The Structure of the NXF2/NXT1 Heterodimeric Complex Reveals the Combined Specificity and Versatility of the NTF2-like Fold

Abstract

NXF1-like members of the Nuclear eXport Factor (NXF) family orchestrate bulk nuclear export of mRNA, while functionally distinct NXF variant proteins carry out separate substrate and tissue specific RNA regulation. Metazoan organisms possess at least one NXF1-like gene and one or more NXF variant genes. Heterodimerization of both proteins with the NTF2-related eXporT protein (NXT) is central to NXF family function, but given the multiplicity of NXF/NXT complexes, the specificity and mechanism of heterodimerization remains unclear. Here, we report the structural and functional analysis of the Caenorhabditis elegans NXF variant, ceNXF2, bound to ceNXT1. Contacts crucial for NXF/NXT heterodimer stability and specificity have been identified, including a probable site for phosphoregulation. The ceNXF2 NTF2 domain bears at least two nucleoporin (Nup) binding pockets necessary for colocalization of ceNXF2/ceNXT1 at the nuclear envelope. Unexpectedly, one Nup binding pocket is formed at the heterodimer interface of the ceNXF2/ceNXT1 complex, demonstrating that NXT binding directly regulates NXF function.

Introduction

Compartmentalization of eukaryotic cells requires that considerable numbers of molecules be continuously transported between organelles. In the nucleus, molecular import and export is regulated by the nuclear pore complex (NPC).^{1; 2; 3} Effective transport of molecules through the NPC requires soluble transport receptors that simultaneously bind cargo and interact with nucleoporin (Nup) proteins. Although multiple models of NPC translocation exist, it is clear that transport receptor binding of phenylalanine-glycine repeat containing nucleoporins (FG-Nups) is a central requirement.^{3; 4; 5; 6; 7}

The Nuclear Transport Factor 2 like (NTF2-like) fold is found in many nuclear transport receptors, including the prototypical NTF2 protein responsible for RAN-GDP import into the nucleus,^{8; 9} and in Nuclear eXport Factor (NXF) proteins that control mRNA nuclear export.^{10; 11; 12; 13; 14; 15} Cargo binding, NPC binding, and self-association are crucial functions of NTF2-like nuclear transport proteins. In the NXF family, heterodimerization of the NXF NTF2 domain with the NTF2-related eXporT protein (NXT) is required for effective nuclear pore association and translocation.^{16; 17} In addition, there are two distinct functional groups within the NXF family: “NXF1-like” members that control bulk nuclear export of mRNA and “NXF variants” that participate in more specific RNA regulation. Because metazoan organisms have at least one NXF1-like gene and one or more NXF variant genes, it remains unclear what differentiates the multiple NXF/NXT complexes in the cell, and what specificity determinants and mechanisms drive NXF/NXT heterodimerization.

mRNA nuclear export in Caenorhabditis elegans is mediated by the NXF1-like protein ceNXF1, orthologous to human NXF1 (hsNXF1).^{18; 19; 20} Although direct binding between ceNXF1 and ceNXT1 has not yet been established, ceNXF1/ceNXT1 heterodimers are expected to form because ceNXT1 coexpression is required for ceNXF1-mediated mRNA export activity in heterologous human cell assays.¹⁶ The C. elegans genome also encodes the NXF variant gene, ceNXF2, and both ceNXF1 and ceNXF2 exhibit domain architecture consistent with other NXF family proteins (Fig. 1).²⁰ However, ceNXF2 has only 2 of the 4 conserved NXF family domains: the Leucine Rich Repeat (LRR) domain and the NTF2 domain. Because NTF2 and Ubiquitin-Associated (UBA) Nup binding domains are both required in hsNXF1 for effective NPC association and translocation activity,^{17; 21} it is unclear whether ceNXF2 would bind the NPC or function in nuclear transport. Indeed, ceNXF2 is cited as a required factor for nuclear retention of the transformer 2 (tra-2) mRNA during C. elegans sexual development, a role directly antagonistic to ceNXF1-mediated export.²² Due to these functional differences, it was not clear if the NXF variant, ceNXF2, would be expected to bind ceNXT1, or if so, how the structural basis and specificity for a ceNXF2/ceNXT1 complex would compare to NXT binding by NXF1-like proteins.

Fig. 1.

NXF family domain architecture and the NTF2-like fold. The NXF family has four conserved domains: the RiboNucleoProtein domain (RNP), the Leucine Rich Repeat domain (LRR), the Nuclear Transport Factor 2 like domain (NTF2), and the UBiquitin-Associated domain (UBA). Protein partners like ceNXT1 share the NTF2-like fold and bind the NXF NTF2 domain.

Because domains of the NXF family are highly conserved and structural information for NXF variants is nonexistent, the extent of structural divergence and basis for functional differences between NXF variants and NXF1-like proteins remain unclear. Within one organism, it is not known how the NXF/NXT interface accommodates sequence divergence between functionally distinct NXF paralogues, or whether divergent NXF paralogues bind NXT with the same specificity determinants. Furthermore, cross-species NXF1-like heterodimers apparently do not form,^{16; 23} but previous structural studies have not identified the regions or residues responsible for such discrimination.^{11; 12; 14} To address these specific questions, we have determined the 1.84Å resolution crystal structure of the ceNXF2/ceNXT1 heterodimer and have carried out structure-based functional assays to better understand the specifics of ceNXF2/ceNXT1 function and NXF/NXT heterodimerization in general.

Results

ceNXF1 and ceNXF2 Both Form Stable Heterodimeric Complexes with ceNXT1

To directly determine whether ceNXF1 and ceNXF2 can interact with ceNXT1, recombinant ceNXF1, ceNXF2 and ceNXT1 were over-expressed in Escherichia coli for in vitro binding studies. His₆-ceNXT1 interacts separately with both ceNXF1 and ceNXF2, forming complexes that can be purified on Talon cobalt resin (Fig. S1 and S2). Consistent with NXF proteins from other organisms, the NTF2 domains of ceNXF1 (residues 352–539) and ceNXF2 (residues 202–405), referred to as ceNXF1(NTF2) and ceNXF2(NTF2), are responsible for ceNXT1 binding, forming stable 1:1 complexes in gel filtration analysis (Fig. S3). Because of the divergent functions and antagonistic relationship of ceNXF1 and ceNXF2,²² it is intriguing that both paralogues form stable heterodimers with ceNXT1.

Structural Overview of the ceNXF2/ceNXT1 Heterodimer

In order to gain greater insight into ceNXF2 function and reveal the molecular details of NXF/NXT dimerization from the perspective of an NXF variant protein, the X-ray crystal structure of the ceNXF2/ceNXT1 heterodimeric complex was solved. With the exception of 9 flexible residues at the C-terminus of ceNXF2 (residues 397–405), both ceNXF2(NTF2) and ceNXT1 were entirely traced in the electron density map, enabling the first complete structural model of an NXF family NTF2 domain. The asymmetric unit consists of 1 copy of ceNXF2(NTF2) and 1 copy of ceNXT1 interacting with pseudo-2-fold symmetry to form the heterodimer (Fig. 2a). The final structural model was refined to 1.84 Å resolution with an R_free of 19.1% (Table 1). ceNXF2(NTF2) and ceNXT1 both exhibit an NTF2-like fold: a conical shaped α/β roll with an N-terminal helical region packed against a dramatically curved, mostly antiparrallel, β-sheet with 5 to 7 strands (β0–β6) and 3 variable loops (L3–L5) (Fig. 2b and 3). The core NTF2 fold of the larger ceNXF2 protein is accentuated by the “NXF insertion” between β1 and α3 (residues V254-A293), which adds an additional helix (α2.5), a short 3₁₀ helical turn (N281-Q285) and two large loops (L1 and L2). Along with an extended C-terminus and an additional insertion in loop L4, the NXF insertion creates a unique “NXF apical surface” in ceNXF2 that is not present in ceNXT1 (Fig. 2b).

Fig. 2.

Comparative structural overview of the ceNXF2/ceNXT1 complex. (a) Cartoon stereo view representation of ceNXF2(NTF2) (red) in complex with ceNXT1 (violet). (b) Electrostatic surface potential (ESP) representation of ceNXT1 and ceNXF2(NTF2), highlighting hydrophobicity and electrostatics of interface, non-polar (grey), basic (blue), acidic (red). Cartoon representation of ceNXT1 (violet) and ceNXF2(NTF2) (red) in same orientation as ESP. Alpha helix “α”, beta strand “β”, loop “L” and insertion (grey) elements of NTF2 fold are labeled on ceNXF2(NTF2) and ceNXT1. Apical surface of ceNXF2(NTF2) is formed by insertion elements, 90°top view. (c) Ribbon representation of ceNXT1 (red) and ceNXF2(NTF2) (red) superposed with human and Candida albicans orthologues. Same orientation as (b) with an additional 90° side view. Coloring according to species (blue = hsNXT1 and hsNXF1, green = caMex67 and caMtr2). Significant structural differences (black stars) and the Nup binding site (black arrow) are indicated. Boxed “1” and “2” denote ceNXT1 pockets. Structural alignments generated by DALI.²⁷ Structural models in all figures from Pymol (www.pymol.org).

Table 1.

Data Collection, Phasing and Refinement Statistics

	“Refinement Set” †	“Phasing Set”†
Data collection and phasing statistics
Space Group	P212121	P212121
Cell dimensions (Å) – a, b, c	42.00, 49.55, 148.68	42.01, 49.55, 148.69
X-ray source	ALS 5.0.2	ALS 5.0.2
Wavelength (Å)	0.979 (Se-met inflection)	0.979 (Se-met inflection)
Resolution (Å)	50-1.84	50-2.70
Total measurements	106261	34827
Unique reflections	27423	9035
Completeness (%)	98.9 (98.1)	99.9 (100)
I/σ	20.9 (3.3)	30.7 (23.0)
Redundancy	3.9 (3.9)	3.9 (4.0)
Linear Rsym	5.0 (32.7)	3.3 (5.0)
Linear Rmerge	6.8 (38.4)	5.0 (7.0)
FOM (acentric)	–	0.528
Refinement statistics
Rcryst (%)	16.6	–
Rfree (%)	19.1	–
ϕψ most favored (%)*	94	–
ϕψ additionally allowed (%)*	6	–
RMSD bond (Å)	0.023	–
RMSD angle (degrees)	1.59	–
Wilson B (Å2)	17.7	–
Average overall B (Å2)	23.8	–
Protein residues	332	–
Water molecules	279	–
Ions	19 Na	–
PDB accession code	3NV0	–

Values for the outermost resolution shell are shown in brackets. FOM, figure of merit; B, atomic displacement parameter; R_free, R factor of 5% of the reflections excluded from the refinement.

^*Ramachandran values from the PROCHECK program of the PDB validation server http://deposit.rcsb.org/validate/

^†Both “Refinement Set” and “Phasing Set” are from the same dataset. “Phasing Set” describes a subset of the inflection dataset (i.e. anamolous data to 2.7 Å resolution) used for SAD phasing. “Refinement Set” describes the full inflection dataset (i.e. to 1.84 Å resolution) that was used for refinement, model rebuilding and phase extension of the phases determined from the “Phasing Set”.

Fig. 3.

Structure-based sequence alignment of NTF2 family transport proteins. DALI structural alignment of ceNXF2(NTF2) with human (hs) NXF1, S. cerevisiae (sc) Mex67, C. albicans (ca) Mex76, and Rattus norvegicus (rn) NTF2, as well as ceNXT1 with hsNXt1, scMtr2, caMtr2 and rnNTF2. Residues not observable in structures are grey. Secondary structure of ceNXF2(NTF2) and ceNXT1 are shown, and the NXF family insertion and yeast L4 loop insertion are indicated. ceNXF2/ceNXT1 interface direct polar contacts are in red boxes, and residues mutated for pull-down assays are colored red. Blue crosses “+” indicate ceNXF2 and ceNXT1 residues that are structurally equivalent to rnNTF2. Residues in uppercase font for hsNXF1, scMex67, and caMex67 are structurally equivalent to ceNXF2. Residues in uppercase font for hsNXT1, scMtr2, and caMtr2 are structurally equivalent to ceNXT1. Invariant residues are boxed in orange and conserved residues in each subfamily are boxed in yellow. Sequence identities range from 11–30 %, and Cα RMSD values for superposition range from 1.9–2.6 Å for ceNXF2(NTF2) and from 1.5–2.6 Å for ceNXT1 (Table S2). PDB accession codes: rnNTF2 (1OUN), ScMex67/Mtr2 (1OF5), caMex67/Mtr2 (1Q40), hsNXF1/NXT1 (1JKG) and ceNXF2/NXT1 (3NV0).

The ceNXF2(NTF2)/ceNXT1 crystal structure reveals notable structural differences between ceNXF2/ceNXT1 and their NTF2 family orthologues (Fig. 2c and 3). Structural comparisons between the NXF variant, ceNXF2, and NXF1-like proteins, hsNXF1¹⁴ and caMex67,¹¹ reveal that all 3 proteins bear the NXF apical surface in their NTF2 domains, but the sequence and conformation of this region vary considerably among the 3 proteins. Specifically, the sequence identity, length, and conformation of the NXF insertion, the L4 insertion, and the C-terminus all differ between ceNXF2, hsNXF1 and caMex67. Interestingly, a majority of the variable NXF apical surface does not contribute to the NXF/NXT dimerization interface, yet as a gross feature it is highly conserved in the NXF family. Therefore, the NXF apical surface may represent a novel site of NXF-specific NTF2 domain function associated with functional differences between various NXF family members. Consistent with this hypothesis, yeast-specific apical loop insertions in Mex67(NTF2) and Mtr2 have recently been directly linked to nuclear export of pre-60S ribosomal subunits²⁴ as well as interaction with Nup84 complexes.²⁵

Structural comparisons between the NXF protein partners ceNXT1, hsNXT1,¹⁴ and caMtr2¹¹ highlight significant structural differences in multiple regions of each NTF2 fold (Fig. 2c and 3). In particular, two unique pockets are found on ceNXT1 that may be indicative of unanticipated ceNXT1 functions. The first pocket, which is occluded in hsNXT1, is roughly 200 ų and analogous to the Ran-GDP binding pocket of NTF2,⁹ exhibiting surprising sequence conservation and a similar overall contact surface. Unlike caMtr2, which possesses a similar pocket, the macroscopic features of the region surrounding the ceNXT1 pocket dock properly with the switch II region of Ran-GDP when modeled. The second ceNXT1 pocket is a rather large solvent exposed cleft (at least ~ 330 ų), not present in caMtr2 and occluded in hsNXT1, that is equivalent in structural position to the FG-Nup binding pocket of hsNXF1. Interestingly, this cleft is even more pocket-like when ceNXF2 is bound to ceNXT1, suggesting a possible dimerization-dependent interaction surface, perhaps involved in Nup binding. Overall, the ceNXF2(NTF2)/ceNXT1 structure exposes a number of structural differences between NXF1-like heterodimers and NXF variant heterodimers, which possibly represent novel interaction sites and warrant further investigation.

The ceNXF2/ceNXT1 Interface Reveals Important Contacts and Specificity Determinants for NXF/NXT Heterodimerization

The ceNXF2/ceNXT1 interface contains roughly 1,900 Ų of buried surface area and consists of a combination of shape and electrostatic complementarity (Fig. 4a and 4c). Despite this large interface, only 12 side-chains from ceNXF2 and 9 side-chains from ceNXT1 are involved in direct polar contacts that connect the two proteins (Fig. 4c and 3). However, this relative lack of polar contacts is consistent with other known NXF/NXT complexes.^{11; 12; 14} Linking ceNXF2 and ceNXT1, there are 3 salt bridges (H280/D79, D323/R134 and R389/D137) and 1 cation-pi interaction (W258/K110). In addition, roughly 2/3 of all the direct contacts exist in an intimately packed, midline region of the interface (Fig. 4b). Of residues involved in direct interface contacts, only 5 from ceNXF2 and 4 from ceNXT1 are homologous to structurally equivalent residues in hsNXF1/hsNXT1 and Mex67/Mtr2 (Fig. 3). Interestingly, all but one of the conserved contacts in ceNXF2/ceNXT1 cluster at the midline region of the interface.

Fig. 4.

Mapping the specificity determinants of NXF/NXT heterodimerization. (a) ceNXF2/ceNXT1 complex shown, highlighting intimacy of the ceNXT1 (violet) and ceNXF2(NTF2) (ESP) interaction. Orientation of ceNXF2(NTF2) in subsequent panels is unchanged. (b) Cartoon representation of ceNXF2(NTF2) (grey) with direct polar contacts (green) and indirect H₂O mediated contacts (cyan) shown. Residues N247, D323, Q379, and R389 (all colored green) are also involved in indirect H₂O mediated contacts. Inset highlights a cluster of 1 cation-pi interaction and 2 salt bridges, residues labeled. (c) ESP representation comparison of NTF2 domain dimerization interfaces from ceNXF2, hsNXF1 and caMex67. The NXF plug motif, common to all three structures, denoted by black box. Inset of the ceNXF2 NXF plug, highlighting key plug forming residues (labeled in black) and the identities of the ceNXT1 side-chains that pack into the pockets (labeled in yellow). (d) Simple schematic representation of the intricate NXF plug interaction between ceNXF2 and ceNXT1. (e) Position and identity of residues from homologous NXF plug interactions, in same orientation as ceNXF2, with NXF1/Mex67 residues (black) and NXT1/Mtr2 residues (red) indicated.

The ceNXF2/ceNXT1 interface overcomes the apparent dearth of direct polar contacts in 2 separate ways. First, the majority of direct contacts, especially those conserved, are situated in a region of the interface where shape complementarity is highest, suggesting extensive VDW interactions help strengthen each individual contact (Fig. 4b–c). Second, 13 side-chains from ceNXF2 and 11 side-chains from ceNXT1 are involved in a network of indirect H₂O-mediated contacts that circumscribe the midline region and supplement interface stability (Fig. 4c). The indirect nature of these “wet” contacts helps explain the relative lack of sequence conservation among paralogous and orthologous NXF/NXT interfaces.

Intimate packing in the ceNXF2/ceNXT1 interface midline is most pronounced at a region referred to as the “NXF plug” because of the “knobs-into-hole” style shape complementarity that resembles a multipin plug/receptacle interaction (Fig. 4c and d). The NXF plug is a cluster of 9 side-chains from ceNXF2, arranged in a 3×3 grid, that accepts 4 residues from ceNXT1, arranged in a 2×2 grid, and contains 5 buried hydrogen bonds. The NXF plug location and architecture is conserved among NXF/NXT interfaces, whereas sequence identity within the plug is more divergent (Fig. 4c and e).^{11; 12; 14} This combination enables distinct networks of buried polar contacts while maintaining extensive shape complementarity. Indeed, ceNXF1(NTF2) homology modeling with ceNXF2(NTF2) using ProtMod (http://ffas.burnham.org/protmodcgi/protModHome.pl) reveals that the ceNXF2 NXF plug has higher sequence identity with the ceNXF1 plug than with the plugs of hsNXF1 or caMex67 (Fig. 4e), suggesting that the NXF plug may serve as a site of speciation.

A number of the direct polar contacts between ceNXF2 and ceNXT1 were mutated and tested for an effect on complex formation (Table 2, Fig. 3, Fig. S1 and S2). Three contact sites, situated in the conserved midline of the dimer interface, are thermodynamically important for stabilization of the ceNXF2/ceNXT1 complex. First, disruption of the buried S294/D13 hydrogen bond between the L2 loop of ceNXF2 and the α1 helix of ceNXT1 does not completely abolish the ceNXF2/ceNXT1 interaction, but the effect on binding is notable given the loss of only a single hydrogen bond. Second, mutation of the conserved D323/R134 salt bridge at the opposite end of the interface midline completely ablates binding, consistent with the homologous salt bridge in hsNXF1/hsNXT1 being required for mRNA export.¹⁴ The compensatory charge swap mutation (i.e. D323R and R134D) fully restores binding between ceNXF2 and ceNXT1, indicating that the presence of a salt bridge, and not the identity of the residues involved, is important for binding. In addition, mutating the homologous D372/R134 salt bridge in ceNXF1/ceNXT1 also destroys complex formation, confirming the importance of this salt bridge in NXF/NXT interfaces. Third, disruption of contacts within the NXF plug of ceNXF2/ceNXT1 significantly inhibits heterodimerization. Specifically, both the double D79Y/Q81Y mutation in ceNXT1 and the single Q379K mutation in ceNXF2 destroy complex formation, presumably by imposing steric hindrance on the tight shape complementarity of the NXF plug and by disrupting the hydrogen bonding network. Effects of the ceNXF2 Q379K mutation are consistent with previous reports of disrupted homodimerization due to a structurally equivalent M118E mutation in NTF2,²⁶ as well as defects in mRNA nuclear export resulting from a distinct I518R mutation in hsNXF1.¹⁴ Mutation of the putative NXF plug contacts in ceNXF1/ceNXT1 also inhibits dimerization, confirming the importance of the NXF plug in NXF/NXT dimerization.

Table 2.

Crucial Interface Contacts and Specificity Determinants of NXF/NXT Heterodimerization

ceNXF2	ceNXT1	Binding		ceNXF1	ceNXT1	Binding		hsNXF1	ceNXT1	Binding
wt	wt	+	:::	wt	wt	+	:::	wt	wt	−
R389A	wt	+
R389D	wt	+
S294A	wt	+
S294D	wt	±
H280A	wt	+
H280D	wt	+
D323R	wt	−
D323A	wt	−
Q379K	wt	−
Q379Y	wt	±
wt	K110Q	+	:::	wt	K110Q	+
wt	R134A	+	:::	wt	R134A	±
wt	R134D	±	:::	wt	R134D	−
D323A	R134A	±
D323R	R134D	+
D323R	R134A	−
D323A	R134D	±
wt	D79Y/Q81Y	−
wt	D79A	+	:::	wt	D79A	+
wt	D79H	+	:::	wt	D79H	±
wt	D137A	+	:::	wt	D137A	+
wt	D137R	+	:::	wt	D137R	+
wt	N98V	+	:::	wt	N98V	+	:::	wt	N98V	−
								E533Q	wt	±
								E533Q	N98V	+

Binding activity measured by pull-down and evaluated by SDS-PAGE

“+” means wild type (wt) binding (≥ 80% of wt binding achieved).

“±” means weak/partial binding (~ 25–75% of wt binding achieved).

“−“ means no appreciable binding (≤ 20% of wt binding achieved).

Outside of the ceNXF2/ceNXT1 midline region, independent disruption of either the R389/D137 salt bridge or the W258/K110 cation-pi interaction has no appreciable effect on dimerization. The same is true in ceNXF1/ceNXT1, where mutation of the putative salt bridge homologous to R389/D137 in ceNXF2/ceNXT1 has no effect on ceNXF1/ceNXT1 complex formation. Therefore, it appears that individual contacts outside the interface midline are not necessary for NXF/NXT complex stability while single contacts along the interface midline are crucial for NXF/NXT heterodimerization.

In addition to identifying thermodynamically important sites in the NXF/NXT interface, mutagenesis also reveals that changing the identity of single NXF plug contacts can radically alter the specificity of NXF/NXT binding (Table 2, Fig. 3, Fig. S1 and S2). It appears that cross-species NXF/NXT heterodimers do not form,^{16; 23} and hsNXF1 does not interact with ceNXT1 (Table 1, Fig. S2). However, it is not known what elements of the NXF/NXT interface are important for specificity within one organism or across organisms. Based on the unique mixture of structural conservation and divergent residue identity in the NXF plug, as well as close proximity to other important contacts (Fig. 4c–e), we hypothesized that the NXF plug represents an NXF/NXT dimerization fingerprint, with differences in sequence identity dictating differences in specificity. To test this hypothesis, mutagenesis of the NXF plug was carried out. A single D79H mutation in ceNXT1 exhibits no decrease in ceNXF2 binding even though the D79Y/Q81Y double mutation in ceNXT1 does disrupt ceNXF2/ceNXT1 dimerization. Remarkably, the same ceNXT1 D79H mutant significantly disrupts ceNXF1 binding, indicating that paralogous NXF/NXT interfaces do not interact entirely in the same way.

A single NXF plug residue in hsNXF1 confers specificity on the hsNXF1/hsNXT1 interaction, and when mutated, relaxes specificity to allow novel cross-species ceNXT1 binding in addition to cognate hsNXT1 binding. Specifically, mutating the central E533 residue of the hsNXF1 plug to glutamine to mimic the plugs of ceNXF1 and ceNXF2 enables cross-species heterodimerization of hsNXF1 and ceNXT1, while not impeding hsNXT1 binding. In contrast, mutating N98 of the ceNXT1 plug to valine to mimic hsNXT1 does not enable ceNXT1 to bind hsNXF1, but it also does not impede ceNXF1 or ceNXF2 binding. Unexpectedly, combining ceNXT1 N98V with hsNXF1 E533Q results in synergistic binding of hsNXF1 to ceNXT1. Interestingly, the glutamine substitution in hsNXF1 is fully functional for hsNXT1 binding, yet the glutamate residue is highly conserved only in NXF1 and NXF2 genes from higher eukaryotes, suggesting that the negative charge may be important for discrimination in NXF/NXT interactions or regulation of heterodimerization.

ceNXF2 Contains an Unexpected Yet Conserved FG Pocket Resembling a Nucleoporin (Nup) Binding Site

The UBA domain that is conserved in other NXF family members and is important for nuclear pore association and transport^{17; 21} is not found in ceNXF2. Thus, it was reasonable to infer that ceNXF2 may not be export competent and, instead, might be a nuclear retention factor.²² However, the ceNXF2/ceNXT1 crystal structure challenges this conclusion by demonstrating that ceNXF2 contains a structurally conserved FG pocket, resembling a Nucleoporin (Nup) binding site, that can accommodate a phenylalanine side-chain. Structural alignment of ceNXF2(NTF2) and hsNXF1(NTF2) using the DALI program²⁷ reveals a hydrophobic cavity in ceNXF2 between α1 and the β3b-β4 hairpin that is of similar location, shape and size as the FG pocket of hsNXF1¹⁴ (Fig. 5a–c). Most of the pocket forming side-chains in hsNXF1 (L383, L386, Q486, L491, A519, P521, and L527) are conserved in the ceNXF2 pocket (V212, L215, V327, L332, V365, P367, V373 respectively), and the 3 side-chains that interact most intimately with the phenylalanine of the FG peptide are identical in both proteins (P367/P521, L332/L491, and L215/L386) (Fig. 5a–c and 3). Interestingly, slight divergence between ceNXF2 and hsNXF1 in the pocket (i.e. ceNXF2 V212, V327, G371, V373 vs. hsNXF1 L383, Q486, S523, L572) creates a slightly larger cavity in ceNXF2 that features a small cleft on either side of the main phenylalanine binding pocket (Fig. 5c vs. 5a).

Fig. 5.

ceNXF2 has a conserved nucleoporin FG repeat binding site. (a and c) ESP representations of ceNXF2 and hsNXF1 comparing the putative FG pocket on ceNXF2 with the known hsNXF1 FG pocket. Key conserved residues that pack against inserted phenylalanine side-chain are labeled. (b and d) Superposition of FG pockets from ceNXF2 (red) and hsNXF1 (blue), in same orientation as (a and c), highlighting conservation. Key pocket forming residues labeled (hsNXF1 in blue, ceNXF2 in red) and secondary structure labels (green) given for orientation. (b) The putative FG pocket of ceNXF2 can accommodate the inserted phenylalanine side-chain and backbone conformation of the FG peptide (yellow with black residue labels) bound to hsNXF1 (PDB: 1JN5). (d) Fortuitous crystal lattice contact in ceNXF2/ceNXT1 proves that the ceNXF2 pocket can bind phenylalanine. The symmetry related L2 Loop of ceNXF2 (green with black residue labels) packs Phe288 into the pocket. ceNXF2 L332 exhibits alternative conformations in the crystal structure.

Direct evidence of phenylalanine binding to the ceNXF2(NTF2) FG pocket is observed in the ceNXF2/ceNXT1 structure due to fortuitous crystal lattice interactions. Specifically, the unique L2 loop of the ceNXF2 apical surface buries 650 Ų of surface area through a crystal contact with the FG pocket of a symmetry related ceNXF2 molecule. At the core of this interaction, Phe288 is flipped out from the L2 loop and buried between conserved Pro367 and L332 side-chains in a manner similar to the binding of the FG peptide by hsNXF1 (Fig. 5d vs. 5b). In addition, the L2 loop and FG peptide backbones are both stabilized by similar intermolecular hydrogen bonds with the cavity wall. Interestingly, Y287 of the ceNXF2 L2 loop, adjacent to the buried F288, packs nicely into a neighboring cleft unique to the ceNXF2 pocket (Fig. 5c and d), suggesting that ceNXF2 could bind GLFG and/or FxFG repeat containing Nups. In addition, surfaces adjacent to the ceNXF2 FG pocket are significantly more polar than in hsNXF1 (Fig. 5c vs. 5a), suggesting potential differences in Nup specificity between the two proteins.

The ceNXF2/ceNXT1 structure yields possible evidence for the presence of two auxiliary Nup binding sites in addition to the conserved FG pocket of ceNXF2. As previously mentioned, ceNXT1 contains a unique surface exposed cleft at the heterodimer interface supplemented by residue contributions from ceNXF2. This site is equivalent in structural position to the FG pocket of hsNXF1 and may serve as a supplementary Nup binding site in the ceNXF2/ceNXT1 heterodimer. At a different location along the heterodimer interface and immediately adjacent the ceNXF2 β3b–β4 hairpin, which demarcates the conserved ceNXF2 FG pocket, lies a crescent shaped cleft that may serve as yet a second possible auxiliary Nup binding site. Considering that in full length ceNXF2, the N-terminus of the NTF2 domain is actually part of a linker connecting the C-terminus of the LRR domain, it is possible that the structured conformation of ceNXF2 residues F202–G208 in the ceNXF2/ceNXT1 structure is not representative of the likely flexible nature of that region in full length ceNXF2. Indeed, if residues F202–G208 are removed from the ceNXF2/ceNXT1 structure, a cleft is found at the dimer interface that is contiguous with the crescent shaped cleft already described. Interestingly, when structurally aligned with the scNTF2 homodimer, both additional ceNXF2/ceNXT1 pockets, although chemically distinct, are at equivalent structural positions as the FxFG binding sites of scNTF2.⁸ Overall, strong structural evidence for a functional FG binding pocket on the putative nuclear retention factor ceNXF2, as well as the possibility for two additional dimerization-dependent Nup binding sites on ceNXF2/ceNXT1, demonstrate that a functional re-evaluation of ceNXF2 is warranted.

The Putative Nuclear Retention Factor ceNXF2 Colocalizes with ceNXT1 at the Nuclear Envelope in a Dimerization-Dependent Manner

Although localization of ceNXF1 to the nuclear rim has been previously shown in HeLa cells,¹⁹ the subcellular localization and possible colocalization of ceNXF2 and ceNXT1 are not known. In keeping with the experimental precedents of using HeLa cells for localization studies of nuclear export proteins,¹⁹ the localization and colocalization patterns of ceNXF1, ceNXF2 and ceNXT1 were analyzed in HeLa cells using fluorescence microscopy. In agreement with previous work,¹⁹ ceNXF1 is localized predominantly in the nucleoplasm and at the nuclear envelope, displaying a punctate pattern in pre-extracted cells (Fig. 6a, Fig. S4b) that is characteristic of hsNXF1 localization to NPCs.^{14; 21} Surprisingly, ceNXT1 is found predominantly in the cytoplasm (Fig. 6a, Fig. S4d), in contrast to the nuclear localization patterns previously seen for its mouse²⁸ and human orthologues.²³ Because hsNXT1 localization to the nuclear rim has been attributed to its association with hsNXF1,¹⁴ ceNXT1 cytoplasmic localization could reflect a failure of ceNXT1 to form productive complexes with hsNXF1 in HeLa cells, consistent with the inability of ceNXT1 to bind hsNXF1 in vitro (Table 2). Finally, ceNXF2 is found primarily in the perinuclear space of the cytoplasm with a few points of possible nuclear rim localization (Fig. 6a, Fig. S4c).

Fig. 6.

A single NPC binding domain is sufficient for ceNXF2/ceNXT1 colocalization at the nuclear envelope. (a–d) HeLa cell localization/colocalization studies, all cells stained with Hoechst 33342 nuclear stain (blue channel). Cell images shown are representative of multiple cells. (a) Cells pre-extracted with 0.1% Triton X-100. (b–d) Cells fixed directly in 4% CH₂O (a–b) Merged images from green, red and blue channels (individual channels in supplemental Fig. 4a–d). (a) Localization of ceNXF1-RFP to nucleoplasm and nuclear rim, localization of ceNXF2-RFP to cytoplasmic perinuclear space, and localization of ceNXT1-EGFP to cytoplasm. (b) Colocalization of hsNXF1-RFP and hsNXT1-EGFP at the nuclear rim. (c) Colocalization of ceNXF1-RFP and ceNXT1-EGFP at the nuclear rim. (d) Colocalization of ceNXF2-RFP and ceNXT1-EGFP at the nuclear rim. (c–d) ceNXT1-EGFP enhances association of ceNXF1-RFP, and especially ceNXF2-RFP, with the nuclear envelope.

To determine whether ceNXF1 and ceNXF2 colocalize with ceNXT1 in vivo, ceNXF1, ceNXF2 and ceNXT1 were cotransfected into HeLa cells and analyzed by fluorescence microscopy. In agreement with the in vitro binding results and expectations from literature, ceNXF1 and ceNXT1 colocalize at the nuclear rim (Fig. 6c), and as a positive control, cotransfected hsNXF1 and hsNXT1 also exhibit nuclear envelope colocalization (Fig. 6b, Fig. S4a). Interestingly, ceNXF2 and ceNXT1 also colocalize at the nuclear rim (Fig. 6d). These results imply that both ceNXF1 and ceNXF2, unlike endogenous hsNXF1, are able to recruit ceNXT1 to the nuclear rim through the direct protein-protein interactions observed in vitro. In addition, coexpression of ceNXF1 with ceNXT1 enhances the nuclear rim association of both proteins, consistent with previous results for hsNXF1/hsNXT1.¹⁶ The nuclear-rim-association enhancing effect of ceNXT1 is even more pronounced on ceNXF2, which changes from weak nuclear envelope association to intense association when coexpressed with ceNXT1.

The ability of ceNXF2 to associate with the nuclear rim is particularly notable because ceNXF2 lacks the UBA domain, known to be important in hsNXF1 for NPC association and mRNA nuclear export.^{14; 17; 21} In addition, ceNXF2/ceNXT1 colocalization data suggest that heterodimerization plays a more central role in the nuclear rim localization of ceNXF2 than it does for ceNXF1 or hsNXF1. As the homodimerization of NTF2 has been shown to directly effect its affinity for FxFG Nup repeats,²⁶ one possible mechanism for dimerization-dependent nuclear rim localization for ceNXF2/ceNXT1 is a Nup binding pocket formed by the dimer interface. Because two large pockets at the ceNXF2/ceNXT1 dimer interface have been identified in structural positions homologous to the FxFG pockets of NTF2,⁸ we hypothesized that these sites may supplement or supplant the conserved FG pocket of ceNXF2 for Nup binding. To test this hypothesis, mutagenesis of the conserved FG pocket, ceNXF2(L215R/L332Y), as well as the two additional dimer interface pockets, ceNXF2(V327W/S328R/T330R) and ceNXT1(E12W/E85R), was carried out and colocalization patterns were again analyzed by fluorescence microscopy. Cotransfection of ceNXT1(wt) and either ceNXF2(L215R/L332Y) or ceNXF2(V327W/S328R/T330R) does not result in nuclear rim colocalization (Fig. 7a and b). In contrast, cotransfection of ceNXF2(wt) with ceNXT1(E12W/E85R) does result in nuclear rim colocalization in a manner dependent on ceNXF2, as transfection of ceNXT1(E12W/E85R) alone does not result in nuclear rim localization (Fig. 7c and d). These data show that while the conserved FG pocket on ceNXF2 is necessary for NPC association, it is not sufficient and is instead supplemented by an adjacent dimerization-dependent cleft that is homologous in location with an NTF2 FxFG binding pocket. Overall, a single NPC binding domain (i.e. the NTF2 domain) is sufficient for ceNXF2/ceNXT1 nuclear rim association, and such localization is intriguing given the proposed function of ceNXF2 as a nuclear retention factor.

Fig. 7.

The conserved ceNXF2 FG pocket and a novel dimerization-dependent Nup binding pocket are both necessary for colocalization of ceNXF2/ceNXT1 at the nuclear envelope. (a–d) HeLa cell localization/colocalization studies, cells fixed directly in 4% CH2O, permeabilized, and stained with Hoechst 33342 nuclear stain (blue channel). Cell images shown are representative of multiple cells. (a) ceNXF2(L215R/L332Y)-RFP and ceNXT1-EGFP do not colocalize at the nuclear rim. (b) ceNXF2(V327W/S328R/T330R)-RFP and ceNXT1-EGFP do not colocalize at the nuclear rim. (c) ceNXF2-RFP and ceNXT1(E12W/E85R)-EGFP colocalize at the nuclear rim. (d) ceNXT1(E12W/E85R)-EGFP does not localize at the nuclear rim without ceNXF2 present.

Discussion

The crystal structure of the ceNXF2/ceNXT1 heterodimer presented here represents the first structure of any nuclear transport protein from C. elegans and the first structure of an NXF family NXF variant bound to its protein partner. Comparison of the ceNXF2/ceNXT1 structure with NXF1-like NXF/NXT heterodimers has identified several significant areas of structural divergence among the NTF2 domains of NXF1-like proteins and NXF variants, including some unexpected surfaces on ceNXF2(NTF2) and ceNXT1 that exhibit unanticipated functions. Structure-based mutational analysis has revealed novel insights into the molecular details and specificity of NXF/NXT heterodimerization. The NXF/NXT interface must exhibit sufficient thermodynamic stability to support function, while being plastic enough to accommodate the divergent residues of multiple paralogous binding partners. In addition, individual contacts within the interface midline are required for NXF/NXT dimerization, while individual contacts at the interface periphery are not. Lastly, the structurally conserved NXF plug, in particular, is crucial for both the thermodynamic stability and specificity of NXF/NXT complexes. Overall, the NXF/NXT specificity determinants described here establish a significant precedent for enabling systematic disruption of specific NXF/NXT complexes in vivo.

While it is postulated that NXF1 function is modulated by heterodimerization, it is not known how NXF1–NXT1 binding is regulated.^{14; 16; 17} Analysis of the NXF plug motif, common in all known NXF/NXT structures, has revealed it as a likely site for dimerization regulation. The overabundance of serine and threonine residues in known NXF plugs (Fig. 4) suggests that phosphoregulation may control NXF/NXT heterodimerization. Because of the intimacy of many NXF plug contacts, steric clash and disrupted hydrogen bonding caused by the phosphorylation of either protein would necessarily inhibit complex formation. Indeed, the hsNXF1 E533Q mutant alone shows the potential regulatory power of a single negative charge not only on heterodimer formation, but also on specificity. Interestingly, preliminary results demonstrate that ceNXF2 can be phosphorylated in vitro (unpublished data), and further efforts to identify specific sites of NXF/NXT phosphoregulation are currently ongoing.

Even though NXF/NXT heterodimerization is central to NXF family function, there is currently no model to explain the mechanism of NXF/NXT association. Binding mechanisms between proteins appear to be governed primarily by protein native topology and mimic mechanisms of protein folding–referred to as the minimal frustration principle.²⁹ High hydrophobicity at the dimer interface and a large ratio of intermonomer native contacts to intramonomer native contacts (called the NC distribution) correlates with an obligate 2-state association mechanism, in which dimerization precedes monomer folding.^{30; 31} In contrast, 3-state dimers, where monomer folding precedes dimerization, are characterized by low interface hydrophobicity and a low NC distribution ratio.^{30; 31} To better understand the mechanism of NTF2 domain dimerization and how it relates to monomer folding, NC distributions^{30; 31} and interface hydrophobicity values³² were calculated for NTF2 homodimers and various NXF/NXT heterodimers from the NTF2 family (Table S1). Overall, comparative native topology data reveal that NXF/NXT heterodimerization proceeds primarily through a 3-state association mechanism with elements of induced fit. A 3-state association mechanism for NXF/NXT suggests an intracellular equilibrium of monomeric and heterodimeric forms of NXF and NXT, similar to the monomer-dimer equilibrium for NTF2.²⁶ Incorporating induced fit into the ceNXF2/ceNXT1 binding mechanism could be important for enabling water molecules to link unsatisfied polar regions at the dimer interface, and also for allowing rearrangement of flexible divergent regions, such as ceNXF2 loops L1 and L2, for maximal thermodynamic stability.

The ceNXF2/ceNXT1 structure and corresponding structure-based functional assays have revealed new insights into ceNXF2 function. ceNXF2 contains an unexpected FG binding pocket analogous to the nucleoporin (Nup) binding site of hsNXF1 as well as a dimerization-dependent Nup binding site not previously observed in other NXF proteins. Heterodimerization of ceNXF2/ceNXT1 appears necessary for nuclear rim association, as the conserved ceNXF2 FG pocket alone is not sufficient. Unlike NXF1-like proteins, the NTF2 domain of ceNXF2 is sufficient for nuclear rim association, indicating that the UBA domain may not be necessary for NPC binding by all NXF family members. Indeed, the type of cargo a nuclear transporter carries has been shown to influence the specific requirements for NPC translocation–hsNXF1 only needs one Nup binding domain when transporting CTE RNA.²¹ Due to the cargo it carries, ceNXF2 may have evolved to lack the UBA domain for the purpose of enabling interaction with the NPC in a way that is distinct from the NXF1-like protein ceNXF1. Considering that the UBA domain is constitutively active for NPC translocation,²¹ nuclear translocation of ceNXF2 should be more tightly regulated than for ceNXF1, mirroring the tight regulation of expression experienced by the tra-2 mRNA cargo that ceNXF2 carries. Finally, the discovery of a dimerization-dependent Nup binding pocket on ceNXF2 elucidates a possible mechanism for cellular regulation of ceNXF2 mediated export pathways, and demonstrates how NXT binding can directly modulate NXF function in general. However, it is important to note that the ability of ceNXF2 to translocate through the NPC has not been directly evaluated and, thus, NPC binding by ceNXF2 is only suggestive of translocation functions. In conclusion, the structure of the ceNXF2/ceNXT1 complex has provided considerable comparative insights into NXF family function, and has also raised interesting questions to be addressed by further experimentation in vitro and in vivo.

Materials and Methods

Plasmid Preparation

For structural studies, nucleotide sequences for ceNXF2(NTF2), residues 202–405, and ceNXT1 (residues 2–137), were subcloned into pETDuet-1 (Novagen) and transformed into BL21 (DE3) E. coli cells. The resulting ceNXT1 construct bears a non-cleavable 18 residue N-terminal His₆ fusion tag for purification. The ceNXF2(NTF2) construct lacks fusion tags. For in vitro protein pull-down assays, the same ceNXF2(NTF2) pETDuet-1 construct was used. The nucleotide sequence for ceNXF1(NTF2), residues 352–539, and hsNXF1(NTF2), residues 371–555, were subcloned into pETDuet-1, yielding constructs with no fusion tags. ceNXT1 and hsNXT1 were subcloned into pET28b (Novagen), yielding constructs with thrombin-cleavable 20 residue N-terminal His₆ fusion tags for purification. The NXF and NXT constructs were co-transformed into E. coli BL21 (DE3) cells for co-expression of NXF/NXT complexes. Point mutations mentioned in Fig. 7 and Table 2 were introduced via QuickChange site-directed mutagenesis (Stratagene). For HeLa cell fluorescence localization assays, full length nucleotide sequences for ceNXF1, ceNXF2, ceNXF2(L215R/L332Y), ceNXF2(V327W/S328R/T330R), and hsNXF1 were all subcloned into the pDsRed1-N1 mammalian expression vector, yielding constructs with C-terminal RFP fusions. Full length nucleotide sequences for ceNXT1, ceNXT1(E12W/E85R), and hsNXT1 were subcloned into the pEGFP-N1 mammalian expression vector, yielding constructs with C-terminal EGFP fusions.

Protein Crystallization and Data Collection

Unlabeled ceNXF2(NTF2) was over-expressed in Luria-Bertani (LB) media while SeMet-labeled ceNXT1 was grown in M9 minimal media supplemented with 30 mg/L L-selenomethione directly prior to induction. Both cultures were grown at 25°C and induced with 0.5 mM isopropyl-b-d-thiogalactopyranoside (IPTG) for 4 hrs. Purification was carried out at 4°C. Cell pellets from each culture were resuspended together in lysis buffer (100 mM NaCl, 50 mM Hepes pH 7, 20 mM imidazole, 7 mM 2-mercaptoethanol (βME) and 1 Roche Complete protease inhibitor tablet) and subjected to gentle sonic dismembration followed by stirring with 0.5× FastBreak cell lysis solution (Promega) for 30 minutes and nucleic acid removal with 0.01 % v/v polyethyleneimine (PEI) for 30 minutes. ceNXF2(NTF2)/ceNXT1 complex was purified with Talon metal affinity resin (Clontech) followed by gel filtration. The 1:1 stoichiometric complex was eluted from superdex 200 16/60 (GE Healthcare) in storage buffer (100 mM NaCl, 50 mM Hepes pH 7, 7 mM βME). The ceNXF2(NTF2)/ceNXT1 complex was crystallized by vapor diffusion in sitting drop plates at 20°C after mixing 1 µl of 122 µM complex with 1 µl of well solution (28% (w/v) PEG 3350, 300 mM KCl). Orthorhombic needle-shaped crystals appeared within 3 days belonging to the p212121 space group (cell dimensions: a = 42.0 Å, b = 49.5 Å, c = 148.7 Å, and α = β = γ = 90°). Crystals were cryoprotected with 35% PEG 3350 and flash frozen in liquid nitrogen. MAD data were collected at 3 wavelengths (peak, inflection, remote) from a single crystal to 1.84 Å resolution (ALS 5.0.2, Berkeley). All oscillation images were integrated, scaled and merged using HKL2000 (www.hkl-xray.com).

Structure Determination and Refinement

SAD phasing of the inflection dataset (λ = 0.979 nM) using Phenix AutoSol³³ identified 5 of 7 Se atoms present in the asymmetric unit, resulting in experimental phases to 2.7 Å resolution (overall figure of merit (FOM) = 0.45, FOM_acentric = 0.528). Iterative rounds of automated model building and phase improvement (solvent flattening and phase extension) with the Phenix AutoBuild³³ produced an electron density map interpretable to 1.84 Å resolution, and a good initial model (R_free = 26%). This initial model was manually evaluated and rebuilt in Coot,³⁴ and refined with the Phenix refinement program.³³ Iterative rebuilding and refinement converged on a final model with R = 16.6% and R_free = 19.1%, and good stereochemistry (94% of residues in the most favored region and 6% of residues in the additionally allowed region).

In Vitro Affinity Pull-down Assays

Co-expression cultures were induced at 25°C with 1 mM IPTG for 3 hours. Cell pellets were resuspended in same lysis buffer as above (however pH 8.0 is optimal for samples containing hsNXF1(NTF2) or hsNXT1, whereas either pH 7.0 or pH 8.0 is equivalent for other proteins tested), and protein was extracted as above. Pull-down assays were performed by applying cleared lysates to pre-equilibrated His SpinTrap columns (GE Healthcare). Columns were washed 3 times with 0.6 ml lysis buffer and eluted with 0.4 ml lysis buffer containing 300 mM imidazole. Protein bands from SDS-PAGE analysis were quantified with in-house software. A systematic procedure was employed to ensure consistent quantitation of gel bands and to help account for slight differences in gel loads and gel staining: 1) the intensity of all NXF and NXT bands on a gel were quantified; 2) within a single gel lane, a relative intensity ratio was calculated between NXF and NXT bands; 3) this NXF:NXT ratio was then normalized against the NXF:NXT ratio from a gel lane loaded with wild type NXF/NXT complex. Table 2 describes how normalized NXF:NXT ratios were used to evaluate NXF/NXT complex formation.

HeLa Cell Fluorescence Localization Assays

HeLa cells in log phase were trypsinized and seeded in a 6-well dish (Corning) containing microscope cover slips. 24 hours later, the cells were transfected (0.5 µg/well for eGFP plasmids and 1.5 µg/well for dsRed plasmids) using Lipofectamin 2000 (Invitrogen). 48 hours after transfection, cells on coverslips were washed and either fixed directly with 4% formaldehyde (CH₂O) or pre-extracted with 0.1% Triton X-100 for 1 min before fixation. Fixed cells were permeabilized and stained with Hoechst 33342 (Invitrogen). Samples were analyzed on either a Leica DMIRE2 microscope, a Zeiss LSM 710 LSCM, or a Bio-Rad Radiance 2100 Rainbow LSCM.

Abbreviations

NPC

Nuclear Pore Complex

Nup

Nucleoporin

FG-Nups

phenylalanine-glycine repeat containing nucleoporins

NTF

Nuclear Transport Factor

NXF

Nuclear eXport Factor

NXT

NTF2-related eXporT protein

LRR

Leucine Rich Repeat

UBA

Ubiquitin-Associated