The molecular mechanism of nuclear transport revealed by atomic-scale measurements

1Nuclear pore complexes (NPCs) are the sole mediators of bi-directional nucleocytoplasmic trafficking. Transport is rapid and reversible, with the entire process of transport factor (TF) docking, passage and release across the NPC taking only a few milliseconds. The NPC consists of a ∼30-nm diameter central channel filled with phenylalanyl-glycyl-repeat-rich nucleoporins (FG Nups) which provide the selective filter. Depletion or deletion of the FG Nups results in leaky, non-selective barriers. Moreover, the FG domains in isolation facilitate selective passage of TFs through nanopores or accumulation in hydrogels.

2It is generally accepted that TF interaction with FG repeats reduce the diffusional barrier to enable selective transport, though the molecular mechanism of TF passage through the NPC remains largely unknown. TFs may alter the properties of the FG permeability barrier. Mesoscale observations of TF-FG repeat interactions in vitro have shown that TFs can change the height of FG brushes on planar surfaces, modulate the transport of inert cargo, assemble with FG Nups into large assemblies, and inhibit amyloid hydrogel formation observed for some FG Nups. Whether these behaviors arise from changes in FG structure or as a result of the multivalent FG-TF interaction remains undetermined, as do their contributions to in vivo nuclear transport.

3Several mesoscale models have attempted to explain how FG Nups prevent the passage of most macromolecules while allowing selective transport of TFs alone and with their cargo. The FG Nups have been proposed to form a selective barrier due to their reversible inter-chain cohesion (hydrogel and bundle models), entropic exclusion (virtual gating model), collapse upon TF binding (reduction in dimensionality model), or a combination thereof (forest model). These models differ in their predictions of FG behavior on their own (ranging from highly mobile to fully self-associated) and upon binding of TFs. For example, the hydrogel, forest and reduction in dimensionality models invoke large changes in FG Nup behavior upon TF interaction. Crucially, none of these models or current observations fully explains how transport can be both selective and rapid, as is seen in vivo.

4These conflicting models of the mechanism of NPC selectivity remain because of a lack of atomic-scale experimental data describing FG Nup behaviors and interactions. Therefore, we used nuclear magnetic resonance (NMR) techniques to provide a rich readout of FG Nup behaviors at atomic detail (Figure 1A). We defined a minimal system including FG attachment, FG repeat type, and TF interaction, the essential features necessary to recapitulate selective transport in vitro. We measured the physical state of FG repeats with and without TFs bound. Because environment strongly affects intrinsically disordered proteins (IDPs), including FG Nups, we mimicked the normal environment of NPCs using Xenopus egg extract, the best characterized environment for in vitro nuclear transport measurements. In addition, we tested the following: cytoplasm of living Escherichia coli using in cell NMR; E. coli high speed lysate; and buffer alone, the latter lacking crowding agents or competitors and being the milieu in which these proteins have been most studied previously.

5We studied Nsp1, the most tested and characterized FG Nup which has been shown in vitro to mimic transport faithfully. We focused on two segments of Nsp1 for which there is a consensus that they prototypically represent the extreme flavors and behaviors of FG Nups; (i) the N-terminal segment of low charge (Asn-rich) and irregularly spaced FG repeats observed to be highly cohesive and form amyloid hydrogels under certain conditions (FG-N) and (ii) the central segment of significant charge (Lys-rich) with regular FSFG repeats typically observed to be highly soluble (FSFG-K). The isolated N-terminal domain forms selective hydrogels in vitro and is able to replace Nup98-FG domain in reconstituted Xenopus nuclei to provide the primary barrier enabling selective transport. We studied these fragments individually and in combination in a wide range of constructs and conditions, including where the FG repeats are tethered to mimic their arrangement in the NPC (Table 1, Figure 1—figure supplement 1).

6FG Nups were fully disordered and highly dynamic in all cellular milieu tested. In NMR measurements, the degree of secondary structure is correlated with the difference between the 13C chemical shift values of the α and β carbons relative to random coil values for each residue. The degree of secondary structure quantified in this way is near zero for all residues in FG-N and FSFG-K constructs in all conditions tested (Figure 1B, Figure 1—figure supplement 1). This behavior of intrinsic disorder, judged from chemical shifts, is strikingly robust, being seen under a wide variety of conditions and also (for FSFG-K constructs) in buffers of varying pH (Figure 1—figure supplement 3).

7Our observations in cellular milieux are in contrast to the behavior of the constructs containing the FG-N domain in buffer alone. Under these conditions, dynamic light scattering and NMR measurements indicate that FG-N forms a hydrogel-like material (Figure 1—figure supplement 4). The in buffer behavior of our construct is consistent with extensive previous observations of β-sheet formation and aggregation of the N-terminal domain of Nsp1 (FG-N). However, when FG-N constructs are observed by NMR inside a living cell, or in the presence of cell lysates or mimics, there is no appearance of high molecular weight components—NMR spectra of FG-N constructs in living E. coli did not change with time over more than 24 hr (Figure 1—figure supplement 5). These data indicate that the cellular milieu is a strong inhibitor of intermolecular FG repeat aggregation. The state of FG repeats in our system is therefore highly dynamic, as has been reported in vivo.

8To understand the differences between protein structure and dynamics in cellular mimics as compared to buffer, we measured spin relaxation parameters (R1, R2, nOe) which quantify the motion and interactions of the residues on timescales of μs to ms (R1, R2,) and ps to ns (nOe). NMR relaxation properties of FG Nups are significantly different between buffer alone and a protein-rich environment (Figure 1C). We measured large increases in R2/R1, indicative of transient spectral changes on interaction between the constructs and the cellular milieu (Figure 1C, Figure 1—figure supplement 2). The increases in R2/R1 were not seen upon increasing viscosity or crowding by inert agents (Figure 1—figure supplement 6), indicating that these changes result from weak binding of the FG repeats to the proteinaceous milieu as suggested by previous studies. The phenylalanines and their adjacent residues show the primary interactions, with the greatest increases in R2/R1, while the spacer sequences remained relatively unaffected. The nuclear Overhauser effect (nOe) data show that the FG-N are highly mobile and flexible (Figure 1C, Figure 1—figure supplement 2), indicating no sequence-specific compaction, folding, or molten globule formation. The cellular milieu is thus in a state of constant, non-specific, dynamic interaction with the FG repeats.

9Our results suggest that interactions of FG repeats with the cellular milieu stabilize the unfolded state by engaging the hydrophobic phenylalanines in transient interactions that decrease their contact with the water, reducing the driving force for hydrophobic collapse and amyloid formation. This is consistent with previous work that demonstrated that mutation of the Fs to more hydrophilic residues inhibits aggregation. Thus, exchange with the cellular milieu maintains the FG Nups in a highly dynamic, disordered state by inhibiting the intramolecular interactions that lead to aggregation in buffer.

10We used the atomic-level readout of our system to interrogate key aspects of FG Nup-TF interactions that have previously been inaccessible. It has long been established using crystallographic, computational, and NMR approaches that FG residues bind to hydrophobic pockets on TFs (reviewed in ). However, the dynamics of this interaction and the behavior of the spacer regions between FG residues upon TF binding have remained uninterrogated, leading to proposals ranging from remaining disordered to significant structural rearrangements. Our assays provide a direct readout of the dynamics of the interaction and the state of the linker regions upon TF binding (Figure 2).

11Many previous in vitro observations of FG-TF interactions lacked demonstration of reversibility and showed strong affinities, incompatible with the rapid transport rates observed in vivo. By monitoring the recovery of the NMR spectral characteristics on dilution of Kap95, the interacting TF, we showed that the FG repeat-TF interactions are fully reversible in our system (Figure 2—figure supplement 1).

12We used measured changes in amide resonances peak position and intensity as a function of Kap95 concentration using 2D Heteronuclear Single Quantum Coherence (HSQC) experiments (Figure 2A). Each (non-proline) residue contributes a peak, with the amide proton resonance frequency on the x-axis and nitrogen resonance frequency on the y-axis. The signal intensity of each peak is sensitive to the mobility and environment of that residue; upon binding to a TF, the effects of slower motion and of differing environment result in the signal of the corresponding peak decreasing. Strikingly, although the phenylalanine repeats themselves are attenuated upon Kap95 interaction, the majority of peaks are minimally affected, showing that the corresponding residues remain disordered and dynamic (Figure 2A). The spacer regions are thus highly mobile even within the interacting state. An extreme of this observation is seen for K332, which shows no attenuation (left subpanel in Figure 2A, lowest trace Figure 2C), and thus this residue remains flexible both in free and when bound to TF. When we examined attenuation as a function of residue number, we observed a striking periodic pattern of minimal attenuation for most residues and significant attenuation for FG repeats (Figure 2B). The degree of signal attenuation falls off relatively uniformly at both sides of each FG repeat, consistent with the transient binding of individual FG residues, with the remaining residues remaining fully disordered and highly dynamic. The minimally attenuated residues include those that have previously been implicated in forming phase changes of the FG repeat regions. We did not observe any major state change of spacers resulting from TF addition, such as caused by the formation or breakage of secondary structure, or gel/sol transitions.

13We observe that the binding interface between FG Nups and TFs at each interacting site is small—just the 2–4 residues surrounding the FG repeat itself—and the residues that do not interact directly with the TF sites remain highly mobile and dynamic. The size of the binding interface is consistent with other known transient interactions. We estimated a minimum effective affinity of Kap95 for the FSFG-K from a titration observing 15N R2 (Figure 2—figure supplement 2). The true Kd is greater than 36 μM, fully consistent with rapid, reversible transport.

14Interaction or binding while remaining primarily disordered is highly unusual, as many functional IDPs are believed to adopt significant secondary structure formation upon interaction, which we do not see here. This behavior is reminiscent of that proposed for random fuzzy complexes, in which one partner remains dynamically disordered and transient interactions predominate during selective recognition.

15Many observations suggest that there are multiple transport routes across the NPC, with FG Nups arranged in a variety of geometries. Thus, because the NPC contains a large number of FG Nups in different arrangements to each other, we investigated the common features of the effects of local packing and attachment on FG repeats (Figure 3, Table 1). We designed chimeric proteins containing the FG domains in homotypic and heterotypic combinations. Some constructs included the tetramerization domain of p53, allowing us to form complexes that mimic the attachment of the FG domains to the NPC. We found essentially no change in chemical shifts and only modest changes in linewidth upon attachment to the p53 tetramerization domain (Figure 3). Any ordered protein of this size (∼120 kDa) would be invisible using our NMR experiments, demonstrating that FG Nups remain dynamic IDPs even in large complexes, as previously measured for other large IDPs. Notably, our largest constructs have a diameter measured by dynamic light scattering of 13 nm (Figure 3B), a significant fraction of the 30-nm diameter of the NPC. Our results reveal that the behavior of a given FG repeat is not affected strongly by the number or type of other FG repeats surrounding it (i.e., no ‘emergent properties’). Our results also strongly indicate that there are no significant FG–FG interactions, as interactions between different types of FGs should impact their average environment. However, we cannot completely exclude that extremely weak, dynamic interactions can occur between FG repeats, and indeed such interactions may modulate the long distance distribution of FG repeat regions in the NPC. The FG Nups thus remain a highly dynamic fluid even at sizes, arrangements, and packing densities commensurate with the NPC. We observed the same pattern of rapid exchange of the FG residues with TF binding sites while the spacer regions remain highly mobile in all environments studied and with all constructs tested, showing the robustness of our findings (Figure 4).

16While all overall behaviors are similar between FG-N and FSFG-K segments, in monomers or in tetramers, there are some intriguing differences. For example, in constructs where both FG types are present (Figure 4), not all FG-N phenylalanine residues are completely attenuated. This may be evidence for specificity of certain FG repeat types to binding particular TFs.

17The FG repeats fill the central channel of the NPC to form a barrier to non-specific macromolecular diffusion. We show that, even when densely packed and in mixed flavors as in the NPC, the FG repeat regions studied here remain fully disordered and highly dynamic. The two domains studied represent the extreme behaviors observed for FG Nups in vitro: FSFG-K is highly soluble and non-cohesive, while FG-N in buffer is highly cohesive, a prototypical hydrogel-forming FG Nup. It is clearly possible to make FG repeats aggregate (Figure 1—figure supplements 4, 5), as has been demonstrated for many proteins. However, our data are inconsistent with any such hydrogel state in cellular milieu. Re-arrangements of internal non-covalent crosslinks within hydrogels are on slow (second to minute) timescales, long enough to form a solid gel resistant to deformation, and too slow for solution NMR analysis. Instead of such a highly internally interacting, slowly moving gel, our results demonstrate that FG repeat regions form a highly dynamic phase, consistent with the rapid rates of nuclear transport. A highly dynamic, fluid state for FG repeats in vivo is in agreement with measurements of the living NPC. FG Nups do not appear to form a molten globule or collapsed state, as has also been proposed, instead behaving as a fully disordered IDP. Our results also allow us to distinguish their IDP class. FG repeats do not appear to populate more folded states either inter- or intra-molecularly upon interaction with either TFs or other FG Nups. Instead, FG repeats lack any secondary structure, even when interacting with cognate binding partners, indicating that IDP mobility may be a key ingredient in nuclear transport.

18Our results highlight the importance of the cellular environment for determining the behavior of the FG Nups because multiple weak interactions with cellular components stabilize the FG Nups in a disordered state. We propose that the selectivity of the NPC is thus maintained by both non-specific and specific cellular interactions, consistent with recent work demonstrating that FG permeability barrier is modulated by TFs. We show that the modulation of permeability of FG brush behavior by TFs is not a result of structural changes of the FG Nups, but must result instead from the rapid, multivalent interactions of TFs with FG repeats.

19Our data, strongly supporting that FG repeats are highly dynamic tethered IDPs, imply that they must form entropic bristles (i.e., strongly sterically hindered regions) around their tether site (the central channel of the NPC). This hindrance increases with increasing size of the passing macromolecules; hence, small proteins can pass more easily than large ones. These larger nonspecific macromolecules interact with the FG repeats with insufficient frequency to overcome the entropic barrier of the FG repeats' polymer bristle structure, thus being effectively excluded. TFs, however, have sufficient interaction frequency to do so—the original tenet of the ‘virtual gating’ idea, now delineated at the atomic scale by these results. In summary, the FG repeat regions, crowded around and within the central channel, may set up an entropic barrier that excludes macromolecules from their vicinity while permitting the approach of small molecules; however, macromolecules that interact with the FG repeats (such as TFs) can overcome this barrier by their multiple and frequent interactions with FG Nups. We are pursuing experiments which we hope in the near future will gage the precise nature and magnitude of the exclusion mechanism.

20Our results suggest that FG disorder is critical to the speed of selective nucleocytoplasmic transport. Selectivity is determined by the free energy gain upon TF-NPC binding, with the off rate (and so speed of transport) constrained by koff = Kd·kon. For a given selectivity (Kd), the residence time within the NPC is constrained by the off rate. IDPs are able to engage far more rapidly than most ordered proteins, allowing for high on and off rates while maintaining selectivity. For FGs in particular, several key lines of evidence indicate that the interaction of FG Nups with TFs is extraordinarily rapid, that is, each FG motif is in extremely fast exchange with TFs. We have directly shown that the binding interface is only 2–4 amino acids (the FG repeat itself). The buried surface area of this interaction is <1000 Å2, consistent with other known transient interactions. The spacer regions remain highly mobile and the degree of signal attenuation falls off relatively uniformly at both sides of each FG repeat (Figure 2B), fully consistent with the transient binding of individual FG residues. Within the NPC, a TF is able to rapidly diffuse, engaging transiently with multiple FG motifs across different FG Nups. The association rate of TFs with individual FG motifs is very fast because the FGs move very rapidly and the TF has multiple binding pockets available for interaction. Taken together, these results indicate that the dynamics of FG-TF interaction are extraordinarily rapid. As a result, the interaction can be strong enough to be selective, and yet remain extremely fast, allowing rapid transit through the NPC.

21In essence, it is the quantity, rather than quality, of interactions possible per molecule with FG repeats that distinguishes a TF from other macromolecules (Figure 5). Transient non-specific interactions of the FG repeats with their milieu help maintain the disordered and dynamic nature of the barrier. Given the low sequence and structural complexity of the FG repeats, such non-specific interactions are to be expected but are not strong enough for these non-specific macromolecules to overcome the FG Nups' diffusion barrier. In contrast, TFs bind specifically to the FG motifs through pockets tuned for this purpose while leaving the spacers between the motifs highly mobile. The fact that the spacers—and the FG repeats as a whole—remain fully mobile even when interacting with a TF also means that the barrier to non-specific passage formed by a dense fluid of FG repeats in the NPC remains fully intact, regardless of even large TF fluxes (a significant issue with models that invoke state changes, which our results and model circumvent).

22While the FG repeats show little evidence of changed behavior upon oligomerization or TF interaction, it is likely that their structured attachment sites and diversity of flavors produce modulation of composition within the NPC. These variations are potentially important for the organization of alternate transport pathways. Our approach and its future extensions will thus be a powerful tool for the detailed characterization of these different nuclear transport pathways. The NPC is a biological example of the idiom ‘many hands make light work’; the combined effect of many weak, transient interactions provides a specific multifunctional transport system, which allows for the simultaneous passage of hundreds of different macromolecules with a wide range of sizes and for robustness to significant alterations in the cell.

23Proteins (Table 1) were expressed in BL21DE3 Gold cells using expression plasmids pET24a or pRSFDuet.

24Protein concentration was determined by BCA analysis (Pierce Co.), except for Kap95 determined by OD280 based on standard amino acid content. IPTG was from Gold Biotechnology. Bovine Serum Albumin (BSA), Pepstatin, PMSF, Protease Inhibitor Cocktail are from Sigma. Xenopus egg extract was the generous gift of Takashi Onikubo and David Shechter. Lyophilized coli extract was prepared by modification of the preciously described method using E. coli BL21 Gold cells grown in LB to OD600 of 0.8, resuspended in Buffer A + 18 mg/l PMSF +0.4 mg/l pepstatin A, lysed by passing through a Microfluidizer, and centrifuged at 112,000 g for 30’ at 4°C.

25The FG-N, and FSFG-K segments were derived as previously described with the following modifications. Expression plasmids (pET24a, pRSFDuet) containing FG Nup fragments were transformed into BL21DE3 Gold cells (Agilent). Cells containing pET24a or pRSF constructs were grown to OD600 of 0.8, induced with 1 mM IPTG and harvested after 2–4 hr. The cells' periplasm was removed by osmotic shock, and lysed under denaturing conditions (8 M urea). Urea was maintained throughout the purification for FG-N, whereas FSFG-K had 8 M urea in the lysis, 3 M urea in the first wash, and no urea in the remaining washes and elution. Proteins were purified on Talon resin in 20 mM HEPES, 150 mM KCl, 2 mM MgCl2, (Buffer A, at pH 7.0), with protease inhibitor cocktail (PIC), PMSF, and pepstatin. Kap95 was prepared as previously described [U-15N] and [U-13C, 15N] materials were prepared using M9 media containing 15NH4Cl and [U-13C] glucose (Cambridge Isotopes Limited, MA) as needed. The tetramerization domain of p53 was used to provide intramolecular crowding by construction of fusions with the tetramerization segment. As with all our constructs, the preparation was analyzed by size exclusion chromatography (SEC) to ensure that we were examining un-aggregated protein of the correct size and quantity (data not shown) and confirmed by dynamic light scattering.

26Experiments were conducted on Bruker spectrometers at 700 MHz and 298 K unless otherwise indicated. In cell NMR methods used the general procedures developed for STINT-NMR, and other IDP studies. Assignment of resonances used the standard triple resonance approach as in;;;;, including HNCO, HNCACO, HNCACM, CBCACINH, HNCA, HNCOCA, CCCONH, and HCCONH. Assignments were conducted in cell (E. coli)/buffer E and in buffer A, for the sets tabulated in Table 2, and were confirmed in other buffers by examination of HSQCs. Relaxation methods used standard procedures. No corrections were applied for possible changes of fast exchange with solvent water, as described for R2(CPMG) for the following reasons. The observed changes for the IDP α-synuclein are less than two fold, and sequence-independent for a change of pH from 7.4 to 6.2 (Figure 1; ) while for FG Nups we observe up to c. sixfold sequence dependent variation of R2. Secondly, the same sequence dependent variations of R2 are seen in buffered E. coli extracts, with controlled pH (Figure 1—figure supplement 6). ‘Titration’ experiments were conducted at 800 or 700 MHz in buffers indicated, by preparation of separate samples for each change of concentration of Kap95.

27Titration experiments for 15N R2 were conducted at 500 MHz in buffer A, by preparing separate samples. Kap95 was fixed at 20 μM and 15N FSFG-K concentration was varied from 12.5 μM to 1200 μM. Each sample was prepared by addition of standard volumes of stocks to obtain the concentrations. No additional normalization was used. Each R2 data acquisition used 6 delay points. For FSFG:Kap95 molar ratio of 1:1 and 0.625:1 the following delay times were used 0, 32.64, 65.28, 97.92, 130.56, 163.2 ms. For FSFG:Kap95 molar ratio of 4:1 the following delay times were used 0, 81.6, 163.2, 244.8, 326.4, 408 ms. For FSFG:Kap95 molar ratio of 7:1 the following delay times were used 0, 81.6, 163.2, 244.8, 326.4, 489.6 ms. For FSFG:Kap95 molar ratios of 60:1, 28.75:1 and 11:1 the following delay times were used 0, 81.6, 163.2, 326.4, 489.6, 652.8 ms. The same delays were used for an FSFG-N alone and the R2 values obtained provided the value of R2 free (R2(0)) in the binding equation. The numbers of scans were optimized for S/N for each sample and total acquisition time per sample was ∼<13 hr.

28The chemical shifts and their standard deviations were used in the standard equation to derive deviations from predicted shift values for the sequences of the assigned proteins. The values for these for Δδ13Cα- Δδ 13Cβ are shown in Figure 1—figure supplement 1.

29For NMR relaxation analysis, we derive approximate correlation times for ps/ns, associated with the different correlation times for hetero nuclear nOe by a direct calculation method that, in part, normalizes across different observation fields/frequencies, avoiding any complexities of spectral density analysis (for example, ), of the unverified applicability of a ‘Model Free’ approach, and of analysis of simulated molecular dynamics which may not be generally applicable to all IDPs.

30The apparent correlation times for motion of backbone amides was derived from use of the heteronuclear 15N[1H] nOe using the standard dipolar treatment, ignoring chemical shift anisotropy contributions. For this pair, the field independent form from the standard formulae (p 135 in ) is given by

31where fH is the 1H frequency (Hertz) and is the apparent correlation time for the ns motion associated with direct dipolar relaxation. We use here to discriminate from the overall correlation time for folded protein motion usually denoted as.

32may be directly obtained as

33The data of Figure 2 do not fit readily in most models of NMR titration for protein/protein interactions, especially for IDPs, where an intermediate folded form is frequent (for example,; ) In the case of slow exchange, it would be expected that at least part of the FG Nup sequence would show multiple peaks associated with some mobility in the ‘bound’ form. In the case of intermediate exchange, a very significant change in line shape associated with the exchange contribution to 1H R2 would be expected. Since neither of these is observed, the spectra are interpreted as arising from fast exchange, with a minimal averaged chemical shift perturbation. This likely arises from multiple interaction sites and of ligand poses with a small average shift change, which does not exclude substantial shifts on interaction of individual sites or poses. The ‘bound’ form with comparable rotational tumbling to that of Kap95 would be expected to be broadened beyond detection for a τc of ∼60 ns. The 1H line widths are only moderately increased suggesting modest 1H Rex contributions. On this basis, and subject to further investigation of exchange phenomena (for example,;; Lepre et al., 2013; ) and of models of interaction, we analyze the data aswhere R refers to observed, free (0) and bound (b) 15N R2's, and f is the degree of occupancy on the TF for the specific residue. Note that the above equation assumes that any Rex additional contribution is negligible. This is justified if we assume that the actual kex is very large, that is, fast (Equation 17b in ). The back calculation of Kd below then reflects a lower bound value subject to more precise, future determination of Rex terms. The relation of Robs to concentrations of FG Nup and TF then permits analysis of apparent equilibrium constants after, aswhere Tf is the concentration of transport factor, and L is the concentration of FSFG-K. From the observed data, Figure 2—figure supplement 2, the individual values of Kd and Rb are approximately determined. We used two methods which provided the same results within expected precision and experimental error. Using a MATLAB (Mathworks, Natick, MA) script, we produced a grid and found the minimum of conventional χ2 with the unknowns Rb and Kd. The resulting grid was smooth with a single minimum (data not shown). In parallel, we used Prism Version 6.05 (GraphPad Software Inc, La Jolla, CA) with the above equation to obtain nonlinear regression curve fitting to the data deriving Rb and Kd. Since our assumptions of limited exchange contribution, site number equivalence of L and Tf, and homogenous sites are significant simplifications, we then concentrate on the value of Kd as a limiting lower value of the gross dissociation constant, estimated as 36 6 μM. Per site of FSFG-K, for the six nominally equivalent FSFG motifs, this would be equivalent to an estimated lower value for a single site Kd of 216 μM. Using Prism, we also derived sequence-specific apparent Kd assuming that the FSFG residues sense ‘different' equilibria. The values obtained–in μM (Std. Error) were Fsfg 31(6), fSfg 55(15), fsFg 44(9), and fsfG 17(6). The narrow range of values supports the assumption that Rex contributions are modest, since contributions are expected to be highly sequence dependent.

34Dynamic Light Scattering (DLS) measurements were made on a Dynapro Plate Reader (Wyatt Instruments, Santa Barbara, CA) at 298 K, in a 384 well plate with typically triplicate samples. Curves in Figure 4G are summations from instrument measurements. The mass-weighted averaged radii (and standard errors of the mean) are 2.46 (0.15) nm; 2.63 (0.14) nm; 6.47 (0.30) nm for FG-N, FSFG-K and FG-N-FSFG-K-tet, respectively. Concentrations used were 8, 4, and 2 mg/ml. Concentration dependencies were small. For the distribution curves the Dynapro software DYNAMICS 7.1.0.25 was used to analyze the experimental correlations with multiple 1 s runs, standard filtering including exclusions of <1% mass peaks, and regularization using the coil setting. The resulting binned regularized data was extracted and for each protein was summed across wells by scaled Gaussian summation at the peak position and width of the log-scale time base.

35Note that cumulant analysis is not practical for multi-component systems with a large dynamic range. In each case the peak in the 1–10 nm regions represent >99% of the scattering mass. Averages and standard errors of the mean were calculated by pooling each mass-weighted average per well.

36Figure 2—figure supplement 4 shows raw decay time data from DLS for FG–N and FSFG-K constructs in standard buffer A, 8 mg/ml. The vertical axis represents the % of mass having a specific decay time, with period in hours after start shown in the NE axis. The presentation of decay time avoids any issue of model for specific molecular shape or density or validity of the Raleigh-Debye approximation. Data are pooled from a triplicate reading using the setting indicated above.

37A: 20 mM HEPES, 150 mM KCl, 2 mM MgCl2, pH 6.5 unless otherwise noted;B: Xenopus egg extract: protein egg extract prepared according to containing 31 mg/ml protein and was diluted with buffer A (3 + 10 v + v) with a final pH of 6.5;C: E. coli lysate: lyophilized protein from high-speed spin dissolved in buffer A, at a final concentration of 30 mg/ml;D: Buffer A plus 100 mg/ml BSA;E: 20 mM Tris pH 7.5, 150 mM NaCl (Burz et al., 2006).

38A: 20 mM HEPES, 150 mM KCl, 2 mM MgCl2, pH 6.5 unless otherwise noted;

39B: Xenopus egg extract: protein egg extract prepared according to containing 31 mg/ml protein and was diluted with buffer A (3 + 10 v + v) with a final pH of 6.5;

40C: E. coli lysate: lyophilized protein from high-speed spin dissolved in buffer A, at a final concentration of 30 mg/ml;

41D: Buffer A plus 100 mg/ml BSA;

42E: 20 mM Tris pH 7.5, 150 mM NaCl (Burz et al., 2006).

43HADDOCK, a rigid body docking algorithm, was used to combine Kap95 and FSFG-K into possible orientations to provide a molecular representation of our findings. Specifically, the crystal structure of Kap95 (3ND2) and a random conformations of FSFG-K was used as inputs and the FSFG-K repeat residues and residues on Kap95 identified by Isgro et al. were chosen as active residues. Models were subsequently altered and images were rendered using Chimera.