Architectural Nucleoporins Nup157/170 and Nup133 Are Structurally Related and Descend from a Second Ancestral Element*

The nuclear pore complex (NPC) constitutes one of the largest protein assemblies in the eukaryotic cell and forms the exclusive gateway to the nucleus. The stable, ∼15–20-MDa scaffold ring of the NPC is built from two multiprotein complexes arranged around a central 8-fold axis. Here we present crystal structures of two large architectural units, yNup170979–1502 and hNup107658–925·hNup133517–1156, each a constituent of one of the two multiprotein complexes. Conservation of domain arrangement and of tertiary structure suggests that Nup157/170 and Nup133 derived from a common ancestor. Together with the previously established ancestral coatomer element (ACE1), these two elements constitute the major α-helical building blocks of the NPC scaffold and define its branched, lattice-like architecture, similar to vesicle coats like COPII. We hypothesize that the extant NPC evolved early during eukaryotic evolution from a rudimentary structure composed of several identical copies of a few ancestral elements, later diversified and specified by gene duplication.

The membrane-enveloped nucleus is the hallmark of the eukaryotic cell. Physical separation of nucleoplasm and cytoplasm necessitates sites for molecular exchange (1)(2)(3). Nuclear pore complexes (NPCs), 2 plugged into circular openings where inner and outer nuclear membranes fuse, perforate the nuclear envelope and form the sole gateway. The NPC is, at ϳ50 MDa, one of the largest protein assemblies in the quiescent cell. It is modular, comprises ϳ30 different proteins, termed nucleoporins (nups), and forms an 8-fold symmetric ring embedded in the nuclear envelope (4). In accord with the symmetry of the complex, each nucleoporin is present in 8 ϫ n copies/NPC. The architecture of the NPC is roughly conserved among eukaryotes, measuring ϳ100 nm in the outer diameter, with a central transport gate ϳ40 nm wide (5)(6)(7)(8). The NPC is a highly dynamic assembly. Some nucleoporins are stably attached, whereas others are more dynamic (9 -11). The main scaffold ring is composed of ϳ15 architectural nucleoporins that anchor to the inner pore wall. A second set of nucleoporins (FG-nups) is characterized by long, phenylalanine-glycine (FG) rich filamentous extensions. These FG fibers emanate into the central cavity of the NPC and define the main transport barrier (12)(13)(14). Ions, metabolites, and macromolecules less than 20 -40 kDa diffuse, for the most part, freely through the transport gate. Larger molecules pass only when bound to dedicated nuclear transport receptors, termed karyopherins, which directly interact with FG-nups (15)(16)(17). The small GTPase Ran regulates the interaction of protein cargo with import or export karyopherins, conferring directionality to these transport processes. This regulation depends on Ran being GTP-bound in the nucleus and GDP-bound in the cytoplasm, a gradient established by the action of cytoplasmic GTPase-activating protein (RanGAP) and nuclear GTP exchange factor (RanGEF).
To better understand the myriad of functions attributed to the NPC, which go far beyond transporting molecules across the nuclear envelope (18,19), we are interested in the structural characterization of the NPC, which begins with the stable scaffold structure. The ϳ15 architectural nucleoporins are organized in two large multiprotein complexes: the well studied Nup84 complex and the more enigmatic Nic96 complex. The components of each are known ( Table  1). The Nup84 complex contains seven universally conserved nucleoporins and adopts a characteristically branched Y shape (20 -22). In this work, the Nup84 complex is referred to as the Y complex. Nup120 and Nup85⅐Seh1 form the two short arms, whereas Nup145C⅐Sec13, Nup84, and Nup133 build the long, kinked stalk. The Nic96 complex likely contains five distinct nucleoporins, two of them duplicated in yeast (23)(24)(25)(26)(27). It connects to the nuclear envelope (28) as well as the FG network (29). These two scaffold complexes likely form ring-like assemblies. Whether these rings are stacked or concentric or arranged some other way is controversial (30 -33). This structural framework is important to the assembly and function of the NPC. Severe defects occur when scaffold nucleoporins are deleted or depleted, including failure to recruit other nucleoporins and diminished transport of protein or RNA across the nuclear membrane (34 -41).
Superficially, the architectural nucleoporins are classified by computational methods as ␤-propeller domains, ␣-helical repeat domains, or tandem combinations thereof (Table 1) (4,42,43). Experimental structural characterization, however, has revealed that this simplistic description does not adequately reflect the reality. For example, Sec13 and Seh1 are predicted as six-bladed ␤-propellers but turn out to be seven-bladed, with the final blade provided in trans by their respective binding partners (31,32,44,45). The four ACE1 nucleoporins are built around a ϳ65-kDa ␣-helical domain. They are distantly related to one another and, strikingly, also to Sec31, the main structural component of the outer coat of the COPII vesicle. This ACE1 domain is a tripartite fold back structure of ϳ28 ␣-helices, distinct from the regular ␣-solenoid domains found in HEAT, TPR, or PPR repeat proteins (46,47), among others. The structural similarity between these ACE1 proteins provided the proof that the NPC and COPII coat derive from a common ancestor (32), as hypothesized previously (42,43).
The ACE1 nucleoporin Nup84 binds Nup133. The structure of a fragment of the human Nup84 ortholog, hNup107 658 -925 , has been solved in complex with hNup133 934 -1156 (48), the C terminus of the protein. This structure showed that the C terminus of Nup133 consists of two ␣-helical blocks. A rigid block of four ␣-helices, residues 934 -1008, forms an interface bundle that binds Nup84(hNup107). A moderately flexible hinge connects this interface bundle to a second ␣-helical unit that forms a distinct lobe at the C terminus of the protein. The N terminus of Nup133 is a ␤-propeller, whose structure is also known (49). hNup133 934 -1156 suggested that Nup133 is not an ACE1 protein.
Here we present crystallographic analysis of two architectural units, yNup170 979 -1502 and hNup107 658 -925 ⅐hNup133 517-1156 , components of both major scaffold complexes of the NPC. Nup170, its paralog Nup157, and Nup133 each consist of an N-terminal ␤-propeller followed by an ϳ80-kDa C-terminal ␣-helical domain. The structures reveal a common ␣-helical architecture for Nup157/170 and Nup133 that is distinct from all other known nucleoporins. This ␣-helical architecture is, with ACE1, another ancestral element of the NPC. We conclude that the basic NPC framework is built from a small set of recognizable structural elements that were already present in multiple copies in the last common ancestor of extant eukaryotes. During the course of evolution, gene duplications occurred and diversified these core elements, generating the complex, multi-functional machine that is the NPC.
Data Collection, Structure Solution, and Refinement-The data for yNup170 979 -1502 were collected at 100 K at microfocus Beamline 24-IDE at the Advanced Photon Source (Argonne, IL). The crystals that diffracted best, to ϳ2.5 Å, were perfectly merohedrally twinned. Untwinned data were obtained to 3.2 Å and used for further analysis. The data were collected from selenomethionine-labeled crystals and processed with the HKL2000 package (50). The phases were determined by selenium single-wavelength anomalous dispersion. Nine of ten possible selenium sites were identified with SHELXD (51). Selenium positions were refined with SHARP (52), which also revealed the additional selenium site, and experimental phases were calculated. The resulting solvent-flattened electron density map was used to build a model with Coot (53).
To improve model quality and attempt to refine against the twinned data diffracting to higher resolution, the C-terminal subdomain yNup170 1253-1502 was crystallized, and 2.2 Å data was collected at Beamline 24-IDC at the Advanced Photon Source from a large crystal formed from several caked layers. The strongest of the observed diffraction patterns was indexed and integrated. Of many specimens tested, morphologically indistinguishable, only this crystal diffracted strongly and belonged to space group I222 ( Table 2). All others belonged to space group P6 3 22, diffracted to 3.5 Å, and were not further analyzed. A molecular replacement solution for yNup170 1253-1502 was found using a partial model from the initially obtained 3.2 Å structure of yNup170 979 -1502 . A complete model for yNup170  was built automatically, with minor intervention, using PHENIX (54). yNup170 979 -1502 was rebuilt incorporating this partial model, improving refinement parameters. We note, however, that the model of the yNup170 C-terminal subdomain did not help process the twinned 2.5 Å data (not shown).
Data for hNup107 658 -925 ⅐hNup133 517-1156 were collected at 100 K at 24-IDE. Because of significant radiation damage, partial data sets were collected and merged from several crystals grown in the same crystallization drop, each exposed at 3-10 spots. The structure was phased by molecular replacement using the minimal 55-kDa hNup107 658 -925 ⅐hNup133 934 -1156 interaction complex (48) as a search model in Phaser (55) in the CCP4 suite (56). The additional 45-kDa domain was built and refined with Coot (53) and PHENIX (54). The data to 3.5 Å were included, despite low I/I, as recommended for low resolution crystallography (57). Anisotropic diffraction was corrected by elliptical resolution truncation and anisotropic B-factor correction using the Diffraction Anisotropy Server (58). The obtained electron density maps allowed positioning of the secondary structure elements, which are essentially all ␣-helical. Connections between helices were mostly visible, allowing tracing of the molecule from the N to the C termini. Observed chain topology and variation in the length of helices allowed us to assign each modeled helix unambiguously to the secondary structure as predicted by the PredictProtein server (59). In the absence of detailed positional markers, the assigned sequence in the deposited data is approximate but is likely erroneous only in a few places and shifted by not more than three or four residues, i.e. one ␣-helical turn. Several nonhelical loops could be traced confidently, including loops that are disordered in the partial structure hNup107 658 -925 ⅐hNup133 934 -1156 (48).
Structure Analysis-Nup170 homologs were retrieved from the NCBI website data base, and a multiple sequence alignment was calculated by the MUSCLE algorithm (60). An average distance tree was used to select representative, divergent sequences. The residues were scored for conservation by the AMAS method in JALVIEW (61). PBD2PQR (62) and APBS (63) software packages were used to calculate surface charge, and the PISA server was used to calculate the accessible surface area (64). MODELLER (65) was used to build a complete model of yNup157 900 -1391 . Pymol was used to generate figures.

RESULTS
Structure of the ␣-Helical Domain of Nup170-Nup170 is predicted to contain two structural domains: an N-terminal ␤-propeller (residues 180 -650) and a C-terminal ␣-helical domain (Fig. 1A) (4,42). By expressing a series of N-terminal truncations of the protein and by limited proteolysis, we defined a stable core of the predicted ␣-helical region, comprising residues 979 -1502 of Nup170 from S. cerevisiae (data not shown). The presumed N-terminal ␤-propeller domain interacts weakly with the ␣-helical domain when separated, indicating flexible attachment (66). yNup170 979 -1502 was expressed recombinantly in E. coli, purified to homogeneity, and crystallized. The structure was solved by single-wavelength anomalous dispersion, using selenomethionine-labeled protein. The asymmetric unit contains one molecule.
The experimental single-wavelength anomalous dispersion electron density allowed for building residues 1020 -1460, revealing a continuous but bipartite stacked ␣-helical domain (Fig. 1). Because of the lack of strong crystal contacts, the C-terminal half of the domain is flexibly positioned. Thus, to aid structure determination, this C-terminal 29-kDa subdomain (residues 1253-1502) was separately expressed and crystallized. The data to 2.2-Å resolution were collected and phased by molecular replacement with the relevant portion of the larger protein as initially modeled. The complete, refined model of the C-terminal subdomain (R free / R work ϭ 27.6/23.3%) was then used to build the structure  Fig. S1. The crystal packing of the 2.2-Å resolution structure is shown in supplemental Movie S1. The data collection and refinement statistics are summarized in Table 2. yNup170 979 -1502 adopts an irregular ␣-helical stack composed of 26 ␣-helices and overall dimensions of 12 ϫ 4 ϫ 4 nm (Fig. 1). We label these helices ␣1-26. The domain begins with helices ␣1/2, ␣3/4, and ␣6/7 forming three consecutive pairs of helices of various lengths, stacked antiparallel, without superhelical twist. Helix ␣5 resides in a loop and does not pair to other helices. Helices ␣8 -13 form an extended zigzag pattern that is rotated by ϳ90°against the ␣1-7 stack. This zigzag can be likened to a stack of three ␣-helical pairs that has been stretched by pulling on its ends. As a result, helices ␣8 -13 extend over ϳ38 Å, reflecting a ϳ50% stretch compared with a tightly packed six-helix stack, which would span only ϳ26 Å. The hydrophobic core of this extended zigzag is poorly packed. Few residues are fully buried. Helix ␣14 is approximately twice as long as its direct neighbors and connects the two ␣-helical subdomains. The C-terminal subdomain forms a crescent only loosely definable as a stack. It starts with ␣15, unexpectedly positioned below, not above, helix ␣14. This helix abuts end-on-end to ␣12 of the N-terminal subdomain. The strictly conserved Arg 1232 is sandwiched between the negatively polarized C termini of ␣12 and ␣15, presumably for charge compensation. Nup170 then continues with helices ␣16 -26 forming a compact hydrophobic core, implying rigidity.
To compare the structure of Nup170 with those of other proteins, we performed structurebased searches with VAST and DALI (67,68). Neither returns significant alignments. No protein aligns to Nup170 over more than six consecutive helices. We conclude that Nup170 has only remote structural similarity to known proteins.
Nup133 517-1156 Adopts a Quadripartite Domain-Nup133 and Nup170 are predicted to have a similar overall topology. Both comprise an N-terminal ␤-propeller domain linked to a C-terminal ␣-helical domain (Figs. 1A and 2A). To compare directly the ␣-helical domains of Nup133 and Nup170, we solved the structure of the complete ␣-helical domain of hNup133 in complex with hNup107. hNup107 658 -925 and hNup133 517-1156 were co-expressed recombinantly in E. coli, purified, and crystallized. The structure was solved by molecular replacement using the 57-kDa Nup107 658 -925 ⅐Nup133 934 -1156 interaction complex (Protein Data Bank code 3CQC) (48). The asymmetric unit contains one heterodimer. Crystallographic analysis was challenging, because crystals were small and difficult to grow, suffered severe radiation damage, and diffracted anisotropically. Based on the resulting electron density map, we were able to assign all the helices of Nup133 and unambiguously determine the overall topology. Most connecting loops are also visible in the electron density. The 105-kDa complex was refined to R free /R work ϭ 37.0/31.2% ( Table 2). The model and electron density for the novel portion is shown in supplemental Fig. S2. hNup133 517-1156 forms an elongated structure composed of 28 helices (Fig. 2). Because the N-terminal ␤-propeller domain of hNup133 has three helices inserted into it (Protein Data Bank code 1XKS) (49), we number hNup133 517-1156 beginning at helix ␣4. The domain can be described as quadripartite. From the N terminus, it starts with a block of 12 long helices, which form a wide and flat plane. These helices are arranged pairwise and antiparallel, except helices ␣9 -10, which form an overhand turn. The following six helices (residues 854 -944) are all short (two or three turns) and zigzag upward, covering a distance of 44 Å. Helices ␣21-24 make up the interface with Nup107 and form a ␣-helical bundle as described previously (48). Finally, helices ␣25-31 fold into a compact C-terminal subdomain.
Comparison with Minimal Interacting Complex hNup107 658 -925 ⅐ hNup133 934 -1156 -The Nup107 moiety in the complex solved here is identical to the minimal interaction fragment previously reported. The portion of Nup133 solved here includes the entire helical portion solved previously and the 418 amino acids that connect it to the N-terminal domain. The C-terminal subdomain of Nup133 forms a crystal contact in the current structure. This subdomain is therefore more stable than in the minimal interacting complex. Several loops not modeled previously are apparent. The metazoan-specific finger helix of Nup107 (labeled ␣6Ј in Fig. 2B) also forms a crystal contact, causing it to bend more than in the previous structure. Otherwise no noteworthy rearrangements occur.
Structural Comparison of Nup170 and Nup133-Sequence-and structure-based alignments suggest remote homology between Nup170 and Nup133. Because of the multi-partite nature of the two proteins and the apparent hinges connecting the subdomains, an overall superimposition is not very informative. However, if the separate subdomains are superposed individually, commonalities become apparent (Fig. 3). The most striking similarity exists between the middle segments of Nup170 and Nup133, where the helices are short and form a characteristically extended zigzag structure. An elastic network model suggests that both molecules may flex about this central zigzag (supplemental Movies S2 and S3). Further, the Nup133 interface bundle and the connection to the C-terminal lobe are particularly conserved. The C-terminal lobe of Nup170 is larger than that of Nup133 (34 versus 27 kDa). N-terminally attached to the zigzag are helices that form tightly packed bundles; however, in Nup133 they form a flat and extended plane, whereas in Nup170 we observe three tightly stacked ␣-helical pairs. This remote but distinctive structural homology is matched by homology observed in the amino acid sequence. A PSI-BLAST search with hNup133 517-1156 predominantly returns homologs of Nup133 and Nup170. Detected similarity spans the entire region that is structurally similar (Nup133 residues 830 -1120 and Nup170 residues 1150 -1380). This search result suggests that Nup170 and Nup133 are more closely related to each other than to other proteins.
yNup170 979 -1502 Has Two Conserved Surface Features-Nup170 is integrated into the structural framework of the NPC and must interact with other nucleoporins to exert its function; however, the direct interaction partners are not yet firmly established. Interaction sites can often be identified by conservation of neighboring surface-exposed residues. We generated a maximally diverse alignment of Nup170 sequences across all eukaryotes (supplemental Fig. S3). Surface representation of Nup170 colored by conservation suggests two conserved surfaces (Fig. 4). Helices ␣11-14 contribute to the first surface, which is mildly conserved and negatively charged. According to the structural alignment, Nup170 helices ␣11-14 correspond to the Nup133 interface bundle, the group of helices by which Nup133 binds Nup107. This alignment is shown in detail in supplemental Fig. S4. The conserved surface on Nup170 corresponds to the Nup133⅐Nup107 interface. In Nup133, this surface is hydrophobic. The double mutant L973E/L976E prevents interaction with Nup107 by placing charged side chains in the  OCTOBER 9, 2009 • VOLUME 284 • NUMBER 41

JOURNAL OF BIOLOGICAL CHEMISTRY 28447
interface (48). The corresponding residues in Nup170, Glu 1234 and Arg 1238 , are charged. If Nup170 binds another protein by this interface, the interaction is not hydrophobic and likely weaker.
The second surface, built by helices ␣16 -18, is more strongly conserved and forms a hydrophobic groove. In the sequence alignment (Fig. 4F), the exposed neighboring residues are conserved: Ser 1305 as serine or threonine, Phe 1308 as phenylalanine or tyrosine, and Phe 1325 as a large hydrophobic residue. Therefore, this region has the characteristics of a typical proteinprotein interface. Interestingly, in the 2.2-Å crystal structure of the short yNup170 1253-1502 , the N-terminal helix, ␣14, is unwound, and the first 10 residues wrap around the surface of the protein and align in the hydrophobic groove (Fig. 4C). One may speculate whether this rearrangement and interaction has physiological relevance. With the full ␣-helical domain present, this motion of helix ␣14 would reorient the N-terminal portion dramatically, causing it to extend opposite the direction observed in this structure. Alternatively, we can consider the intramolecular interaction of these 10 residues as a serendipitous crystal artifact that may mimic the interaction with the natural binding partner. In the later case, this structure would suggest the physical mechanism by which such interaction occurs.
Comparison of Nup170 to Nup157-In S. cerevisiae, Nup170 has a paralog, Nup157, not present in metazoa (26), likely a consequence of the whole genome duplication that occurred in the ancestor of Saccharomyces (69). We modeled the homologous ␣-helical domain of Nup157, which can be done confidently (sequence identity is 41% for the 60-kDa segment in question). We note that the exposed residues at the second conserved surface of Nup170 are identical in Nup157, except residue 1328 is serine instead of alanine. If this hydrophobic groove is a protein interface, we predict that Nup170 and Nup157 bind to the same molecule. The most apparent difference between the ␣-helical domains of Nup170 and Nup157 are three insertions in the very N-terminal portion of the ␣-helical domain of Nup170, not part of this structure. These Nup170 extensions are predicted as disordered loops. Thus, it is likely that the structures of the two proteins have a very similar topology across the entire ␣-helical region.

DISCUSSION
The size, complexity, and heterogeneity of the NPC make this complex a formidable challenge to structural biology. However, because the NPC is assembled from subcomplexes in a modular fashion, the high resolution structure can be approached by a divide-and-conquer strategy (4). Two multiprotein complexes compose the principal scaffold architecture: the Y complex and the Nic96 complex (30). We report here one structure from each: hNup107 658 -925 ⅐hNup133 517-1156 of the Y complex and yNup170 979 -1502 of the Nic96 complex. Structural similarities between Nup133 and Nup170 indicate that these nucleoporins descend from a common ancestor. Because ACE1 proteins are also found in both major structural complexes of the NPC, we conclude that both complexes employ similar structural principles. The extant NPC likely derived from duplication and diversification of a few ancestral genes.
The structure of Nup133 provides a significant portion for a now nearly complete high resolution model of the heptameric core of the Y complex, as cartooned in Fig. 5. This complex has been studied extensively by crystallography and electron microscopy (20,31,32,45,48,49,70). A central hub connects two short arms to a long and kinked stalk (20,21). The nucleoporins that build this Y connect via binary interactions with strong affinities (32,48). Crystal structures of most elements are now known. Seh1⅐Nup85 and Nup120 form the short arms (32,70). Sec13⅐Nup145C is the proximal segment of the kinked stalk (31); Nup84 is the chain link between Nup145C and the distal Nup133.
Electron micrographs of the Y complex suggest that some segments can articulate with respect to one another (20,71). Our hNup107 658 -925 ⅐hNup133 517-1156 reveals one molecular determinant of that flexibility. The extended zigzag at the center of the Nup133 ␣-helical domain provides an accordion-like transition that allows the flat, N-terminal ␣-helical plane of Nup133 to flex against Nup84(hNup107). This motion can be simulated using an elastic network model (72). In addition, there is a second hinge between the Nup84(hNup107) binding interface and the compact, C-terminal, all ␣-helical domain (49).
The Nic96 complex is less well studied than the Y complex. At high resolution, the ACE1 domain of Nic96 is known (73,74), as well as the 15-kDa homodimerization domain of Nup53(hNup35) (75). Both structures are uncomplexed, and the direct interaction partners within the Nic96 complex are not known with certainty. Nup157/170 is another member of this complex. Here we report the structure of a major potion of the ␣-helical stack domain of Nup170. Like Nup133, Nup170 . Surface conservation of Nup170 suggests two protein-protein interfaces. A, amino acid sequence conservation among Nup170 genes from maximally diverse eukaryotes was mapped on the protein, gradient-colored from white (not conserved) to orange (strongly conserved), orientated as in Fig. 1B. A conserved groove is boxed. B, structure rotated 180°compared with A, with conserved surface patch boxed. C, surface groove boxed in A shown magnified. Structure of C-terminal subdomain at 2.2-Å resolution is superposed and shown in aquamarine as a cartoon. Helix ␣14 as white cartoon, extending down and left, in the conformation seen in the full domain, as well as in aquamarine as the well ordered, extended peptide seen in the 2.2-Å structure of the isolated C-terminal subdomain. The key residues are labeled. D, surface patch boxed in B is magnified and shown as a cartoon with exposed residues labeled. Partially transparent surface representation is colored by calculated surface charge, in a gradient from negative (red) to neutral (white) to positive (blue). E, homologous section of Nup133 colored and labeled as in D with extent of interface to Nup107 delimited by a solid black line. F, sequence alignment of maximally diverse selection of eukaryotic Nup170 sequences, colored by conservation as in A-C. Helical segments are shown as red cylinders and labeled. The bar graph shows accessible surface area (Å 2 ) for each residue. Yellow circles mark conserved, buried arginine 1232; two hydrophilic surface residues, glutamine 1234 and arginine 1238, that would be buried were this surface, shown in D, indeed a protein-protein interface, as in Nup133, shown in E; and residues serine 1305, phenylalanine 1308, and phenylalanine 1325, lining the groove shown in C. OCTOBER 9, 2009 • VOLUME 284 • NUMBER 41 contains a flexibly tethered N-terminal ␤-propeller, followed by a large, ϳ70-kDa, ␣-helical domain. The ␣-helical domains of Nup133 and Nup170 both are divided into rigid segments connected via flexible hinges. Pairwise superposition of these segments accentuates the structural resemblance between the two proteins, particularly in the central zigzag and in the interface bundle (Fig. 4).

Structures of Nup133 and Nup157/170
The Nic96 complex is apparently not as stably associated as the Y complex. The many different interactions that have been reported for the Nic96 complex indicate that it has a role as a connector. The interaction with the FG-Nup-Nsp1 complex (29) on the one hand and Ndc1 on the other (76) indicate that the Nic96 complex spans the width of the NPC scaffold.
The crystal structures solved here solidify the notion that the major scaffold complexes of the NPC are structurally related. On that basis, we suggest that the Nic96 complex adopts a branched structure and that its components are joined through binary interactions. In Nup133, the interaction with Nup84 tethers the protein to the NPC (48). It is known that, as in Nup133, the ␣-helical domain of Nup170 is necessary and sufficient to target the protein to the NPC (66). We show here that the surface by which Nup133 binds Nup84 is also conserved in Nup157/170 but is not as hydrophobic, arguing for perhaps a more dynamic interaction.
Weaker interactions and protein-peptide interactions will play auxiliary roles in the assembly of both scaffold complexes and in their function. For example, the N-terminal 29 residues of the mRNA export factor Gle1 tether it via hNup155(yNup157/170) to the NPC (77). An interaction between Nup120 and Nup157/170 may join the two complexes (78).
A comparison of a variety of scaffold nucleoporins is now possible. One class shares the common ACE1 domain (32), also found in the COPII coat protein Sec31 (44). This class includes Nup85, Nup145C, Nic96, and Nup84. ACE1 consists of three ␣-helical modules, termed crown, trunk, and tail. These proteins fold back onto themselves to form a U-turn within the crown module. The trunk is composed of two ␣-helical units running in opposite directions, capped by the tail module. The tail of Nup84(hNup107) is shown here in complex with Nup133, and the tails of other ACE1 domains also support protein-protein interfaces (32,73). The ACE1 trunk domain, with multiple helices embedded in the hydrophobic core, confers greater rigidity than structures built of stacked ␣-helical pairs, which are more flexible. For example, all of the nuclear transport receptors are constructed from repeated ␣-helical pairs or triples, and the functional significance of flexibility is well documented (79). In Nup133 and Nup157/170, the helices do not fold back on one another as in the ACE1 trunk. Instead they typically pair, albeit without a recognizable repeat pattern. Consequently, they are more flexible than ACE1 proteins. Further, Nup133 and Nup157/170 show greater variation than the ACE1; although a central core has much the same accordionlike structure in both, flanking this core there is substantial variation.
The overall topologies of each class of scaffold nucleoporin are shown in Fig. 6. The ␣-helical domains of Nup133 and the ACE1 Nic96 are equally long (Fig. 6, A and B); however, the N terminus of the ACE1 is at the middle of the domain rather than at one end. Nup120 has a more convoluted architecture and is shorter (Fig. 6C). Its ␣-helical domain is interrupted by a blade of the N-terminal ␤-propeller, with which it integrates to form one compact entity. Taken together, these three domain classes are not really structurally related, other than all being ␣-helical. The assembly principles of vesicle and nuclear pore membrane coats are related, and some of their components evolved from common ancestors. The structural similarity between the ACE1 proteins of the NPC and of the COPII vesicle coat established the common ancestry (32), as hypothesized previously (42,43). Both coat assemblies also incorporate the bifunctional protein Sec13. We expect that the scaffold structure of the NPC is as open and lattice-like as the COPII coat and has similar connectivities in parts. However, Nup133 and Nup170 have no known structural homologs outside the NPC, and their specific integration into the NPC scaffold needs to be further analyzed. In addition, the NPC core scaffold contains two other large ␣-helical proteins, Nup188 and Nup192 (80,81), whose structures may reveal new surprises.
To build a rudimentary NPC scaffold, one might need only a few different structural building blocks. In an early eukaryote, we speculate, multiple copies of a small number of distinguishable elements formed a complete NPC scaffold. Gene duplications then created families of related nucleoporins. The members of each family evolved divergently into the distinct, nonredundant structural elements of extant NPCs.
This theory for the origin of the NPC explains the perplexing observation that many structural nucleoporins are not essential in yeast. Under stressed conditions, these genes can still partially complement one another. Given that the same structural elements can be detected in the NPCs of all extant eukaryotes, this process of diversification must have occurred in the early eukaryote. No extant eukaryote has a scaffold significantly simpler than the yeast or human NPC (82). The NPC likely evolved early in eukaryotic evolution, by multiplication of a few structural elements. Today, these multiplications are still evident, although the amino acid sequences of these ancestral elements have diverged greatly.
With a growing inventory of nucleoporin structures, the next task will be to determine the higher order assembly of the NPC. The anatomy of the NPC has been delineated by cryo-electron tomography (6 -8) and a computational analysis produced a draft of the NPC (30), but there are still various ways in which these nucleoporins might arrange to form the entire assembly. The lattice-like structure of the NPC provides few spatial restraints, and interactions between complexes are largely still unknown. However, the structures of the architectural nucleoporins, those presented here and those already available, narrow the speculations about the NPC assembly. These structures now allow us to probe the NPC by genetic manipulation of specific structural elements. Some models already can be ruled out. Altogether, from a combination of structural, computational, and cell biology research, the structure of the NPC is quickly emerging.