 complex, we found direct contacts between Nup57 and Nsp1 and Nup57 and Nup49, as well as Nup57 and Nic96. These are consistent with the literature (15,44,67).
A number of additional contacts between NPC subcomplexes have been established (Fig. 7). Since the NPC is likely to be flexible on multiple levels (68), analysis of the interactome data is not trivial. As in the case of the Y-complex, flexibility is likely to create different microenvironments for other subcomplexes as well, such that detected interactions might only exist for a subset of Nups at any given time. Another point is that contacts that match established intrasubcomplex contacts could "hide" secondary, intersubcomplex interactions or be mutually exclusive. Further, the NPC needs to assemble into the confined space of the circular openings in the nuclear envelope. It is likely, although not tested, that the assembly process itself requires substantial conformational flexibility of subcomplexes. Therefore, an interaction detected in our screen may reflect a temporary contact with high relevance in the NPC assembly process, yet it may not occur in the completed NPC. Finally, it is becoming clear that the modular NPC can exist in different compositions, possibly to fulfill specialized functions (69,70). Since our screen cannot differentiate between all these scenarios, the data may easily be overinterpreted. Despite these limitations, it is reasonable to assume that the interactome map is fairly comprehensive as it is a nearly complete coverage of previously known and structurally characterized interactions and thus provides a prime resource for pairwise interactions between Nups of the NPC. Y2H Analysis-PPI matrix screening was performed as described previously (25,28). In brief, nonautoactivating baits (L40ccU MATa yeast strains) were mated six times in 384-array format with prey strains using two and three independently transformed bait and prey yeast colonies, respectively. Interacting bait-prey pairs were identified by growth on selective agar plates (Leu-Trp-Ura-His). Only baitprey pairs that showed growth at least two times were considered for further evaluation. As multiple clones and configurations of the same protein pair were tested in several replicas, we combined all data obtained for each interacting pair assigning a score (1,3). The PPI data are reported in Table S1 and have been submitted to the IMEx (http://www.imexconsortium.org) consortium through IntAct (72) and assigned the identifier IM-25241.
Microscale Thermophoresis-Purified proteins were labeled following the manual of Nanotemper technology, i.e. Monolith NT TM Protein Labeling Kit BLUE-NHS 647. The concentration of labeled proteins was adjusted to ϳ50 -100 nM. Dilution series of up to 16 unlabeled protein concentrations were prepared in a final volume of 20 l for the MST measurements. Capillaries were filled and loaded into the Monolith NT.115, and the thermophoresis was performed using the Monolith NT.115 Nanotemper TM Technology.
Protein Complementation Assays in Mammalian Cells-U2OS cells were seeded in 24-well plates on cover slips and cotransfected with pairs of 50 ng Venus-F1 and 50 ng Venus-F2 plasmid DNA. 12-24 h after transection cell were grown further in fresh DMEM/FBS (5%) for additional 24 h. The cells were fixed at room temperature for 10 min in the dark with 2% paraformaldehyde in DPBS. Cells were stained with DAPI (16 g/ml) for 2 min at room temperature. To visualize the YFP signal indicative of a protein interaction, a confocal fluorescence microscope (LSM700, Zeiss) at excitation wave length 525 nm was used.
Computational Analyses-Network analyses and visualization were performed with Cytoscape, R and custom PERL scripts for network link randomization procedures. Co-AP scores (25) were calculated using composite interaction profiles from Supplemental Fig. 4 of Alber et al. (38). R01GM77537; K. E. K. and N. C. L., T32GM007287). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. □ S This article contains supplemental material.
The authors declare that no competing interests exist.