Complex Formation among the RNA Export Proteins Nup98, Rae1/Gle2, and TAP*

Most nucleocytoplasmic traffic through the nuclear pore complex is mediated by soluble receptors of the importin/exportin or karyopherin family. mRNA export is unique in that no receptor of this family has been implicated in trafficking of the bulk of mRNAs. Instead, many diverse proteins have been linked to mRNA export, but an all-encompassing model remains elusive. Understanding how these proteins interact with each other is central to the development of such a model. Here, we have focused on the interactions between three proteins implicated in mRNA export, Nup98, Rae1/ Gle2, and TAP. We have defined the binary complexes that form among these proteins. We find that Gle2 requires two sites within TAP for stable interaction. Strik-ingly, rather than a general affinity for all nucleoporin FG repeats, TAP has highest affinity for a specific region within the GLFG domain of Nup98, indicating that not all repeats are identical in function. We have established that the ternary complex can form through simultaneous binding of both Gle2 and TAP to adjacent sites on Nup98. In contrast, Nup98 competes with TAP for Gle2 binding; when bound to Nup98, Gle2 no longer interacts directly with TAP. From these interactions, we propose that Gle2 may act to deliver TAP to Nup98 and that this may represent the first in a series of interactions between an export complex and a nucleoporin. Traffic between the nucleus and cytoplasm proceeds exclu-sively through the nuclear pore complex, a multiprotein complex of (cid:1) 125 MDa in vertebrate cells (1–3). parallel, preformed Gle2-GBD to demonstrate vitro

Traffic between the nucleus and cytoplasm proceeds exclusively through the nuclear pore complex, a multiprotein complex of ϳ125 MDa in vertebrate cells (1)(2)(3). This immense structure is assembled from surprisingly few constituent proteins or nucleoporins; ϳ30 proteins make up the NPC in both yeast and metazoans (4,5). This seemingly low number may reflect the highly symmetrical nature of the pore. The bulk of the pore structure has 8-fold symmetry around a central axis perpendicular to the plane of the envelope with most nucleoporins present in 8, 16, or 32 copies per pore. The exact mechanism by which the nuclear pore carries out translocation is still uncertain, although several models are under investigation.
In addition to the pore complex itself, a number of soluble proteins are essential for nucleocytoplasmic transport. A large family of nuclear transport receptors termed importins/exportins or karyopherins recognize cargoes through specific import or export signal sequences (6 -8). Assembly and disassembly of the receptors and their cargo are highly regulated by the GTPase Ran. Strict compartmentalization of the Ran accessory factors, the cytoplasmic Ran GAP and the nuclear Ran GEF (RCC1), generates a Ran gradient across the nuclear envelope, which regulates the assembly and disassembly of receptorcargo complexes (9). Different members of the importin/ exportin receptor family specialize in the trafficking of specific cargoes. Import of the classical, basic nuclear localization sequence is mediated by the importin ␣/␤ heterodimer, and other protein classes, such as ribosomal proteins, have dedicated import receptors. The workhorse of nuclear protein export is Crm-1/exportin-1, which binds and transports leucine-rich nuclear export sequence-containing proteins. Similarly, ribosomal subparticles, small nuclear RNAs, and tRNAs are transported by exportin family members.
In contrast, no exportin family member has emerged as the direct mediator of mRNA export, although two recent reports suggest a contribution of transportin 2/karyopherin ␤2B to the export of at least some mRNAs (10,11). Instead, a plethora of proteins have been implicated in mRNA export, although the interactions of all these factors and how they carry out mRNA trafficking is still unclear (reviewed in Refs. [12][13][14]. The best characterized candidate for the mRNA export receptor has been the protein TAP or NXF in metazoans and its yeast homolog, Mex67. TAP is a modular protein with domains involved in RNA binding, nuclear pore binding, and interactions with numerous other proteins, themselves implicated in mRNA processing, splicing, or export (15)(16)(17). TAP contains two sites for binding to FG repeat nucleoporins. One binding site is found at the C terminus of TAP and is part of the ubiquitin-associated fold (UBA) 1 (18). The second binding site is formed upon interaction of a central domain of TAP with the co-factor p15/NXT (16,19,20). Two pore binding sites within TAP are required for efficient mRNA export; however, it appears that two copies of either type can function as well as one copy of each (21).
In addition to TAP, another essential player in mRNA export is a TAP binding partner, REF/Aly or Yra1 in yeast (reviewed in Refs. 12 and 13). REF/Aly is recruited to mRNAs during splicing, at least in part through binding to the splicing factor UAP56, a component of the exon junction complex, which is assembled ϳ20 nucleotides upstream of exon-exon junctions in spliced mRNAs (22). REF/Aly, in turn, is thought to recruit the TAP/NXT heterodimer to the mRNP complex. Also associated with the spliced mRNA complex are several proteins involved in cytoplasmic nonsense mediated decay (reviewed in Ref. 23). Some of these mRNA export proteins transit to the cytoplasm as part of the exported mRNP complex.
A number of additional proteins have been implicated in mRNA export, although their contributions are not fully characterized. In most cases, mutation of the yeast homolog generates a temperature-sensitive accumulation of nuclear poly(A) ϩ RNA. Dbp5 is a shuttling DEAD box RNA helicase that associates with the cytoplasmic face of the nuclear pore (24). Gle1 and Rae1/Gle2 also bind to the nuclear pore and are implicated in mRNA export (25)(26)(27). In Saccharomyces cerevisiae, a complex of Sac3p and Thp1p binds to Sub2p (yeast UAP56), and mutations in either Sac3p or Thp1p lead to mRNA export defects (28,29). Both Sub2p and Yra1p (yeast REF/Aly) are also members of the TREX complex that is implicated in both mRNA transcriptional elongation and export (30). Some of these proteins also bind to TAP/Mex67p; for others, molecular interactions are uncharacterized. Thus, many factors have been implicated in various systems and through various approaches, but the complicated interrelationships and mechanistic contributions to mRNA export are not fully understood.
We have found in Xenopus egg extracts that the nucleoporin Nup98 is present as a stable complex with one of its binding partners, Rae1/Gle2, a factor implicated in mRNA export in both yeast and metazoans. Moreover, both Nup98 and Gle2 can bind directly to TAP. In characterizing the individual binding interactions among these proteins, we have found that the Gle2-TAP interaction requires both a specific region within the p15/NXT binding domain of TAP and sequences within the N-terminal 60 amino acids of TAP whose role in TAP function has been unclear. TAP, in turn, shows a strong preference for a subset of nucleoporin repeat motifs within Nup98. Although both Gle2 and TAP can be simultaneously associated with Nup98, Gle2 bound to Nup98 no longer interacts directly with TAP. From these results, we suggest a possible model in which Gle2 aids in delivery of TAP to Nup98 and the nuclear pore complex.

EXPERIMENTAL PROCEDURES
Immunoprecipitation of Nup98-Gle2 Complexes-Xenopus egg extracts, wheat germ agglutinin (WGA)-binding proteins, and Nup98depleted extracts were prepared from Xenopus egg extracts as previously described (31). For immunoprecipitation, WGA-binding proteins were diluted with 4 volumes of a mild RIPA buffer (50 mM Tris-Cl, pH 7.5, 150 mM NaCl, 0.1% Nonidet P-40, 10 g/ml aprotinin, 10 g/ml leupeptin) and spun briefly at 14,000 ϫ g to remove any aggregated protein. Affinity-purified anti-Nup98 antibody (31) or nonspecific rabbit IgG was added for 2 h at 4°C, the sample was diluted 5-fold in RIPA, and incubated for 2 h at 4°C with Protein A-Sepharose. The beads were washed five times with 10 volumes of RIPA buffer, and the bound protein was eluted by heating in Laemmli gel sample buffer. Proteins were resolved by SDS-PAGE (8%) and detected by silver staining (Silver Stain Plus; Bio-Rad). When antigen was pretreated with SDS, the detergent (final concentration 1%) was added to the WGA-binding proteins, incubated for 10 min at room temperature, and then diluted 80-fold in RIPA buffer and used for immunoprecipitation.
Purification of Xenopus Gle2-For purification of the 42-kDa doublet, 25 1-ml aliquots of frozen Xenopus egg extract were pooled and diluted with 3 volumes of ice-cold TEN buffer (30 mM Tris, pH 7.2, 300 mM NaCl, 1 mM dithiothreitol, 10 g/ml aprotinin, 10 g/ml leupeptin, 1 mM phenylmethylsulfonyl fluoride). The diluted extract was centrifuged in an SS34 rotor at 10,000 rpm, 4°C for 1 h to pellet any aggregates or residual membrane vesicles, and then incubated overnight at 4°C with a volume of WGA-Sepharose equivalent to onetenth the starting egg extract volume. Beads were washed three times with 10 volumes of cold TEN buffer, four times with 10 volumes of Mono Q buffer (20 mM Tris, pH 7.8, 50 mM KCl, 2 mM MgCl 2 ), and then eluted by two sequential elutions (4°C, 45 min each) with 1 bead volume of 0.25 M GlcNAc, 8 mM trichitotriose in Mono Q buffer. The elutions were pooled and loaded onto a 1-ml Hi-trap Q FPLC column (Amersham Biosciences) and eluted with a linear 0 -0.5 M NaCl gradient. Fractions containing the 42-kDa doublet were pooled and used for preparative SDS-polyacrylamide gels (8%). After staining with Coomassie Blue, the upper and lower bands of the doublet were cut out individually and used for peptide sequencing.
Cloning of Xenopus Gle2-Three peptide sequences were obtained from Edman sequencing (KFTASCDK, NKPTGFALGSI, and NYFLR-NAAEELKP) and compared with the GenBank TM and expressed sequence tag data bases using the BLAST program (32). Matches were found with both human and mouse sequences as well as homology to cDNA sequences of Schizosaccharomyces pombe Rae1 and S. cerevisiae Gle2. Following alignment of homologs from all four species, regions of identity were used to design primers for degenerate PCR. A 142-bp PCR product corresponding to amino acids 149 -189 of Xenopus Gle2 was obtained and used to screen a cDNA library derived from Xenopus oocytes (Peter Klein, Howard Hughes Medical Institute, University of Pennsylvania) from which a full-length Xenopus Gle2 clone was isolated and sequenced.
Antibody Production and Immunoblots-To raise antibodies to the xGle2 protein, the full-length Xenopus Gle2 gene was cloned into the pET-28 vector to express xGle2 protein with an N-terminal hexahistidine tag. This protein was produced in bacteria, solubilized in urea, and purified on a column of His-bind resin according to the manufacturer's instructions (Novagen, Madison, WI). Purified protein was then used to immunize rabbits (Zymed Laboratories, S. San Francisco, CA). Affinity purification of anti-xGle2 antibodies was performed using nitrocellulose strips containing Xenopus Gle2 protein, as previously described for purification of antibodies to Xenopus Nup98 (31). Immunoblots were performed as previously described (33). Affinity-purified antibodies were used at a 1:1000 dilution in 2% BSA, 1ϫ PBS, 0.2% Tween 20.
Production of GST-TAP Beads-The full-length or 61-619 human TAP genes pGEX-4T were obtained from Bryan Cullen (Howard Hughes Medical Institute, Duke University Medical Center). For protein expression, pGEX-TAP was induced in BL21 cells overnight at 18°C with 0.5 mM isopropyl-1-thio-␤-D-galactopyranoside. Cells were lysed in the presence of 1ϫ Complete protease inhibitors (Roche Applied Science) using a French press, and GST-TAP was purified on glutathione beads (Amersham Biosciences) according to the manufacturer's directions. After elution, glutathione was removed using a centrifugal concentrator (Ultrafree; Millipore Corp.; Bedford, MA), and GST-TAP was rebound to beads at ϳ2.5 g of protein/l of beads. Because GST-TAP was subject to some degradation even in the presence of protease inhibitors, the exact amount of bound protein was adjusted to give equal amounts of full-length TAP.
For preparation of TAP/NXT beads, GST-TAP (F617A) and untagged p15/NXT were expressed separately in BL21 cells. After induction of protein, cells were lysed, and the lysates were combined at a ratio of one part GST-TAP to two parts p15/NXT and incubated for 1.5 h at 4°C to allow binding. The resulting GST-TAP/NXT complexes were then purified on glutathione beads as described above. Formation of complexes was confirmed by SDS-PAGE and Coomassie staining or immunoblotting. Rabbit anti-p15/NXT was a gift from Bryce Paschal (University of Virginia).
In Vitro Binding Assays-GST or GST-TAP beads (10 l per assay) were preblocked in 400 l of 10 mg/ml BSA in 1ϫ PBS, 0.1% Tween 20 (PBST) with 1ϫ Complete protease inhibitors for 1 h at 4°C. The blocking solution was then removed, and beads were used in binding assays. Human Nup98 in pcDNA3.0, mouse Gle2 or Xenopus Gle2 in pCS2, or 6ϫMyc-mGle2 in pCS2-MT was transcribed and translated in vitro using the TNT kit (Promega, Madison, WI) with [ 35 S]Met, according to the manufacturer's instructions. Translation reactions (typically 5 l) were diluted in 400 l of 1 mg/ml BSA in PBST and incubated with preblocked GST or GST-TAP beads for 1 h at 4°C. Beads with bound proteins were then washed five times with 400 l of PBST at 4°C. After removal of all supernatant, bound proteins were eluted by boiling with 25 l of 1.5ϫ Laemmli gel sample buffer. Typically, 6 l of eluate was loaded per lane on SDS-PAGE. Gels were soaked for 30 min in 1 M sodium salicylate, dried, and exposed to film overnight.
For binding between GST-TAP and translated p15/NXT, 5 l of p15/NXT translation was diluted in 25 l of unprogrammed reticulocyte lysate and incubated with GST or GST-TAP beads as above. Beads were washed, and bound proteins were analyzed as described above.
For prebinding of Gle2 to the N-GBD of Nup98, the N terminus of Nup98 (amino acids 1-224) was cloned into pET-28, and the T7 tagged Nup98 fragment was produced in vitro using the TNT kit. MGle2 (5 l) was mixed with 7 volumes of Nup98 N-GBD translation and incubated 30 -60 min to allow binding. This binding reaction was then diluted to 400 l in 1 mg/ml BSA in PBST, incubated with 10 l of preblocked GST-TAP beads, and then analyzed as above. In parallel, a duplicate binding reaction was diluted in 1 mg/ml BSA in PBST and incubated with preblocked anti-T7 beads to immunoprecipitate the Nup98 Gle2binding domain (GBD or GLEBS) fragment along with its associated Gle2. Bound proteins were eluted and analyzed as above.

Nup98 in Xenopus Extracts Is Stably Bound to Rae1/Gle2-
Nup98 is a dynamic component of the nuclear pore complex that carries multiple copies of the nucleoporin repeat motif, GLFG. The Nup98 repeat domain has been proposed to bind to importin-␤/karyopherin family members that participate in nuclear trafficking (34,35). Additionally, Nup98 contains a binding site for the mRNA export factor, Rae1/Gle2 (GBD) (26). Gel filtration analysis of Xenopus extracts led to a predicted mass of ϳ450 kDa for Nup98, suggesting that it exists either as a homomultimer or in a stable complex with one or more of its binding partners (36). In Xenopus extracts, the nuclear pore complex exists in a disassembled form, equivalent to the disassembly seen during mitosis. In order to determine the form in which Nup98 exists in these extracts, we used Nup98-specific antibodies for immunoprecipitation, followed by SDS-PAGE and silver staining (Fig. 1). Along with Nup98, we consistently observed a protein doublet of ϳ42 kDa (Fig. 1, lanes 4 and 5). This doublet was not found in the absence of Xenopus protein or when control antibodies were used (Fig. 1, lanes 1-3). Coprecipitation is due to protein-protein interactions, since pretreatment of Xenopus extracts with SDS to disrupt interactions led to loss of the 42-kDa bands (data not shown). Further, these proteins co-fractionated with Nup98 on both lectin and MonoQ chromatography, and the co-precipitated complexes remain intact when treated with either 2 M NaCl or 2 M urea, confirming the existence of a stable complex (data not shown). We obtained sequence information for several peptides derived from the protein doublet (see "Experimental Procedures"), and these, together with antibody reactivity, demonstrated that both bands correspond to the Xenopus homolog of Rae1/Gle2.
Cloning and Localization of Xenopus Rae1/Gle2-In order to produce reagents with high specificity for Xenopus Rae1/Gle2, we cloned the Gle2 gene from a Xenopus cDNA library. The Xenopus gene was sequenced (GenBank TM accession number AY256464) and proved to be 92% identical to human and mouse Rae1/Gle2/mRNP41 (37,38). As is true for each of the previously cloned vertebrate genes, Xenopus Gle2 is ϳ40% identical to either of its yeast counterparts, S. pombe Rae1 and S. cerevisiae Gle2, which themselves share only ϳ50% identity (27,39). Thus, the S. cerevisiae and S. pombe homologs are nearly as divergent from each other as each is from the well conserved metazoan proteins. Antibodies generated against bacterially expressed Xenopus Gle2 recognized both bands of the 42-kDa doublet ( Fig. 2A). Multiple intracellular localizations have been proposed for Gle2 in different systems, primarily using epitopetagged protein. To further investigate this, we localized endogenous Xenopus Gle2 using our affinity-purified antibody. Gle2 was found primarily within the nucleus of Xenopus cells, with a small but reproducible fraction in the cytoplasm (Fig. 2B). Whereas Gle2-GFP was localized identically to the endogenous protein, GFP-Gle2 showed a proportionately greater association with the nuclear envelope (data not shown).
To determine whether all of the Gle2 existed in a complex with Nup98 or whether a fraction might potentially function independently, we probed Nup98-depleted extracts for the presence of remaining Gle2 (Fig. 2C). The vast majority of Gle2 was depleted along with Nup98; however, a small but significant fraction remained even when Nup98 was not detectable. (Fig. 2C, lanes 3 and 4).
TAP Is a Direct Gle2 Binding Partner-Given this indication that not all Gle2 was complexed with Nup98, we investigated other potential interactions in which Gle2 might participate. Both Nup98 (40) and Rae1/Gle2 (26,27,39) have been implicated in the export of mRNA from the nucleus. The protein TAP has been shown by multiple groups to be a major factor in the export of mRNA (reviewed in Ref. 13). TAP binds directly to nucleoporins and also to several other proteins with contributions to RNA processing and transport. We therefore asked whether TAP might be a binding partner for Nup98 and Gle2 in Xenopus extracts. When Xenopus extract was incubated with either GST or GST-TAP immobilized on beads, both Nup98 and Gle2 were observed to bind specifically to TAP (Fig. 3A). The relative amount of Gle2 bound in the TAP binding assays always appeared to be in excess of the amount co-precipitated with Nup98. This led us to ask whether there might be a direct interaction between TAP and Gle2. To test this possibility, the same binding assay was carried out using extracts that had been depleted of Nup98. The residual Gle2 present in the Nup98-depleted extracts bound to TAP beads in a Nup98independent manner (Fig. 3A, lanes 4 and 5). As further confirmation of these interactions, we translated Nup98 and Gle2 individually in reticulocyte lysates and assayed the direct interaction of each protein with GST-TAP (Fig. 3B). Both Nup98 and Gle2 bound strongly and specifically to TAP. To ensure that each of these interactions was direct and not bridged by the presence of the partner protein present in the reticulocyte lysate, we probed the lysates with antibodies to both Nup98 and Gle2. Although when mRNA was added to the lysates we could readily detect the translation products with the appro- priate antibody, we did not observe either Gle2 or Nup98 in unprogrammed lysate (Fig. 3C). Although the formal possibility exists that the antibodies did not recognize the rabbit proteins, they do cross-react with the human, mouse, and Xenopus homologs, indicating broad specificity. Given that we could readily detect the few nanograms of translated human or mouse protein, it is most probable that the rabbit protein would be detected if present. Thus, we conclude that TAP interacts directly with both Nup98 and Gle2.
Interaction between Gle2 and TAP Requires Two Regions of TAP-In order to understand the interactions between these three proteins and the contribution of such interactions to nuclear export, we set out to determine the binding sites for each of the partner pairs. The site of interaction between Gle2 and Nup98 has been previously mapped to a short sequence between amino acids 181 and 224 of Nup98 (26). Gle2 is a WD40 repeat protein, a class of proteins that typically form highly ordered ␤-propeller structures, not always amenable to deletion analysis (41). Point mutations at multiple sites within Gle2 reduce binding to Nup98 (D294A (26); data not shown). Given the lack of a three-dimensional model from which to predict non-structure-disrupting mutations, we have not attempted to map specific binding surfaces within Gle2.
In contrast, TAP is a 619-amino acid, modular protein with several independently folded domains and multiple reported binding partners (15,16,18). Most studied are the two distinct binding sites for repeat nucleoporins; one site is found within the ubiquitin-associated fold (UBA domain) at the C terminus  1 and 3) added plasmid DNA template were immunoblotted with antibodies specific for Nup98 (lanes 1 and 2) or Gle2 (lanes 3 and 4).
of TAP (amino acids 600 -619), and a second nucleoporin binding site is formed when the co-factor p15/NXT becomes complexed with residues 372-555 of TAP. Although binding of this site to nucleoporin FG repeats requires p15/NXT, direct molecular contacts appear to occur only between TAP and the FG of a single repeat motif, and the role of p15/NXT is indirect (16). In contrast to the nucleoporin binding function of the C terminus of TAP, the first 372 amino acids of TAP bind the CTE RNA of Mason-Pfizer monkey virus and mediate its export from the nucleus (43). Additionally, this N-terminal half of TAP has been reported to bind to multiple proteins including splicing factors, components of the exon junction complex, transportin 2/karyopherin ␤2B, and E1B-AP5 (11,13,44,45). To map where Gle2 and Nup98 each interact with TAP, we made a series of C-terminal truncations in TAP and tested the ability of each to bind Nup98 and Gle2. In parallel and as a control, we also confirmed the expected interaction with p15/NXT. We found that amino acids 601-619 were essential for the interaction between Nup98 and TAP; no significant binding was observed when TAP was truncated at amino acid 600. This is somewhat different from the results of Bachi et al. (45), who found that significant binding of Nup98 was maintained by TAP deleted of residues 567-613, although this mutant TAP was strongly impaired in binding to other FG repeat nucleoporins, including Nup153 and p62. As expected, p15/NXT bound to TAP 1-583 but no longer bound to the shorter TAP 1-507 (Fig. 4B, lower panel). The results of the Gle2 binding assays indicated that Gle2 interacts with a subregion of the p15/NXT binding site within TAP. TAP 1-445 retained full Gle2 binding although TAP 1-372 was unable to interact with Gle2 (Fig. 4B, middle panel). Thus, sequences between 372 and 445 are essential for interaction with Gle2. Collectively, these data demonstrate that Gle2 can interact with TAP even in the absence of Nup98 or NXT binding, reinforcing the conclusion that Gle2 binds directly to TAP.
In order to further define the boundaries of the Gle2 interaction domain, we carried out deletions from the N terminus of TAP. Unexpectedly, when the first 60 amino acids were removed from TAP, the protein was no longer able to interact with Gle2. However, this deleted protein retained structural For some assays, the same truncations were expressed in vitro as HA-tagged proteins. B, GST or GST-TAP beads (10 l) containing the form of TAP indicated by FL or the construct number were used in binding assays with in vitro translated Nup98 and Gle2 as in Fig. 3B. For binding between TAP and p15/NXT, the same TAP truncations were produced with the HA tag at the N terminus. The different HA-TAP proteins and p15/NXT were translated in vitro, and equal volumes of 15/NXT and the indicated HA-TAP protein were incubated together to allow binding. The resulting complexes were isolated on anti-HA beads. C, GST, GST-TAP, or GST-TAP 61-619 (⌬1-60) beads were used to bind in vitro translated Nup98 or Gle2. D, HeLa cells were transfected with HA-TAP 1-619 (FL) or HA-TAP 61-619 (⌬1-60) expression vectors either with or without co-transfected Gle2 plasmid. Cell lysates (lanes 1-6) were used for immunoprecipitation with anti-HA. Bound material was analyzed by PAGE and immunoblotting. The positions of each form of TAP are indicated (arrowheads). An anti-HA cross-reacting band (asterisk) appeared on immunoblots of lysates (lanes 1-6) but immunoprecipitated relatively inefficiently (lanes 7-12).
integrity, since it was still able to bind Nup98 (Fig. 4C). This requirement for the N terminus of TAP in binding Gle2 was confirmed in vivo. HeLa cells were transfected with HA-tagged TAP or HA-tagged TAP⌬1-60. Parallel samples were also cotransfected with untagged human Gle2. Antibodies to the HA tag were then used to immunoprecipitate TAP from cell lysates, and the co-precipitation of Gle2 was assessed on immunoblots. We found that whereas full-length HA-TAP co-precipitated either the endogenous or transfected Gle2, ⌬1-60 TAP did not interact with Gle2 in vivo, confirming the results of the in vitro binding assay (Fig. 4D, compare lanes 7 and 9 with lanes 8 and  10). Thus, binding between TAP and Gle2 requires multiple sites in the TAP protein. The region of TAP between amino acids 372 and 445 constitutes a subset of the p15/NXT binding site and is composed primarily of several stretches of ␣-helix (16). In contrast, the N-terminal 118 amino acids of TAP are relatively unstructured. This region has not been crystallized, but secondary structure predictions indicate only a few short stretches of ␤-sheet with multiple potential loops (46). Thus, the two regions of TAP implicated in Gle2 interaction exhibit no obvious structural similarities. Between these regions lie both the RNP-like domain of TAP and the leucine-rich repeat domain, structures implicated in binding to CTE RNA. However, it is clear that these domains are not sufficient for interaction with Gle2.
TAP Binds to a Specific Subset of Nup98 GLFG Repeats-We next asked whether we could determine precisely where Tap bound to Nup98. Complexes between TAP and a nucleoporin FG repeat have been analyzed at the structural level (16,18). Both the TAP/NXT nucleoporin binding site and the C-terminal 600 -619 binding site directly contact only the two amino acids, phenylalanine and glycine, that comprise the core of the repeat motif. This would suggest that all copies of a repeat motif should bind with approximately similar affinity. To test this model, we again used the in vitro binding assay to ask whether a specific TAP interaction site was present within the Nup98 repeat domain. Various subdomains of Nup98, as diagrammed in Fig. 5, were translated and then incubated with either GST or GST-TAP beads. Somewhat surprisingly, we found that the 155-amino acid N-terminal domain of Nup98, which contains GLFG (3) and FG (13) motifs, did not interact with TAP. In contrast, all constructs that contained the entire central repeat domain bound strongly to TAP. When the central GLFG domain was expressed alone, binding was weaker than when the domain was flanked by sequence at either end. This most likely results from defects in folding in the absence of any other domains, since the effect was not specific for either N-or C-terminal flanking domain (compare lanes 14, 20, and 22). The 373-amino acid central GLFG domain contains six copies of the GLFG motif and 16 copies of the FG motif, several of which FIG. 5. TAP binds a specific subset of GLFG repeats. A series of Nup98 constructs were prepared as diagrammed. Each Nup98 protein was translated in vitro and used in pull-down assays with GST or GST TAP beads. The left panel shows 0.12 l of each translation reaction. In the right panel, odd-numbered lanes correspond to material bound to GST, and even-numbered lanes contain protein bound to GST-TAP beads. Assays were carried out as in Fig. 3B. occur in the context of presumably degenerate GLFG motifs, such as TLFG, SLFG, and GAFG. Surprisingly, we found that the TAP binding site was contained within a further subset of the repeat domain. Constructs containing the N-terminal half of the repeat domain actively bound to TAP, whereas the Cterminal half of this domain did not bind to TAP. It is unclear precisely what functionally distinguishes these two regions. The N-terminal half of this domain is enriched in GLFG motifs (five GLFG and seven FG), whereas the C-terminal half has more FG and degenerate GLFG motifs.
For the above binding assays, the beads contained full-length TAP but not p15/NXT. Thus, binding should occur between Nup98 and the nucleoporin binding site at the C terminus of TAP, amino acids 600 -619. Just as it was unclear whether a binding site that interacts only with the FG of the repeat motif would prefer a particular subset of repeats, it was unknown whether the second, p15/NXT-dependent nucleoporin binding site of TAP would similarly show a preference for certain repeats. Consequently, we prepared beads carrying the F617A mutant of TAP, in which the C-terminal binding site is disrupted (45), complexed with p15/NXT. In these complexes, only the site generated by interaction of TAP and p15/NXT is functional for nucleoporin binding. We again carried out the same set of binding interactions with identical results (data not shown). The TAP and p15/NXT complex exhibits the same preference for the amino-terminal half of the central repeat domain of Nup98. As previously reported, in the absence of p15/NXT, we saw minimal binding between the F617A mutant and Nup98.
Analysis of Ternary Complex Formation-We had now established how each of the binary complexes could form. As a next step toward understanding the role of these interactions in export, we wanted to determine whether and how a tripartite complex could assemble. By determining which ternary complexes could form among this set of proteins, we could hypothesize an order of interaction and reveal a step or steps in RNA export. We first asked whether Gle2 prebound to Nup98 retained the ability to interact directly with TAP. This complex should mimic Gle2 at the nuclear pore. Translated Gle2 was mixed with a 7-fold excess of a translated, N-terminal fragment of Nup98 that contained the Gle2 binding site but was unable to interact with the nucleoporin binding sites of TAP. Following an incubation to allow association between Gle2 and the Gle2 binding domain, this complex was diluted and used in a GST-TAP binding assay. Formation of the Gle2-Nup98 complex was confirmed by immunoprecipitation using antibodies to the T7 tag at the N terminus of the Nup98 fragment. We found that, as expected, Gle2 and the Gle2 binding Nup98 fragment were bound tightly together (Fig. 6, lane 8). As before, when Gle2 alone was incubated with GST-TAP, binding was specific and strong. However, when Gle2 was prebound to the Gle2 binding domain, the vast majority of TAP binding was lost (Fig. 6, lane  4). Only minimal association was observed between GST-TAP and the translated Nup98 fragment that contained the Gle2 binding site (Fig. 6, lane 6). Thus, Gle2 can bind to TAP but only when Gle2 is not bound to Nup98. This argues that if the ternary complex occurs, it is not formed by Gle2 stably bridging an interaction between TAP and Nup98. Further, this result indicates that Gle2 at the nuclear pore complex is not a target for TAP binding, suggesting that the TAP/Gle2 interaction may occur only in the nucleoplasm.
To determine whether a ternary complex might arise from simultaneous binding of Gle2 and TAP to Nup98, we made use of an N-terminally tagged form of Gle2 that we found was impaired in binding to TAP (Fig. 7, compare lanes 3 and 7) but remained fully capable of binding to Nup98 (Fig. 7, lane 12; data not shown). Myc-tagged Gle2 and Nup98 were translated and mixed to allow complex formation as before. This complex was then assayed for binding to TAP. Since Gle2 prebound to Nup98 would not interact efficiently with TAP, and since the Myc-Gle2 used in this assay only weakly bound to TAP, we reasoned that Gle2 should be found on TAP beads only if the interaction is bridged by Nup98. This was the case; the amount of Gle2 present when the complex was isolated on GST-TAP was equivalent to the amount of Gle2 present when the same complex was isolated in parallel using antibodies to Nup98 (Fig. 7, compare lanes 11 and 12). Thus, whereas TAP does not bind Gle2 at the pore, Tap can bind to its adjacent site on Nup98 when Gle2 is also bound to Nup98. DISCUSSION In this report, we have demonstrated the existence of binary and ternary interactions between the nuclear transport factors Nup98, Rae1/Gle2, and TAP. Despite the ability of Gle2 and TAP to interact in the absence of Nup98, we found that Gle2 and TAP no longer interact directly when bound to Nup98. Rather, TAP establishes interactions with Nup98, whereas Gle2 binds to Nup98 at a distinct adjacent site. These observations raise the intriguing possibility that the TAP-Gle2 complex is an intermediary in the transport process that participates in the delivery of TAP to the nuclear pore.
A Stable Complex between Nup98 and Gle2 in Egg Extracts-Nup98 has been proposed to interact with a number of different proteins including Rae1/Gle2, karyopherins (importin ␤, trans-  1 and 2) or Gle2-GBD complex (lanes 3 and 4) incubations. In parallel, the same preformed Gle2-GBD complex was immunoprecipitated with anti-T7 antibodies to demonstrate complex formation. In vitro translated GBD showed minimal binding to TAP in the absence of Gle2 (lanes 5 and 6). portin), RCC1, and other nucleoporins (tpr, Nup96, Nup88) (26,34,35,(47)(48)(49). Therefore, it is perhaps surprising that Nup98 would be stably associated specifically with Gle2 in Xenopus extracts. Nup98 is the only known binding site for Gle2 at the nuclear envelope; however, both Nup98 and Gle2 shuttle between the nucleus and cytoplasm (26,33). Whether these partners shuttle as a complex has not been established. The steady state localization of the two proteins in Xenopus cells is not identical, although both proteins can be found within the nucleoplasm. Whereas a substantial amount of Xenopus Nup98 is associated with intranuclear structures we termed GLFG bodies (33), Gle2 does not appear to be concentrated in these structures. It is entirely possible that the interaction between Nup98 and Gle2 is dynamic during interphase in vivo. In support of this, our initial photobleaching experiments using Gle2-GFP indicate that Nup98 and Gle2 shuttle between the nucleus and cytoplasm at different rates (33). 2 The nuclear pore is disassembled into conserved subcomplexes at mitosis, and these most likely correspond to the complexes observed in Xenopus egg extracts (36). Thus, even if Nup98 and Gle2 are only transiently bound in vivo, the interaction could be stabilized when nucleocytoplasmic trafficking ceases during mitotic disassembly.
Gle2 has recently been proposed to have a mitotic role as a checkpoint protein (50,51). A partial localization to the kinetochore was demonstrated using a FLAG-tagged Gle2 protein (52). We have not observed a kinetochore localization pattern with antibody to endogenous xGle2, but careful analysis by deconvolution microscopy is under way and should address this possibility more rigorously. Neither we nor others have observed Nup98 at kinetochores during mitosis; however, checkpoint regulation could be a role for the small fraction of Gle2 that seems to be independent of Nup98. Possibly, the sequestration of the great majority of Gle2 into a complex prevents interference with the checkpoint role of this abundant protein.
We note that Gle2 always appears as a doublet in Xenopus extracts and that both forms interact with Nup98. The nature of this doublet is as yet unknown, but it is not converted to a single band by treatment with a broad specificity phosphatase. Intriguingly, identical observations were recently reported for Xenopus Bub3, the closest homolog of Rae1/Gle2 and a checkpoint pathway protein (53).
Gle2-TAP Interactions-Gle2 has the capacity to interact directly with TAP. The presence of Gle2 among a group of TAP-binding proteins was previously observed in yeast. In S. cerevisiae, a direct interaction between Gle2 and Mex67 (yeast TAP) was reported, although in S. pombe direct binding could not be detected, and it was proposed that Gle2 bound indirectly as part of a multiprotein complex. In HeLa cell nuclear extracts, Gle2 was one of a number of proteins that bound to GST-TAP, and in vitro binding assays suggested a direct interaction between Gle2 and the C-terminal half of TAP (45). Here we have demonstrated that Gle2 binds directly to a region within the p15/NXT binding domain of TAP. Specifically, we found that amino acids 371-455 are required for this interaction. Based upon the crystal structure of this domain of TAP, these residues correspond to the ␣-helix opposite from the TAP/NXT binding interface as well as the short helices and loop that form the "insertion loop" (16). The relative positions of the p15/NXT and Rae1/Gle2 binding interfaces would suggest that these two factors could interact with TAP simultaneously. In agreement with this prediction, we did not observe a significant reduction of Gle2 binding to the TAP/NXT complex. 3 Bacchi et al. (45) had previously reported that the C-terminal 372-619 region was sufficient for Gle2 binding; unfortunately, we were unable to confirm this, since, in our hands, this Cterminal half of Tap was not soluble when expressed in bacteria. GST-TAP 372-455, a region we identified as essential for Gle2 binding, was soluble but on its own was insufficient to bind to Gle2 in vitro.
We found that deletion of the first 60 amino acids from otherwise full-length TAP dramatically impaired Gle2 binding, both in vitro and in vivo. No function has been definitively ascribed to this region of TAP, which is poorly conserved in the yeast Mex67 homologs. ⌬1-60 TAP was reported to be less efficient than full-length TAP in promoting the export of intronless RNAs. This deletion had no effect on the export of an intron-containing RNA that would necessarily interact with the splicing machinery before export (54). Possibly, this region aids in splicing-independent recruitment of factors that are otherwise assembled on the mRNA after processing. E1B-AP5 is an hnRNP-like protein that binds to the N-terminal half of TAP (amino acids 1-372) (45). Like Gle2, E1B-AP5 requires the first 60 amino acids of TAP for binding but also requires a second domain; TAP 1-265, in which the leucine-rich repeat domain is deleted, showed a marked reduction in E1B-AP5 binding (45). Sub2 (the yeast homolog of UAP56), which inter- acts with the exon junction complex and is part of the TREX complex, also binds to Tap and is required for export of intronless RNAs, although it is not clear how or when this protein would be recruited to intronless transcripts (55).
Gle2 and mRNA Export-The precise contribution of Rae1/ Gle2 in mRNA export is unclear. In S. pombe, point mutations in Rae1 are temperature-sensitive and synthetically lethal with deletion of the nonessential S. pombe Mex67 (56). In S. cerevisiae, the Gle2 null is ts lethal and extremely growth impaired at the permissive temperature. Point mutations in both Rae1 and Gle2 result in defects in poly(A) RNA export at the nonpermissive temperature (27,39). In HeLa cells, overexpression of the Nup98 GBD, which should competitively inhibit interaction between Gle2 and Nup98, was reported to block mRNA export, suggesting an essential role for this interaction in metazoans (26,57). However, the mouse Rae1/Gle2 knockout has no mRNA export defect (50). It is as yet unclear how to reconcile these results. We found that Gle2 prebound to Nup98 could no longer interact with TAP. These data could suggest a molecular mechanism whereby overexpressed GBD interacts with Gle2 and prevents not only pore targeting of Gle2 but also Gle2-TAP binding. However, it is uncertain why this should have a stronger mRNA export phenotype than deletion of Gle2. Possibly, binding to Nup98 alters Gle2 interaction with as yet unidentified partners.
Although the Gle2-Nup98 complex no longer interacts with TAP via Gle2, the ternary complex still forms via bridging by Nup98. Indeed, the TAP binding site is located immediately adjacent to the Gle2 binding site on Nup98. This same positioning was observed in S. cerevisiae Nup116 using the twohybrid system (58). It is intriguing to speculate that Gle2 might aid in directing TAP to Nup98 and then, once bound, release TAP to interact directly with the adjacent repeat motifs and presumably progress to other FG nucleoporins. In keeping with this, Sabri and Visa (59) observed that Gle2 was associated with the nuclear pore only when the Balbiani ring granule, an RNA export complex, bound to the basket of the pore.
Interaction between TAP and FG Repeats-It is not apparent why TAP should exhibit a preference for certain FG motifs. Nothing obviously distinguishes this stretch of repeats, yet the same preference is observed whether TAP interacts with Nup98 via the C-terminal UBA domain or through the TAP/ NXT heterodimer site. The structure of each TAP domain bound to a nucleoporin repeat peptide has been determined, and the major molecular contacts observed were between TAP and the phenylalanine of the FG motif (16,18). This finding would predict that all repeats are recognized identically. When a structural analysis of a nucleoporin repeat domain was attempted (18), only the FXFG motif was visible; the rest of the domain was too disordered for diffraction analysis. Rexach and colleagues (42) have proposed that FG repeat domains may correspond to "natively unfolded proteins." This class of proteins possesses a characteristic amino acid composition and typically elutes from gel filtration columns at greater than the predicted molecular weight. In keeping with this model, the Nup98 Gle2 complex was estimated at 450 kDa (36), yet we have been unable to discern any in vitro homotypic interactions of either Nup98 or Gle2. If the Xenopus complex corresponds to one copy of each partner to yield a predicted size of 130 kDa, the observed molecular mass of 450 kDa would be consistent with the natively unfolded model. However, our finding that not all repeats function equivalently suggests that, whereas the structure may be too disordered for crystallographic analysis, there may be some contextual or structural information present. Resolution of this issue will be important to defining the roles of the FG repeat nucleoporins.