Nup60 ‘suspension cable’ oligomerizes via its helical region

Nup60 was identified as the basket’s membrane-anchored suspension cable, containing SLiMs that interact directly with Mlp1 and Nup2 (ref. 40). To determine how the basket docks onto the core of the NPC, we employed an artificial intelligence (AI)-based SLiM screen, extending AlphaFold-Multimer’s success in structure prediction towards identifying unknown protein–protein interactions. We segmented Nup60 into pieces of equal length (31 amino acids (aa)) covering all conserved SLiMs and determined the average interface predicted template modelling (ipTM) score for every binary combination with all Nups (Fig. 1a). Out of 264 combinations tested, two pairs stood out by their high ipTM score: the helical region (HR) of Nup60 (HR; aa119–149) with itself, and a SLiM at aa285–315 with Nup85. The HR of Nup60 is predicted to adopt a helix-turn-helix fold with two copies arranged in an antiparallel orientation, held together by a conserved hydrophobic patch that is buried at the dimer interface (Fig. 1b; see Extended Data Fig. 1a,b for model confidence). The solvent-exposed dimer surface is largely hydrophilic (Extended Data Fig. 1c–f). To verify a potential oligomerization of Nup60, we performed a two-step split-affinity purification of differentially tagged Nup60 HR constructs co-expressed in Escherichia coli. We fused the Nup60 HR (aa48–162; including an N-terminal linker) with either a His6 or FLAG tag and appended a NusA (57 kDa) or SUMO (14 kDa) tag to create a size difference. Following a two-step affinity purification on Ni-NTA and anti-FLAG beads, we retrieved both NusA and SUMO-tagged Nup60 species (Fig. 1c), indicating an assembly of at least two Nup60 HR copies.

Fig. 1: The Nup60 suspension cable oligomerizes via its HR.

a, Heatmap of the average ipTM scores for every possible pairwise combination between S. cerevisiae Nup60 SLiMs and full-length Nups as estimated by the AlphaFold-Multimer model. Nup60 query SLiMs are stretches of 31 amino acid residues covering regions evolutionarily conserved across yeast orthologues (yellow and dark green shading in Nup60 cartoon). N2BM, Nup2-binding motif. b, The AlphaFold-Multimer model of Nup60(110–160) dimer shown in cartoon representation (left) and a combination of cartoon and surface representations (right) with surface coloured by sequence conservation, Coulombic electrostatic potential and hydrophobicity (from left to right). The oval outline (orange) depicts a conserved hydrophobic patch buried at the dimer interface. For model confidence and dimer surface representations, see Extended Data Fig. 1. c, SDS–PAGE of a two-step split-affinity purification of differentially tagged Nup60(48–162) constructs co-expressed in E. coli. The protein eluates were analysed by SDS–PAGE and Coomassie staining. d, Dilution ITC profiles (isotherms) of NusA-His6-Nup60(48–162) and NusA-His6 samples (170 μM) are shown. By applying a simple dimer dissociation model, the dissociation constant (KD) of NusA-Nup60(48–162) was estimated to be 76.3 ± 7.9 μM (mean ± s.d., n = 3). Source numerical data and unprocessed gels are available in Source data.

Source data

To characterize Nup60 oligomerization quantitatively, we performed dilution isothermal titration calorimetry (ITC) experiments with the recombinant NusA-His6-Nup60(48–162) protein. The dilution isotherms displayed a hyperbolic profile, characteristic of a simple dimer dissociation model, whereas NusA-His6, used as a negative control, did not exhibit such behaviour (Fig. 1d and Extended Data Fig. 1g,h). The dissociation constant (KD) of the Nup60 HR was 76.3 ± 7.9 μM (mean ± s.d.). Additionally, we performed size exclusion chromatography coupled with multi-angle light scattering experiments. Consistent with Nup60 dimerization, we observed a concentration-dependent shift of the NusA-His6-Nup60(48–162) peak towards higher molecular mass (Extended Data Fig. 1i). However, a low peak resolution prevented us from accurately estimating the molecular mass of the putative dimer peak. We cannot exclude the possibility that Nup60 undergoes higher-order oligomerization, beyond just forming a dimer. Conversely, there was no discernible shift in the NusA-His6 negative control, thereby ruling out an influence of the tag on Nup60 dimerization (Extended Data Fig. 1j). Dimerization of Nup60, as explained below, is expected to affect the connectivity of the basket, its docking onto the NPC core and may stabilize the head-to-tail arrangement of Y-complexes in the nuclear ring of the NPC.

Nup60 interacts with the Nup85•Seh1 arm of the Y-complex

The second-best hit identified in our AlphaFold-based SLiM screen suggested an interaction between Nup60 and Nup85, a subunit of the Y-complex. Surprisingly, this stretch of 31 amino acids (aa285–315) is embedded in a larger region of Nup60 (aa240–318, termed Nup60 Mlp1-binding motif (MBM)) that was previously shown to recruit the basket subunit Mlp1 (ref. 40). This finding raised the possibility that the Y-complex and the Mlp1 filaments bind to Nup60 in immediate proximity. Mlp1 was not detected in this binary screen, possibly because it interacts with Nup60 only as a coiled-coil homodimer (see further below). Moreover, the expected interaction between Nup60 and Nup2 (ref. 40) was not detected, suggesting that AlphaFold encounters challenges in predicting SLiM-mediated interactions.

To experimentally verify a putative Y-complex–Nup60 interaction, we aimed to reconstitute it in vitro. To this end, we engineered a stable S. cerevisiae Nup60 construct, which lacks the N-terminal lipid-interacting AH (aa1–47) and the disordered FG-repeat region (aa381–504). This Nup60 construct (aa48–539ΔFG; abbreviated as Nup60*) was purified from E. coli via an N-terminal glutathione S-transferase (GST)-tag and immobilized on glutathione (GSH) beads as a bait (Fig. 2a). The Y-complex was affinity purified from an S. cerevisiae strain expressing Nup133 with a C-terminal TAP tag (Fig. 2b). When incubated with each other, the Y-complex bound Nup60* (Fig. 2c and Extended Data Fig. 2a), but did not bind to GST used as a negative control. Next, we deleted conserved regions of Nup60, which overlap or lie adjacent to the previously identified direct binding site for Mlp1. Deleting the previously mapped Mlp1 binding site on Nup60 (aa240-309) also reduced the affinity for the Y-complex (Fig. 2c). Deletion of the adjacent Nup60 region (aa200–239) similarly resulted in a partial loss of Y-complex interaction (Fig. 2c). Interestingly, deletion of both Nup60 regions (aa200–309) completely abolished the interaction with the Y-complex (Fig. 2c). This indicates that the region spanning aa200–309 of Nup60 is necessary for direct interactions with the Y-complex. We then asked whether this region of Nup60 is sufficient for binding to the Y-complex. To address this, we generated a stable recombinant GST-Nup60 construct (aa200–318) and performed binding assays with the Y-complex (Fig. 2d,e and Extended Data Fig. 2b). The Y-complex specifically interacted with this minimal Nup60 region, whereas further N- or C-terminal truncations of the Nup60 construct resulted in a loss of interaction (Fig. 2e). Thus, a region spanning aa200–318 of Nup60 is both necessary and sufficient for a direct interaction with the Y-complex.

Fig. 2: Nup60 interacts with the Nup85•Seh1 arm of the Y-complex.

a, Top: the motif organization of S. cerevisiae Nup60. The motif organization of GST-Nup60*-StrepII constructs used for the in vitro binding assays is shown. Motif boundaries and constructs are drawn to scale with relevant amino acid positions indicated. N2BM, Nup2-binding motif. b, The cartoon representation of the Y-complex organization is shown. c, SDS–PAGE gel of in vitro binding assays performed with GSH bead-immobilized recombinant GST-Nup60* proteins (green circles) and the Y-complex (blue circles) affinity purified from S. cerevisiae (NUP133-TAP nup60Δ). The protein eluates were analysed by SDS–PAGE, Coomassie staining and immunoblotting against both GST (Nup60) and CBP (Nup133) tags. Bands were identified by MS (black asterisk: E. coli chaperone DnaK, which co-purifies with GST-Nup60* fusion proteins). For input, see Extended Data Fig. 2a. d, The motif organization of S. cerevisiae Nup60 and the GST-Nup60-StrepII constructs used for the in vitro binding assays is shown. e, SDS–PAGE of in vitro binding assays performed with GSH bead-immobilized recombinant GST-Nup60 proteins (green circles) and the Y-complex (blue circles) affinity purified from S. cerevisiae (NUP133-TAP nup60Δ). Protein eluates were analysed by SDS–PAGE, Coomassie staining and immunoblotting against both GST (Nup60) and CBP (Nup133) tags. For input, see Extended Data Fig. 2b. f, Inter-protein DSS cross-links identified by MS in the native eluate of GST-Nup60(200–318)•Y-complex. For DSS cross-linking, see Extended Data Fig. 2d. The cross-link between aa141 of Nup145C and aa198 of Seh1 is consistent with a previous report43. Also, aa201 of Nup145C is close to aa198, explaining why it cross-links to Seh1. The cross-link between aa856 of Nup133 and aa676 of Nup84 has also been reported earlier43 and satisfies the DSS distance restraints (<30 Å between Cα atoms) when mapped onto the crystal structure66. g, SDS–PAGE gel of in vitro binding assays performed with GSH bead-immobilized recombinant GST-Nup60 proteins and recombinant Nup85(1–564)•His6-Seh1 heterodimer. The protein eluates were analysed by SDS–PAGE, Coomassie staining and immunoblotting against both GST (Nup60) and His6 (Seh1) tags. For input, see Extended Data Fig. 2c. Source numerical data and unprocessed gels are available in Source data.

Source data

To precisely determine the binding site of Nup60(200–318) on the heptameric, ~580-kDa Y-complex, we performed cross-linking coupled to mass spectrometry (XL–MS) experiments using disuccinimidyl suberate (DSS) as a lysine-specific cross-linker. Recombinant Nup60(200–318) cross-linked specifically to the Nup85 and Seh1 subunits of the Y-complex, which constitute one of its short arms (Fig. 2f and Extended Data Fig. 2d). Guided by these results, we tested whether a recombinant Nup85(1–564)•Seh1 heterodimer (termed Nup85*•Seh1), purified via an N-terminal His6 tag on Seh1 (ref. 54), directly interacts with Nup60* (Fig. 2g and Extended Data Fig. 2c). Indeed, these experiments confirmed that the region between aa200 and aa318 of Nup60 is sufficient to bind to Nup85*•Seh1. Both the regions spanning aa200–239 and aa240–318 play a role in this interaction, with the latter displaying a higher affinity for Nup85*•Seh1. In summary, we have unambiguously identified Nup85•Seh1 as a direct interaction partner of Nup60.

Mlp1 forms a homodimer of coiled coils and flexible hinges

Based on these findings, our objective was to clarify the manner in which the structurally elusive basket filaments are linked to Nup60, and consequently, the Y-complex. Despite decades of effort, the overall topology of the basket filaments remains unknown due to the difficulties of studying these large coiled-coil proteins in isolation or within the NPC context. Earlier EM studies of two separate N-terminal fragments of the human Tpr (aa1–398 and aa774–1370) had reported rectilinear particles55. However, the manner in which these fragments are connected remained unclear.

We engineered Mlp1 to carry an N-terminal maltose-binding protein (MBP) tag (40 kDa), large enough to be visualized by EM. We succeeded in purifying a stable S. cerevisiae Mlp1(17–1,137) construct, expressed in insect cells. This construct comprises ∼60% of the full-length protein and 77% of the structured parts without the disordered C-terminus (Fig. 3a). After MBP-affinity purification on amylose beads and native elution with maltose-containing buffer, the protein exhibited a predicted size of ~174 kDa without major impurities (Fig. 3b). This protein was examined by heavy metal rotary shadowing EM. We observed filamentous particles containing multiple rod-like segments (Fig. 3c; for an overview, see Extended Data Fig. 3a). Notably, we consistently observed two globular knobs at one end of the particles, which we interpret as MBP tags (Fig. 3c), because the knobs were not present in a construct lacking MBP (Extended Data Fig. 3b). This provides direct experimental evidence that Mlp1 forms a parallel homodimer, probably consisting of rigid coiled-coil segments interrupted by flexible hinges. The number of visually distinct segments (defined as a particle segment, which can be traced by a straight line) varied from one up to six; however, most particles contained either four or three segments (Fig. 3d). The end-to-end length (defined as the sum of all segment lengths) of the four-segment particles ranged from 130–170 nm with a median length of ~150 nm (Fig. 3e). We then measured the lengths of individual segments in the largest subpopulation of particles, which contained four distinct segments (Fig. 3f). Counting from the N-terminal knobs of the MBP tag, the first, third and fourth segments displayed similar length distributions with medians between 37 and 43 nm. The second segment was noticeably shorter with a median length of only ~26 nm.

Fig. 3: Mlp1 forms a homodimer with coiled coils and flexible hinges.

a, Domain organization of S. cerevisiae Mlp1 based on previously published reports and AlphaFold-Multimer predictions. Domain boundaries are drawn to scale with relevant amino acid positions indicated. CC, coiled-coil region (dark magenta) and N60BD (bar above the cartoon); CTE, C-terminal extension. For AlphaFold-Multimer predictions of Mlp1, see Extended Data Fig. 4a. b, SDS–PAGE gel of recombinant MBP-3C-Mlp1(17–1,137)-His6 produced in Hi5 insect cells. Protein eluate was analysed by SDS–PAGE and Coomassie staining (black asterisk: proteolytically cleaved MBP tag; band was identified by MS). c, Representative rotary shadowing EM images of MBP-3C-Mlp1(17–1,137)-His6 particles (white arrowheads: N-terminal MBP tags; yellow lines in insets, particle traces used for segment measurements). Scale bar, 40 nm. For a more accurate representation of particle heterogeneity, see a gallery of images in Extended Data Fig. 3a. d, Histogram of MBP-3C-Mlp1(17–1,137)-His6 particles based on the number of visually distinct segments (n = 65). e, Histogram of end-to-end lengths (nm) of four-segment MBP-3C-Mlp1(17–1,137)-His6 particles (n = 31). f, Quantification of individual segment lengths (nm) of MBP-3C-Mlp1(17–1,137)-His6 particles (boxplots, n = 31). Segments are counted from the N-terminus (first segment is closest to MBP tags). Data presented in the boxplots as median and interquartile range (IQR) with whiskers extending to points that lie within 1.5 IQRs of the lower and upper quartile. Source numerical data and unprocessed gels are available in Source data.

Source data

The AlphaFold-Multimer prediction of Mlp1(17–1,137) features three extended coiled coils, each with a length of ~34–38 nm (Extended Data Fig. 4a), which closely aligns with the length of the first, third and fourth segments observed through EM. The second segment of Mlp1 is predicted to have a shorter coiled-coil structure and may correspond to the Mlp1 Nup60-binding domain (N60BD) (as explained below). AlphaFold-Multimer predictions also suggest that the overall architecture is shared between yeast Mlp1 and human Tpr (Extended Data Fig. 4a,b). In sum, we provide experimental evidence that Mlp1(17–1,137) forms a coiled-coil parallel homodimer with flexible hinge regions at distinct positions. Flexible Mlp1 hinges therefore are key features of the basket filaments, contrary to models that portray them as entirely rigid56.

Nup60 couples Mlp1 recruitment and docking on the Y-complex

The Nup60 aa200–318 region that we identified as an interaction site with both Mlp1 and the Y-complex consists of three adjacent SLiMs (termed SLiMs A, B and C). The proximity of three SLiMs within a short stretch suggested that Nup60 organizes, and potentially co-regulates, the recruitment of Mlp1 filaments to the Y-complex. To disentangle this complexity, we generated a set of GST-Nup60(200–318) constructs with different SLiM deletions (Fig. 4a), immobilized them on GSH beads and assessed their affinity to either recombinant Nup85*•Seh1 or Mlp1 by in vitro binding assays. Deleting SLiM A reduced the interaction with Nup85*•Seh1 (Fig. 4b, upper panel; compare lane 2 with 4), and correspondingly, SLiM A alone exhibited some affinity for Nup85*•Seh1 (compare lane 2 with 3). An additional deletion of SLiM C, retaining SLiM B, completely abolished the interaction with Nup85*•Seh1 (compare lane 4 with 5). Conversely, deletion of SLiM B did not affect Nup85*•Seh1 binding (compare lane 4 with 6). The linker connecting SLiMs A and B also displayed no detectable affinity for Nup85*•Seh1 (compare lane 4 with 8). Importantly, SLiM C was sufficient to robustly interact with Nup85*·Seh1 (compare lane 4 with 10).

Fig. 4: Nup60 couples Mlp1 filament recruitment and Mlp1 docking onto the Y-complex.

a, Motif organization of S. cerevisiae GST-Nup60-StrepII constructs used for the in vitro binding assays. b, Top: SDS–PAGE gel of in vitro binding assays performed with GSH bead-immobilized recombinant GST-Nup60 proteins (green circles) and Nup85(1–564)•His6-Seh1 heterodimer (blue circles). Bottom: SDS–PAGE gel of in vitro binding assays performed with GSH bead-immobilized recombinant GST-Nup60 proteins (green circles) and NusA-His6-Mlp1(382–620) (magenta circle). Protein eluates were analysed by SDS–PAGE and Coomassie staining (black asterisk: E. coli chaperone DnaK, which co-purifies with GST-Nup60 fusion proteins). c, A cartoon summary of in vitro binding assays with higher affinity interactions depicted in solid lines (Nup60 SLiM B-Mlp1 N60BD and Nup60 SLiM C-Nup85*•Seh1) and lower affinity interactions depicted in dashed lines (Nup60 SLiM A-Mlp1 N60BD and Nup60 SLiM A-Nup85*•Seh1). d, DSS cross-links identified by MS mapped onto AlphaFold-Multimer model of Nup60(260–318)•Nup85•Seh1. The segment of the Nup60 cartoon corresponding to the GST-Nup60(285–318) construct used in the DSS cross-linking experiment is coloured in green, whereas the rest is shown in grey. The identified cross-links are depicted in dashed lines connecting Cα atoms of cross-linked amino acid residues (orange indicates satisfied distance restraint and grey indicates violated distance restraint). For clarity, the unstructured N-terminal extension of Nup85(1–46) and the internal loop of Seh1(248–291) are not shown. For cross-linking, see Extended Data Fig. 5a,b. An AlphaFold3 model of Nup60(260–318)•Nup85•Seh1 recapitulated the positioning of polypeptide chains predicted by AlphaFold-Multimer. e, Zoom-in of satisfied cross-links as in d with corresponding amino acid residues indicated. f, The DSS cross-links identified by MS mapped onto AlphaFold-Multimer model of Nup60(260–318)•Mlp1(390-620) dimer. The segment of Nup60 cartoon corresponding to GST-Nup60(267–284) construct used in the DSS cross-linking experiment is coloured in green, whereas the rest is shown in grey. Identified cross-links are depicted in dashed lines connecting Cα atoms of cross-linked amino acid residues (orange indicates satisfied distance restraint and grey indicates violated distance restraint). For cross-linking, see Extended Data Fig. 5a,c. An AlphaFold3 model of Nup60(260–318)•Mlp1(390–620) dimer recapitulated the positioning of polypeptide chains predicted by AlphaFold-Multimer. g, Zoom-in of satisfied cross-links as in f with corresponding amino acid residues indicated. Source numerical data and unprocessed gels are available in Source data.

Source data

We then tested the same set of GST-Nup60 constructs for their capacity to bind a recombinant fragment of Mlp1. We used the previously identified N60BD of Mlp1 (aa382–620, Mlp1 N60BD)40, which was N-terminally tagged with NusA-His6 for improved protein stability and expression. The GST-Nup60(200–318) construct exhibited robust affinity for Mlp1 N60BD (Fig. 4b, lane 2 of the lower panel). Deleting SLiM A from Nup60 largely retained the affinity for Mlp1 N60BD (compare lane 2 with 4). Correspondingly, SLiM A alone exhibited only a weak affinity for Mlp1 N60BD (compare lane 2 with 3). An additional deletion of SLiM C had no major effect on Mlp1 N60BD binding (compare lane 4 with 5). Consistently, SLiM C alone exhibited only a minor affinity for Mlp1 N60BD (lane 10). Further dissection of SLiMs B and C showed that deletion of SLiM B had a strong effect on Mlp1 N60BD binding (compare lane 4 with 6). Importantly, when testing the affinities of SLiMs A, B and C for Mlp1 N60BD in isolation, SLiM B displayed the highest affinity for Mlp1 N60BD (compare lanes 3, 9 and 10). This establishes the 18-aa-long Nup60 SLiM B as a critical contact point for the Mlp1 filaments.

Comparing the affinities of the three Nup60 SLiMs for Nup85*•Seh1 and Mlp1 N60BD, SLiM B binds exclusively to Mlp1 N60BD, SLiM C binds primarily to Nup85*•Seh1 and SLiM A has the capacity to interact with both Nup85*•Seh1 and Mlp1 N60BD under the conditions tested (Fig. 4c). These data define the detailed molecular architecture of the junction connecting the NPC basket and the NPC core. The interaction pattern of this tripartite junction suggests that SLiM A may sense the presence of basket filaments and Y-complex simultaneously, acting as a ‘coincidence detector’.

XL–MS confirms Nup60 interactions with Nup85 and Mlp1

For a refined understanding of the NPC basket-core connection, we modelled Nup60 interactions with either Nup85•Seh1 or Mlp1 by AlphaFold-Multimer. The Nup60 region spanning SLiMs B and C (aa260–318) is predicted to form two α-helices connected by a linker, with each SLiM containing a short α-helix (Fig. 4d). The α-helix of Nup60 SLiM C wedges between the ‘crown’ and ‘trunk’ of the Nup85 α-solenoid fold54 (Fig. 4d). In contrast, no contacts were predicted between SLiM B and Nup85 or Seh1, consistent with our in vitro binding assays (Fig. 4b). AlphaFold-Multimer also did not predict an interaction of Nup60 SLiM A with Nup85•Seh1, possibly because their affinity is too low (Fig. 4b, lane 3). To test the validity of the model, we performed XL–MS experiments with the reconstituted Nup60(285–318)•Nup85*•Seh1 complex and mapped the DSS cross-links onto the predicted structure (Fig. 4d,e and Extended Data Figs. 5a,b and 6a–c). Supporting the structural model, the K307 residue of the Nup60 SLiM C α-helix cross-linked specifically to the Nup85 K163 residue, positioned in a groove between the crown and trunk of Nup85. The measured distance between these cross-linked lysine residues satisfies the maximal distance restraints imposed by the DSS cross-linker (<30 Å between Cα atoms). A second cross-link was detected between the Nup60 K286 residue and Nup85 K238, providing additional support for the predicted relative orientation of SLiM C with respect to Nup85 (Fig. 4d). Notably, despite the large number of surface-exposed lysine residues in Nup85*•Seh1 (29 in Nup85* and 22 in Seh1), only Nup85 K163 and K238 were cross-linked to Nup60 SLiM C, suggesting that the intrinsically disordered Nup60 protein binds to the Y-complex in a defined orientation.

We then modelled Nup60 interactions with Mlp1 following the same approach. The Mlp1 N60BD construct (aa390–620) was predicted to form two dimeric coiled-coil segments, arranged in a V-shape with an acute angle of ~45° (Fig. 4f). The apex of the V-shaped Mlp1 structure, which connects and orients the two coiled-coil segments, is decorated by short α-helices, located on opposite sides of the apex. Nup60(260–318) forms two α-helices connected by a linker similar to the Nup60•Nup85•Seh1 model (Fig. 4d,f). In contrast, SLiM B is predicted to wedge into a groove on the C-terminal coiled-coil segment of Mlp1 N60BD close to the hinge region, whereas SLiM C remains unbound (Fig. 4f). Interestingly, the Mlp2 N60BD is predicted to interact with SLiM B of Nup60 in a similar fashion (Extended Data Fig. 6h). Noting that a segment of human Tpr is predicted to adopt a structure almost identical to Mlp1 N60BD (Extended Data Fig. 4), we asked whether AlphaFold-Multimer predicts a similar interaction mode between human Nup153 and Tpr, which have been shown to interact directly in the past57. We ran an unbiased prediction using full-length Nup153 and a Tpr(390-620) homodimer. Interestingly, AlphaFold-Multimer predicted that Nup153(305–320) forms an α-helix that interacts with the Tpr(390–620) homodimer in a fashion identical to the Nup60 SLiM B interaction with the Mlp1 N60BD (for a superposition of the predicted structures, see Extended Data Fig. 6g). The predicted interactions are consistent with previously performed in vitro binding assays demonstrating a direct interaction between recombinant Nup153(228–611) and Tpr(172–651) (ref. 57). In a yeast two-hybrid assay, Tpr(172–651) interacted with Nup153(228-439), but neither interacted with Nup153(1–244) nor Nup153(337–611) (ref. 57), which is in line with our structural modelling. Hence, despite low overall sequence conservation between yeast Nup60 and human Nup153, as well as between yeast Mlp1 and human Tpr, the predicted interactions are well conserved.

To further test the validity of the model, we performed XL–MS experiments with the reconstituted Nup60(267–284)•Mlp1 N60BD complex and subsequently mapped the identified DSS cross-links onto the structure (Fig. 4f,g and Extended Data Figs. 5a,c and 6d–f). The majority of identified inter-protein cross-links (14 out of 15) mapped onto the C-terminal coiled-coil segment of Mlp1 N60BD (Fig. 4g) even though both coiled-coil segments of the V-shaped Mlp1 model contain a large number of surface-exposed lysine residues (26 in the C-terminal and 12 in the N-terminal segment). The 15-residue stretch (aa559–573) within the Mlp1 N60BD forms a cross-linking hotspot, where Mlp1 K559, K563, K566, K568 and K573 are identified as being cross-linked to both K267 and K268 of the Nup60 SLiM B (Fig. 4g). In each of these instances, at least one chain of the Mlp1 N60BD dimer satisfies the DSS distance restraints. Moreover, the cross-link between K267 of Nup60 and K545 of Mlp1 N60BD is satisfied by one of the Mlp1 chains. The cross-link between K267 of Nup60 and K537 of Mlp1 N60BD, although violated by distance, can be explained by the flexibility of the SLiM B N-terminus. Only two cross-links, involving K267 and K268 of Nup60 and K590 of Mlp1 N60BD, remain unexplained. Thus, the AI-predicted interaction patterns are in excellent agreement with our reconstitution and XL–MS experiments and confirm a division of labour between SLiM B (Mlp1 contact) and SLiM C (Nup85 contact).

Mutational analyses verify the interaction surfaces

To verify the critical interfaces of the Nup60•Nup85•Mlp1 junction, we introduced rationally designed point mutations. The interaction between Nup60 SLiM C and Nup85 is predicted to involve a salt bridge between Nup60 R306 and Nup85 D216 or the adjacent E218 (note that out of the five high-confidence models generated by Alphafold-Multimer, residue R306 can accommodate two alternative salt bridges) (Fig. 5a). A second salt bridge is formed between Nup60 K309 and Nup85 E177 or E176. Hydrogen bonding between Nup60 N296 and the Nup85 E218-P219 peptide group is expected to further stabilize the interface (Fig. 5a). Notably, the NPY triad (aa296–298) of Nup60 is conserved across yeast species (Extended Data Fig. 7a). Hence, we generated mutations targeting these interactions in Nup60 and Nup85 (alanine substitutions or charge inversions) and tested the ability of recombinant GST-Nup60(285–318), spanning SLiM C, to interact with Nup85*•Seh1. The in vitro binding assays showed that alanine substitutions of Nup60 N296 and Y298 reduced the affinity for Nup85*•Seh1, while mutating P297 to alanine had a minor effect (Fig. 5b and Extended Data Fig. 7b). A double mutant, in which both Nup60 R306 and K309 were mutated to E, also had a disruptive effect (Fig. 5b). Mutating the relevant Nup85 residues also disrupted the Nup60 interaction, although to a lesser extent (Fig. 5b).

Fig. 5: Mutational analyses verify the interaction surfaces.

a, Cartoon representation of the interface between Nup60 SLiM C (green) and Nup85 (blue) as predicted by AlphaFold-Multimer model of Nup60(260–318)•Nup85•Seh1. b, SDS–PAGE gel of in vitro binding assays performed with GSH bead-immobilized recombinant GST-Nup60(285–318) mutants and wild-type Nup85(1–564)•His6-Seh1 heterodimer (left), and wild-type GST-Nup60(285–318) and Nup85(1–564)•His6-Seh1 mutants (right). Protein eluates were analysed by SDS–PAGE and Coomassie staining. For inputs, see Extended Data Fig. 7b. c, A cartoon representation of the interface between Nup60 SLiM B (green) and Mlp1 dimer (magenta) as predicted by AlphaFold-Multimer model of Nup60(260–318)•Mlp1(390–620) dimer. d, SDS–PAGE gel of in vitro binding assays performed with GSH bead-immobilized recombinant GST-Nup60(267–284) mutants and wild-type NusA-His6-Mlp1(382–620) (left), and wild-type GST-Nup60(267–284) and NusA-His6-Mlp1(382–620) mutants (right). The protein eluates were analysed by SDS–PAGE and Coomassie staining. For inputs, see Extended Data Fig. 7c. e, The AlphaFold-Multimer model of the Nup85(101–430)•Nup60(260–318)•Mlp1(450–580) dimer complex at the interface between the NPC core and nuclear basket. For cartoon representation, model confidence and surface conservation, see Extended Data Fig. 8a–c. An AlphaFold3 model of Nup85(101–430)•Nup60(260–318)•dimeric Mlp1(450–580) recapitulated the positioning of polypeptide chains predicted by AlphaFold-Multimer. f, SDS–PAGE gel of Y-complex•Nup60(200–318)•Mlp1(382–620) assembly reconstitution. Anti-FLAG beads were coated with VHH-SAN8-SNAP-FLAG (VHH-SAN8*) nanobody specific for the Nup84 subunit of the Y-complex59. The Y-complex affinity purified from S. cerevisiae (NUP133-SNAP-TAP nup60Δ) was immobilized on VHH-SAN8* beads and co-incubated with recombinant Nup60(200–318)-StrepII and NusA-His6-Mlp1(382–620) proteins. To generate Nup60(200–318)-StrepII, purified GST-TEV-Nup60(200–318)-StrepII protein was incubated with TEV protease and the mix was added to the reconstitution mixture. Top: protein eluates were analysed by SDS–PAGE and Coomassie staining. Bottom: the identity of the Nup60(200–318)-StrepII band (green circle) in the Coomassie-stained SDS–PAGE gel was verified by immunoblotting against the C-terminal StrepII tag. It was also verified by MS analysis of the excised band. Unprocessed gels and blot are available in Source data.

Source data

For the Nup60 SLiM B interface with Mlp1, AlphaFold predicted a salt bridge between Nup60 D281 and Mlp1 R532, and a network of hydrogen bonds involving Nup60 S270 and N271, and Mlp1 D524, E561 and E564 (Fig. 5c). To assess the relevance of these residues, we examined how mutations impacted the binding strength between recombinant GST-Nup60(267–284), spanning SLiM B, and Mlp1 N60BD. Mutating Nup60 D281 to K reduced the affinity for Mlp1, whereas mutating Nup60 S270 to D abolished binding to Mlp1 (Fig. 5d and Extended Data Fig. 7c). A double mutation, in which both S270 and N271 were mutated to A, had a similarly strong effect (Fig. 5d). Conversely, mutating Mlp1 D524 to A, Mlp1 R532 to E or mutating both E561 and E564 to A strongly reduced the affinity between Mlp1 N60BD and the Nup60(267–284) (Fig. 5d). In summary, our integrative approach, combining in vitro binding assays (Fig. 4b), XL–MS (Fig. 4d–g) and point mutagenesis (Fig. 5a–d), provides robust support for the validity of the AlphaFold models.

Reconstitution of the basket–outer ring junction in vitro

Having identified the Nup60•Mlp1 and Nup60•Nup85 binary interfaces, we set out to obtain a structural model of the entire Nup85•Nup60•Mlp1 junction. Because large multi-protein structures can pose a challenge for AlphaFold, it became necessary to truncate the proteins further. We used the crown-trunk junction (aa101–430) of Nup85 (ref. 54), two copies of Mlp1 covering the hinge region and shortened coiled-coil segments (aa450–580), and Nup60(260–318), spanning SLiMs B and C, as before. The predicted model of the composite ternary complex recapitulates the main features of the binary interactions, yet offers additional insights (Fig. 5e). Nup60 now simultaneously interacts with Mlp1 and Nup85 via SLiM B and SLiM C, respectively. In doing so, it stitches the Y-complex and basket filaments together. The hinge region of Mlp1 lies directly opposite to Nup85, yet, the two proteins exhibit no noticeable contacts, suggesting that the Nup85•Nup60•Mlp1 junction could be intrinsically flexible. These predicted features align with a host of previous observations suggesting that the Mlp1/Tpr filaments can exhibit extensive movements on the NPC core58. Of note, the predicted interfaces of the tripartite junction are evolutionarily conserved across yeast species, highlighting their structural relevance: the surface patch of Mlp1, which interacts with Nup60 SLiM B, displays a high degree of conservation across yeast species (Extended Data Fig. 8a–c). Similarly, multiple residues in the groove of Nup85, which accommodates the Nup60 SLiM C, are conserved (Extended Data Fig. 8c).

To reconstitute a supramolecular assembly comprising the Nup60 suspension cable, Y-complex and part of the Mlp1 filaments, we devised a step-wise approach. First, the affinity-purified Y-complex was incubated with recombinant Nup60(200–318) spanning SLiMs A, B and C (generated by TEV protease cleavage of GST-TEV-Nup60(200–318)). The Y-complex was captured on nanobody-coated anti-FLAG beads using a FLAG-tagged nanobody (VHH-SAN8-SNAP-FLAG), which binds Nup84 (ref. 59). Next, NusA-His6-Mlp1 N60BD was added, and after washing, the interacting proteins were eluted using a FLAG peptide. As expected, Nup60(200–318) alone bound to the Y-complex (Figs.