Crystal structure of a SFPQ/PSPC1 heterodimer provides insights into preferential heterodimerization of human DBHS family proteins

Members of the Drosophila behavior human splicing (DBHS) protein family are nuclear proteins implicated in many layers of nuclear functions, including RNA biogenesis as well as DNA repair. Definitive of the DBHS protein family, the conserved DBHS domain provides a dimerization platform that is critical for the structural integrity and function of these proteins. The three human DBHS proteins, splicing factor proline- and glutamine-rich (SFPQ), paraspeckle component 1 (PSPC1), and non-POU domain–containing octamer-binding protein (NONO), form either homo- or heterodimers; however, the relative affinity and mechanistic details of preferential heterodimerization are yet to be deciphered. Here we report the crystal structure of a SFPQ/PSPC1 heterodimer to 2.3-Å resolution and analyzed the subtle structural differences between the SFPQ/PSPC1 heterodimer and the previously characterized SFPQ homodimer. Analytical ultracentrifugation to estimate the dimerization equilibrium of the SFPQ-containing dimers revealed that the SFPQ-containing dimers dissociate at low micromolar concentrations and that the heterodimers have higher affinities than the homodimer. Moreover, we observed that the apparent dissociation constant for the SFPQ/PSPC1 heterodimer was over 6-fold lower than that of the SFPQ/NONO heterodimer. We propose that these differences in dimerization affinity may represent a potential mechanism by which PSPC1 at a lower relative cellular abundance can outcompete NONO to heterodimerize with SFPQ.

The Drosophila behavior human splicing (DBHS) 2 family of proteins are ubiquitous nuclear proteins implicated in many aspects of nuclear functions (1). Defined by the highly conserved DBHS region that consists of two tandem RNA recognition motifs (RRMs), a NonA/paraspeckle domain (NOPS) and a C-terminal coiled-coil, the members of this protein family have been reported as multifunctional proteins, playing important roles in RNA biogenesis and transport, and subnuclear body (paraspeckle) formation via direction interaction with structural noncoding RNA as well as DNA damage repair (1).
The DBHS family of proteins are exclusively found in the animal kingdom; invertebrate or lower vertebrate typically encode a single DBHS protein, whereas higher vertebrates encode more than one (2). In humans, three DBHS proteins: SFPQ, PSPC1, and NONO, share greater than 70% identity within the DBHS region, flanked by low complexity N-and C-terminal domains that vary in sequence and length among the three paralogues (Fig. 1). DBHS proteins are obligatory dimers; the three human DBHS proteins form either homodimer or heterodimers among the three proteins in vitro (3)(4)(5) and all three possible heterodimers (SFPQ/NONO, SFPQ/ PSPC1, and PSPC1/NONO) have been shown in mouse Sertoli cells (6).
The dimeric nature of the DBHS family of proteins has been confirmed by the crystal structures of human NONO/PSPC1 heterodimer (7), SFPQ homodimer (5), and Caenorhabditis elegans NONO-1 (2) in our previous studies. The crystal structures of the DBHS proteins consistently revealed an intimate dimer interface contributed by all four domains within the DBHS region. The dimer formation is largely driven by the hydrophobic interactions of the second RRM with the NOPS domain as well as the antiparallel coiled-coil interaction, whereas the first RRM is dispensable in dimer formation (2,5,7). Dimerization of the DBHS proteins is not only critical to their structural integrity but also to their functions such that dimerization incompetent DBHS proteins generated by mutation of key hydrophobic residues failed to form subnuclear bodies termed paraspeckles (7).
Combinatorial heterodimerization among the three DBHS proteins diversifies the functional repertoire and also contributes functional overlap and redundancy, evidenced by functional compensation of the loss of DBHS protein by the other paralogues. Specifically, NONO deficiency in murine embryonic fibroblasts derived from NONO knockout mice led to spontaneous up-regulation of PSPC1, which formed a complex with SFPQ in place of NONO and compensated the loss of NONO in DNA repair (8). Similar up-regulation of SFPQ and PSPC1 was observed in patients with syndromic intellectual disability with little or no NONO protein expression due to the mutations in NONO (9). The data suggest that up-regulation of SFPQ and PSPC1 may compensate the down-regulated NONO at least partially; however, the distinct contributions of NONO as in SFPQ/NONO or PSPC1/NONO are irreplaceable by SFPQ/PSPC1 in regulation of synaptic transcription, leading to intellectual disability (9). Taken together, these observations indicate that dimerization among the three DBHS paralogues is likely interdependent as well as spatially and temporally dynamic. Thus, different DBHS heterodimers may have different tissue-or cell-type specific functions, contributing to the myriad of nuclear functions assigned to the DBHS proteins. However, the relative affinity and mechanistic details of the combinatorial dimer states are still unclear.
As a first step toward better understanding the relative affinity and mechanism of combinatorial dimerization of the three human DBHS proteins, we have carried out analytical ultracentrifugation on the SFPQ containing dimers (SFPQ homodimer, SFPQ/PSPC1, and SFPQ/NONO). In addition, we have determined the crystal structure of SFPQ/PSPC1 and analyzed the structural differences of the SFPQ/PSPC1 heterodimer with the previously characterized SFPQ homodimer.
In this study, His-tagged PSPC1 (residues 61-320) was coexpressed with untagged SFPQ (residues 276 -535) and the resulting SFPQ/PSPC1 heterodimer was successfully purified, confirming that the three human DBHS proteins can form all three possible heterodimers. One of the important observations from the course of purification of all six possible dimeric complexes (three homodimers and three heterodimers) is that human DBHS proteins display a preference for heterodimerization; when two DBHS proteins are co-expressed, the less expressed protein is exclusively found in the heterodimer (supporting Fig. S1), indicating that dimerization affinities of het-erodimers are stronger than those of homodimers and/or heterodimers are more stable than homodimers.

Differential dissociation constants of SFPQ-containing dimers
As a first step toward characterizing the relative affinities for dimer formation among the SFPQ-containing dimers, we conducted sedimentation velocity experiments in the analytical ultracentrifuge comparing all three possible SFPQ-containing dimers: SFPQ homodimer, SFPQ/PSPC1, and SFPQ/NONO heterodimers at an initial total protein concentration of 5 M. All three dimers have the same construct boundaries defined by the dimerization domain (RRM1, RRM2, NOPS, and partial coiled-coil) (Fig. 1). It should be noted that the exclusion of the extended coiled-coil domain beyond the dimerization domain was necessary to avoid complication of data analysis because of oligomerization mediated by the extended coiled-coil (5).
The data from sedimentation velocity experiments were fitted to a continuous size (c(s)) distribution model and the results demonstrated that all three samples at an initial concentration of 5 M exist as two major species in solution with sedimentation coefficients of 2.3 and 3.8 S, consistent with monomer and dimer, respectively (Fig. 2, A-C, Table 1). There is also a small proportion of a higher order oligomer with a modal sedimentation coefficient of ϳ5.7 S, particularly in the SFPQ homodimer sample ( Fig. 2A). More importantly, the relative proportion of dimers and monomers differed in each sample with significantly more dimeric species present for the heterodimeric complexes, particularly SFPQ/PSPC1 (Fig. 2C), compared with the SFPQ homodimer, indicating that the dimerization affinities of the SFPQ-containing heterodimers are indeed stronger than that of the SFPQ homodimer. This assertion was confirmed by performing sedimentation equilibrium experiments at multiple rotor speeds and initial protein concentrations. The data were analyzed by a global nonlinear regression method and best fitted to a monomerdimer equilibrium for the SFPQ homodimer and to the 1:1 complex (i.e. AϩB N AB) for the SFPQ-containing heterodimers. The resulting apparent dissociation constants showed that the SFPQ/PSPC1 heterodimer (K D AB3 AϩB of 0.55 M) is almost 20-fold tighter than the SFPQ homodimer (K D 231 of 10.54 M) (Fig. 2, D-F, Table 1). By contrast, the SFPQ/NONO heterodimer (K D AB3 AϩB of 3.65 M) is ϳ3-fold tighter than the SFPQ homodimer (Table 1). A noticeable difference (Ͼ6-fold) in the apparent dissociation constant between the two SFPQ-containing heterodimers is also observed ( Table 1).

Crystal structure of a SFPQ/PSPC1 heterodimer
To gain insight into the preferential heterodimerization of human DBHS proteins, we have crystallized the SFPQ/PSPC1 heterodimer encompassing the dimerization domain (  Crystal structure of a SFPQ/PSPC1 heterodimer and supporting Fig. S2). The structure of the SFPQ/PSPC1 heterodimer was solved by molecular replacement, and refined to 2.3-Å resolution (Fig. 3). Despite high sequence similarity between the two proteins within the dimerization domain (74 and 93% sequence identity and similarity, respectively) (supporting Fig. S2), several key residues that are different in the two proteins allowed unambiguous assignment of SFPQ and PSPC1 (supporting Fig. S3). The final model consists of one heterodimer in the asymmetric unit.
Consistent with the previously characterized structures of human DBHS proteins, the structure of the SFPQ/PSPC1 heterodimer displays a pseudosymmetrical dimer, reflecting high sequence identity within the dimerization domain (Fig. 3). The second RRM domain (RRM2), NOPS, and coiled-coiled domain provide the major dimerization interface. Largely hydrophobic in nature, the interactions of the NOPS domain with the RRM2 and coiled-coil domain of the dimer partner are followed by the right-handed coiled-coil interactions between the two proteins.
The dimerization domain without the first RRM domain (RRM1) superposes well with each other (root mean square difference (r.m.s. deviation) of 1.05 Å for 146 common C␣ positions), whereas the r.m.s. deviation for the overall superposition of the two protein chains is 2.53 Å (224 C␣). The major devia- Plotted are the data at 12,000 rpm (circles) and 18,000 rpm (triangles) for each protein at an initial concentration of 1.5 M with solid lines representing the nonlinear best fits to homo-or heterodimerization equilibrium models. Insets, residuals plotted as a function of radial position for the best fits obtained at both rotor speeds. Note that data and fits shown are representative of global nonlinear least squares analyses employing multiple protein concentrations (0.5-15 M) and rotor speeds that yielded the dimerization dissociation constants reported in Table 1. Global reduced 2 values were Ͻ1.0 for all nonlinear best fits. a Molecular mass of a dimer based on amino acid composition. b Taken from the ordinate maximum of the relevant peak in the c(s) distribution best-fit ( Fig. 2). c Taken from the ordinate maximum of the relevant peak in the c(M) distribution best-fit (data and fits not shown). d Dimerization dissociation constant determined from global nonlinear least squares analyses of sedimentation equilibrium experiments at multiple speeds and initial protein concentrations (data and fits not shown). Estimated error was determined from Monte-Carlo error analysis of the nonlinear best-fit parameters (1000 iterations, 95% confidence).

Crystal structure of a SFPQ/PSPC1 heterodimer
tion from the 2-fold symmetry is due to the arrangement of the two N-terminal RRM1 domains relative to each other. Although the RRM1 domains from each protein superpose well individually (r.m.s. deviation of 0.685 Å, 77 C␣), the observed rotation angle of the two RRM1 domains is 170.4 o , breaking the 2-fold symmetry. In addition, the relative position of RRM1 in SFPQ to the dimer core is tilted by 27 o toward the dimer core compared with that of RRM1 of PSPC1, contributing to further deviation from the 2-fold symmetry. This observation is consistent with the structure of a PSPC1/NONO heterodimer (PDB code 3SDE), the first DBHS protein structure we have characterized, where similar pseudo-symmetry was observed with the major deviation from the 2-fold symmetry contributed by the dispositions of the two RRM1 domains (7). In contrast, in the structure of the SFPQ homodimer (PDB code 4WII), the two RRM1 domains are positioned in a more symmetrical way (rotation angle of 178.8 o ) with no significant deviation in the dispositions of each RRM1 to the dimer core (Fig. 4, A and B) (5). It is, therefore, not surprising that the use of the structure of PSPC/NONO (PDB code 3SDE) as a search model in phasing by molecular replacement found a straightforward structure solution, whereas a more complicated search strategy was required to produce readily interpretable electron density maps when the SFPQ homodimer (PDB code 4WII) was applied as a search model.

Structural comparison between the SFPQ/PSPC1 heterodimer and SFPQ homodimer
Due to high sequence similarity between SFPQ and PSPC1 within the dimerization domain (74 and 93% sequence identity and similarity, respectively), the overall structure of the SFPQ/ PSPC1 heterodimer is indistinguishable from that of the SFPQ homodimer (PDB code 4WII) (Fig. 4A) (5). Analysis of interface using PISA (11) shows the common principal interface involving all four domains in both structures with over 40% of residues directly involved in the dimer interface (supporting Table  S1). The principal dimer interface is largely driven by hydrophobic interactions evidenced by a highly negative gain of solvation energy on complex formation (⌬ i G ϭ Ϫ50.4 kcal/mol with p value 0.027 for the SFPQ/PSPC1 heterodimer; ⌬ i G ϭ Ϫ46.1 kcal/mol with p value 0.061 for the SFPQ homodimer) as calculated by PISA (11).
Closer inspections on the amino acid sequences of the two proteins and superposition of the structures of the SFPQ-containing dimers, however, reveal subtle differences in local primary and tertiary structures. Although most of the primary sequence variations are scattered throughout the dimerization domain, four regions that have a stretch of amino acids with sequence variations are observed (supporting Fig. 2A). With exception of Region 2, these regions are involved in dimer interfaces with the resulting local structures showing subtle differences in the two structures; Region 1, located prior to the RRM1, contributes to the asymmetry of the two RRM1 arrangements described in the previous section, whereas Regions 3 and 4 are located in the NOPS and C-terminal coiled-coil, respectively, where further local structural differences are observed (Fig. 4A). Although most of the intermolecular interactions in these regions made by networks of hydrogen bonds and salt  Table  S2) and detailed below.
Region 1 (residues 276 -287 in SFPQ; 61-80 in PSPC1), showing the most sequence divergence between SFPQ and PSPC1, takes on a different secondary structure. In the struc- Crystal structure of a SFPQ/PSPC1 heterodimer ture of the SFPQ homodimer (PDB code 4WII), this region forms a short ␣-helical structure that is not involved in the dimer interface (5). The corresponding residues of PSPC1 in the structure of the SFPQ/PSPC1 heterodimer form a strand with no secondary structure. Interestingly, an additional hydrophobic residue (Phe-70) in Region 1 of PSPC1 participates in the dimer interface between the two RRM1 domains in the heterodimer, contributing to the asymmetry in the arrangement of the RRM1 domains (Fig. 4, B and C). The different hydrophobic environment associated with this asymmetry at the interface of the two RRM1 domains may contribute toward favoring heterodimerization.
Region 3 (residues 464 -471 in SFPQ; 249 -256 in PSPC1), located in the middle of the NOPS domain, appears to be flexible with different conformations observed in the two structures (Fig. 4, A, D, and E). The electron density of Region 3 of SFPQ (residues 462-474) is in fact disordered in the structure of the SFPQ/PSPC1 heterodimer and thus the residues in this region are not modeled in the final structure (Fig. 4D). Side chains of the majority of the residues in Region 3 of PSPC1 and the SFPQ homodimer are found in the solvent-accessible area, thus not contributing to the dimer interface. However, the first residue of Region 3 (Ala-464 in SFPQ and Met-249 in PSPC1) is directly involved in the dimer interface through hydrophobic interaction with a residue from RRM2 of the dimer partner (Fig.  4, D and E). Interestingly, the partnered hydrophobic residue in RRM2 shows amino acid variation: Tyr-381 in SFPQ, whereas Val-166 in PSPC1. The resulting hydrophobic interaction in the heterodimer shows Met-249 of PSPC1 interacting with Tyr-381 from RRM2 of SFPQ, whereas Ala-464 in Region 3 of SFPQ (not observed in the SFPQ/PSPC1 structure) is likely to interact with Val-166 from RRM2 of PSPC1 when extrapolated from the structure of the SFPQ homodimer (Fig. 4, D and E).
Region 4 (residues 525-535 in SFPQ; 310 -320 in PSPC1) is located at the end of anti-parallel coiled-coil interface. Although all of the residues of Region 4 in PSPC are observed, residues beyond 527 of SFPQ in the heterodimer structures are not observed with a similar observation made in the SFPQ homodimer. In addition, the majority of the residues in this region belong to the beginning of the "coiled-coil interaction motif" (residues 528 -555 in SFPQ) characterized in our previous study, which mediates polymerization of the SFPQ dimers (5). Therefore, it is unlikely that the sequence variations in this region directly influence the dimer interface. Variations in these residues may have, if any, impacts on choice of polymerization partners among the DBHS dimers, which are yet to be characterized.
However, closer inspection of the dispositions of the distal coiled-coil domain reveals that highly conserved residues at the C-terminal end of the NOPS domain (Phe-486, Tyr-490, and Trp-494 for SFPQ; Phe-271, Tyr-275, and Trp-279 for PSPC1) take on drastically different side chain conformations in the two structures (Fig. 4, D-F). Although the structure of the SPFQ homodimer shows similar side chain conformations for these residues in each chain of the dimer (Fig. 4E), the side chains of the corresponding residues take on drastically different conformations in the heterodimer setting (Fig. 4, D and F). Importantly, two of these residues (Tyr-275 and Trp-279 for PSPC1 numbering) have been shown previously to be critical for DBHS dimerization; mutation of these residues to alanine in NONO and PSPC1 prevents homo-and heterodimer formation in yeast two-hybrid analyses (7). Similar observations on the different side chain conformations of these residues were also made in the structure of a NONO/PSPC1 structure. Thus, observation of different side chain conformations in the structure of the SFPQ/PSPC1 heterodimer further reinforces the idea that structural plasticity associated with these residues may influence choice of dimerization partner.
Due to the intimate interaction between NOPS and RRM2 of the partner DBHS protein (RRM2Ј), we further inspected the dimer interface between NOPS and RRM2Ј and found a cluster of hydrophobic interactions between NOPS and RRM2Ј that show differences in the amino acid identity (Fig. 4, G-I). The residues involved in this hydrophobic pocket belong to the RRM2Ј (Val-435 for SFPQ numbering) and the distal end of NOPS (Ala-481, His-483, Tyr-488, and Ser-491 for SFPQ numbering). This region of NOPS also harbors the three conserved aromatic residues mentioned above that show different side chain conformations. The two clusters of hydrophobic residues are found in the different faces of the helix, however; whereas the three conserved aromatic residues face the distal coiled-coil domain of the partner protein (CCЈ), two of the residues (Tyr-488, and Ser-491 for SFPQ numbering) involved in the latter hydrophobic pocket are located in the outer face of the helix with the other two residues (Ala-481 and His-483 for SFPQ numbering) sitting on the hinge region before the beginning of the helix. With the exception of the alanine residue (Ala-481 for SFPQ; Ala-266 PSPC1), the amino acid identity of the four residues involved in this hydrophobic pocket varies among the three DBHS proteins. In combination with the three conserved aromatic residues, the variations of amino acid sequence in this hydrophobic pocket may contribute to the apparent differences in the dissociation constants of the homodimer and heterodimers.
In summary, it appears that preferential heterodimerization of the SFPQ-containing dimers is governed by synergistic effects of subtle variations in the local structures observed in the dimer interfaces encompassing all four domains (two RRMs, NOPS, and coiled-coil) within the dimerization domain.

Discussion
Since the first identification and characterization of members of the DBHS protein family in the early 1990s, a myriad of nuclear functions have been designated to this family of proteins (1). Although exhibiting a large overlap of functions, the three human DBHS paralogues, SFPQ, PSPC1, and NONO, are incapable of complete functional compensation after loss of the other members in some cases (9). This suggests specific roles for each paralogue in different temporal and spatial contexts, presumably acting through the divergent N and C termini flanking the conserved DBHS domain. It has been established that the functional unit of the DBHS proteins is a dimer (3)(4)(5)7); the three paralogues, a product of evolutionary duplication and divergence of an ancestral gene, homoand heterodimerize with each other to add extra layers to the Crystal structure of a SFPQ/PSPC1 heterodimer functional complexities of the DBHS proteins in higher vertebrates.
In this study, built on our previous studies where preference to heterodimerization has been observed from the co-expression and purification of the three human DBHS proteins using a bacterial expression system, we confirmed through sedimentation analyses that the dimerization affinity for the SFPQ-containing heterodimers is indeed tighter than that of the SFPQ homodimer. Surprisingly, the apparent dissociation constant for the SFPQ/PSPC1 heterodimer is more than 19-fold lower than that of the SFPQ homodimer, whereas the SFPQ/NONO heterodimer is ϳ3-fold tighter than the SFPQ-homodimer, suggesting that the majority of DBHS proteins are most likely to exist as heterodimers in vivo. This imposes significant implications in interpretation of the past literature on the human DBHS proteins where individual proteins have been addressed separately to probe their nuclear function. Thus, future work may need to revisit and validate the roles assigned to each individual paralogue taking into account their dimerization states.
The observed 19-fold difference in the apparent dissociation constant between the SFPQ/PSPC1 heterodimer and the SFPQ homodimer was unexpected given that the two paralogues share high sequence similarity within the dimerization domain. Reflecting this similarity, the structural determinants for preferential heterodimerization of SFPQ/PSPC1 compared with the homodimeric SFPQ cannot be attributed to a few key residues; contributed by all four domains (RRM1, RRM2, NOPS, and coiled-coil domain), synergistic effects from small variations in the identity or conformation of many residues largely in the hydrophobic pockets of the dimerization interface appear to favor heterodimerization.
Intrinsic differences in dimerization affinity among all six possible human DBHS dimers are most likely to be the major determinant of the dimerization partner choice in vivo. However, other factors including the differential expression level of the three paralogues and post-translational modifications may also contribute to the equilibrium and dynamics of homo-and heterodimerization in vivo.
Although all three paralogues are ubiquitously expressed in many cell types, relative expression levels are different; SFPQ is highly expressed across all tissue types, whereas varied expression levels of NONO and PSPC1 are observed in the Human Protein Atlas (www.proteinatlas.org) 3 (12). In particular, the expression level of PSPC1 is highly tissue-and cell-type specific; very low expression of PSPC1 is observed in heart muscle and ovary tissue. In HeLa cells, PSPC1 is much less abundant than SFPQ and NONO (6), whereas a similar expression level of PSPC1 was observed in mouse Sertoli cell (13). In addition, the presence of multiple transcript variants producing isoforms of each paralogue adds extra layers of complexities in the combinatorial dimers of the three DBHS paralogues. Some of these variants lack a nuclear localization signal located at the C terminus (14, 15), suggesting that localization of DBHS proteins can be altered depending on the identity and isoform of the dimerization partner. This is particularly relevant with the emerging function and importance of the cytoplasmic pool of DBHS family proteins in neuronal development. Localization of the DBHS family proteins is predominantly nuclear; however, recent studies have shown that a non-nuclear pool of SFPQ is essential for axon motor development and viability (16,17).
The effects of post-translational modifications on dimerization of DBHS proteins are yet to be characterized; however, various post-translational modifications on DBHS proteins including phosphorylation and methylation have been shown to alter their nuclear activities and localization of the proteins from the nucleus to the cytoplasm in some cases (18 -20). It is plausible that the altered cellular activities and localization of DBHS proteins may directly reflect altered dimerization due to post-translational modifications.
In summary, we demonstrate that the apparent dissociation constants of the SFPQ-containing heterodimers are lower than that of the SFPQ homodimer, confirming that heterodimerization is indeed preferred to homodimerization among the SFPQ-containing dimers. A noticeable difference (Ͼ6-fold) in the apparent dissociation constant between the SFPQ/PSPC1 and SPFQ/NONO heterodimers is also observed. These differences in dimerization affinity may represent a potential mechanism by which less expressed PSPC1 can outcompete more abundant NONO to heterodimerize with the most abundant SFPQ in many cell and tissue types. Despite the noticeable difference in the apparent dissociation constants, very subtle differences in the dimerization interfaces are observed in the crystal structure of the SFPQ/PSPC1 heterodimer compared with that of the SFPQ homodimer, reinforcing the concept that the synergistic effects of subtle variations in the sequence and local structures result in stronger affinity in the DBHS heterodimers.

Protein expression and purification
For coexpression and purification of the SPFQ/PSPC1 heterodimer, pETDuet-1-PSPC1(61-320) and pCDF-13-SFPQ(276 -535) were co-transformed into Rosetta2 (DE3) (Novagen). The cells were grown in LB growth medium supplemented with 100 g ml Ϫ1 of ampicillin, 50 g ml Ϫ1 of spectinomycin, and 50 g ml Ϫ1 of chloramphenicol and protein expression was induced with isopropyl 1-thio-␤-D-galactopyranoside (0.5 mM) at an OD of 0.8 -1.0. The cells were harvested after further incubation at 25°C for 16 h. The cell pellet was suspended in buffer A (20 mM Tris-HCl (pH 7.5), 250 mM NaCl, 10% glycerol) and disrupted using a TS series benchtop cell disruptor (Constant Systems Ltd.). The soluble fraction was applied to a nickel-chelating column (Hi-Trap HP, GE Healthcare) and the bound protein was eluted by applying a linear gradient of 25-500 mM imidazole in buffer A. Fractions containing His-PSPC1/SFPQ were pooled and subjected to TEV protease digestion with 1:100 ratio (w/w) at 4°C overnight. Uncleaved heterodimer and His-TEV protease were removed by the second nickel-affinity chromatography. Flow-through fractions containing the cleaved PSPC1/SFPQ heterodimer were concentrated and applied to a HiLoad 16/600 Superdex 200 size exclusion column, pre-equilibrated with 20 mM Tris-HCl (pH 7.5), 0.25 M NaCl. Expression and purification of the SFPQ homodimer (residues 276 -535) and SFPQ(276 -535)/ NONO(53-312) heterodimer was described previously (5). The purified proteins were typically concentrated to 7-10 mg/ml and stored at Ϫ80°C until further use.

Sedimentation velocity
Sedimentation velocity experiments were conducted at 20°C using a Beckman Coulter XL-A or XL-I analytical ultracentrifuge with an An-50 Ti 8-hole rotor using similar methods reported previously (21,22). Briefly, double-sector quartz cells were loaded with 380 l of protein samples at an initial concentration of 5 M and 400 l of reference buffer (20 mM Tris-HCl (pH 7.5), 500 mM NaCl). Data were collected at 230 nm and rotor speed of 40,000 rpm in continuous mode using a using a step size of 0.003 cm without averaging. The partial specific volume of SFPQ, SFPQ/NONO, and SFPQ/PSPC1, buffer density, and buffer viscosity were computed using SEDNTERP (23). Data were analyzed using the continuous size-distribution model employing the program SEDFIT (24 -27) with a continuous size distribution model.

Sedimentation equilibrium
Sedimentation equilibrium experiments were performed in a Beckman Coulter XL-I analytical ultracentrifuge in an An-50 Ti 8-hole rotor employing similar methods reported previously (21,22). Briefly, 120 l of protein samples (0.5-15 M) and 140 l of reference buffer (20 mM Tris-HCl (pH 7.5), 500 mM NaCl) were centrifuged at rotor speeds of 12,000 and 18,000 rpm until sedimentation equilibrium was attained (t ϭ 36 and 48 h for each speed). Experiments were carried out using a step size of 0.001 cm with 10 averages at a wavelength optimized at 234 nm. Data were fitted to a monomer-dimer equilibrium model using SEDPHAT (21,22,28).

Crystallization and X-ray diffraction data collection
Crystals of the SFPQ/PSPC1 heterodimer were grown by hanging-drop vapor diffusion experiments at 20°C. 1.5 l of SFPQ/PSPC1 (10 mg/ml) was mixed with 1.5 l of reservoir solution (50 mM MES (pH 5.5), 0.2 M NaCl, 3% (w/v) aminohexanoic acid, 25% (w/v) PEG3350) and equilibrated against 0.5 ml of the reservoir solution. Crystals were transferred briefly to a drop containing artificial reservoir solution with 30% (w/v) glucose before cryocooling. Diffraction data were recorded on beamline MX2 at the Australian Synchrotron at a wavelength of 0.954 Å at 100 K. The data were processed with XDS (29), and merged and scaled with AIMLESS (30). Data collection and merging statistics are summarized in Table 2.

Structure solution and refinement
The crystal structure of the SFPQ/PSPC1 heterodimer was solved by molecular replacement using PHASER (31) within the CCP4 suite (32). The structure of the PSPC1/NONO heterodimer (PDB code 3SDE (7)) was used as search model after removing all nonprotein atoms. The structure of PSPC1/NONO found a solution with a log likelihood gain of 792 and a Z score of 25.3, locating one dimer in the asymmetric unit. Difference Fourier maps (F o Ϫ F c ) from the first round of refinement with auto-BUSTER (33) revealed several key residues that are different among SFPQ, PSPC1, and NONO, which allowed unambiguous assignment of SFPQ and PSPC1. At this stage, the correct amino acid sequence was incorporated, and iterative model building with COOT (34) and refinement with autoBUSTER was carried out. The final model consists of two chains with SFPQ in Chain A (residues 290 -461 and 475-527) and PSPC1 in Chain B (residues 66 -320), and 136 water molecules. The quality of the model was validated using MOLPROBITY (35). The refinement statistics are included in Table 2. The atomic coordinates have been deposited in the Protein Data Bank as entry 5WPA.