Specific Protein Domains Mediate Cooperative Assembly of HuR Oligomers on AU-rich mRNA-destabilizing Sequences*

The RNA-binding factor HuR is a ubiquitously expressed member of the Hu protein family that binds and stabilizes mRNAs containing AU-rich elements (AREs). Hu proteins share a common domain organization of two tandemly arrayed RNA recognition motifs (RRMs) near the N terminus, followed by a basic hinge domain and a third RRM near the C terminus. In this study, we engineered recombinant wild-type and mutant HuR proteins lacking affinity tags to characterize their ARE-binding properties. Using combinations of electrophoretic mobility shift and fluorescence anisotropy-based binding assays, we show that HuR can bind ARE substrates as small as 13 nucleotides with low nanomolar affinity, but forms cooperative oligomeric protein complexes on ARE substrates of at least 18 nucleotides in length. Analyses of deletion mutant proteins indicated that RRM3 does not contribute to high affinity recognition of ARE substrates, but is required for cooperative assembly of HuR oligomers on RNA. Finally, the hinge domain between RRM2 and RRM3 contributes significant binding energy to HuR·ARE complex formation in an ARE length-dependent manner. The hinge does not enhance RNA-binding activity by increased ion pair formation despite extensive positive charge within this region, and it does not thermodynamically stabilize protein folding. Together, the results define distinct roles for the HuR hinge and RRM3 domains in formation of cooperative HuR·ARE complexes in solution.

Several mRNAs that are rapidly degraded, often with halflives of less than 1 h, include those encoding oncoproteins and signaling proteins such as cytokines, chemokines, and inflammatory mediators (3). A common feature of many rapidly degraded mRNAs is the presence of an AU-rich element (ARE) 2 in their 3Ј-untranslated regions. The sequence of this cis-element is highly variable, although it frequently contains one or more AUUUA pentameric motifs within or near a U-rich region. The length of an ARE is also variable and can range from 30 to 120 nucleotides (4). AREs serve as binding sites for a variety of nuclear and cytoplasmic proteins, Ͼ20 of which have been identified to date (3,5). However, interactions between ARE-binding proteins and their cognate RNA targets can yield a diverse range of consequences. For example, AUF1, tristetraprolin, and K homology splicing regulatory protein (KSRP) can direct rapid decay of mRNAs containing high affinity binding sites (6 -10). Conversely, members of the Hu family of proteins inhibit degradation of many ARE-containing mRNAs, possibly by antagonizing recruitment of competing mRNA-destabilizing factors (11)(12)(13)(14). Finally, proteins of the TIA-1/TIAR family do not appear to influence mRNA decay kinetics directly; rather, they control the subcellular localization and translational efficiency of targeted transcripts (15,16).
The four mammalian members of the Hu family of AREbinding, mRNA-stabilizing proteins share significant sequence similarity with the Drosophila RNA-binding protein ELAV (embryonic lethal abnormal vision) (17). HuR, alternatively called HuA, is the only ubiquitously expressed protein of the four (18). Expression of the remaining three family members (HuB (sometimes referred to as Hel-N1), HuC, and HuD) are limited to neuronal tissues (17,19). The primary structures of the Hu proteins are well conserved (Ͼ68% pairwise identity among all proteins) and share an identical domain arrangement, with two tandemly arrayed RNA recognition motifs (RRMs) near the N terminus, followed by a hinge region and finally a third RRM near the C terminus (18). To date, characterization of the RNA-binding activity of the Hu proteins has been largely limited to the neuronal isoforms. Binding studies with HuD deletion mutants indicated that RRM1 is essential for RNA binding, but that high affinity association with ARE substrates also requires RRM2 or RRM3 (20). Crystal structures of an HuD deletion mutant containing the two N-terminal RRMs bound to ARE fragments show that both of these RRMs specifically contact poly(U) sequences within RNA substrates, spanning a total of eight to nine nucleotides and inducing a pronounced kink in the RNA backbone between sequences contacted by RRM1 versus RRM2 (21). An NMR structure of the corresponding domain of HuC bound to an ARE fragment yielded similar data (22). However, the function of the C-terminal RRM domain is much less clear. A HuD binding study indicated ARE-binding activity for RRM3, although inclusion of this domain made relatively minor contributions (2-fold) to binding affinity (20). By contrast, an earlier report indicated that HuR RRM3 facilitates association with poly(A) RNA sequences (23). The cellular role of RRM3 also remains unresolved. Overexpression studies using HuR truncation mutant proteins have alternatively demonstrated that RRM3 is either required or dispensable for the mRNA-stabilizing activity of this protein, although these contradictory results may be cell type-specific (13,24). Finally, the basic hinge region located between RRM2 and RRM3 contains a nuclear localization sequence that allows HuR to shuttle between the cytoplasm and the nucleus, although the protein is predominantly nuclear (25)(26)(27). To date, the hinge domain has not been implicated in the association of Hu proteins with RNA substrates.
In this study, we constructed recombinant wild-type and mutant HuR proteins lacking affinity tags and characterized their RNA-binding properties using combinations of electrophoretic mobility shift assays (EMSAs) and fluorescence anisotropy-based binding experiments. The use of differentially sized RNA substrates spanning the ARE from tumor necrosis factor ␣ (TNF␣) mRNA permitted resolution of the minimal RNA-binding sites necessary for HuR recognition and cooperative binding. Together, these experiments identify RNA and protein determinants contributing to cooperative formation of multisubunit HuR complexes on ARE substrates and define specific roles for both the basic hinge and C-terminal RRM domains in this process.

EXPERIMENTAL PROCEDURES
RNA Substrates-Synthesis, 2Ј-hydroxyl deprotection, and purification of the RNA substrates used in this study were performed by Dharmacon Research or Integrated DNA Technologies. The RNA substrate ARE 38 includes the core ARE sequence from TNF␣ mRNA (Table 1). Other substrates designated ARE xx are subsections of this TNF␣ mRNA sequence. The RNA substrate R␤ encodes a fragment of the rabbit ␤-globin mRNA coding sequence. Fluorescein groups were linked to the 5Ј-ends of some RNA substrates during solid-phase synthesis and are designated by the prefix "Fl" where applicable. Lyophilized RNA samples were resuspended in 10 mM Tris-HCl (pH 8). Quantification of RNA yields and fluorophore labeling efficiency was performed by absorbance spectroscopy as described previously (28,29). For preparation of 5Ј-32 P-radiolabeled RNA substrates, 5Ј-hydroxyl RNA oligonucleotides were incubated with [␥-32 P]ATP (PerkinElmer Life Sciences) and T4 polynucleotide kinase (Promega) as described (30), yielding specific activities of 3-5 ϫ 10 3 cpm/fmol.

Construction of Recombinant HuR and HuR Deletion Mutant
Expression Vectors-A cDNA fragment encoding the complete open reading frame of human HuR was amplified by reverse transcription-PCR using RNA purified from the monocytic leukemia cell line THP-1 and then subcloned into pGEM-7Z(ϩ) (Promega) to generate plasmid pGW01. The fidelity of the HuR open reading frame was verified by automated DNA sequencing. HuR cDNA subfragments encoding the complete 306-amino acid open reading frame, residues 1-185 (HuR⌬Hinge3), and residues 1-242 (HuR⌬RRM3) were amplified from the pGW01 template by PCR utilizing DNA primers incorporating unique terminal restriction sites. These cDNA fragments were then subcloned downstream (HuR and HuR⌬Hinge3) or upstream (HuR⌬RRM3) of an intein/chitinbinding domain tag using plasmids pTYB11 and pTYB1, respectively, from the IMPACT-CN (intein-mediated purification with an affinity chitin-binding tag) system (New England Biolabs). The sequence and orientation of all cDNA inserts were verified by automated DNA sequencing.
Preparation and Characterization of Recombinant Proteins-Recombinant HuR or HuR deletion mutant proteins (see Fig.  1A) were prepared in Escherichia coli ER2566 cells using a variation of a procedure described by Meisner et al. (31). Transformed cells were grown with shaking at 37°C in SOB medium supplemented with 10 mM MgCl 2 and 50 g/ml ampicillin until reaching A 600 Ϸ 0.6 -0.8. Protein production was induced by addition of isopropyl 1-thio-␤-D-galactopyranoside (1 mM), and cell growth was continued at 25°C for 5 h. Cells were recovered by centrifugation at 3300 ϫ g for 20 min. In pilot experiments, optimal protein buffer conditions were selected through a solubility screening test monitored by dynamic light scattering using a Zetasizer Nano series instrument (Malvern Instruments) essentially as described (32). This procedure indicated optimal solubility for HuR and HuR deletion mutant proteins in a high salt buffer composed of sodium phosphate (pH 7.0), 500 mM NaCl, and 1 mM EDTA (data not shown), which was subsequently termed HuR column buffer.
Cells were resuspended in 15 ml of chilled HuR column buffer containing 20 M phenylmethanesulfonyl fluoride and disrupted on ice by sonication. All subsequent purification steps were performed at 4°C. Cell lysates were clarified by centrifugation at 20,000 ϫ g for 20 min. The supernatant was loaded onto a chitin affinity column (New England Biolabs) equilibrated with HuR column buffer. The column was washed with 10 bed volumes of HuR column buffer to remove unbound or nonspecifically bound proteins, followed by addition of 3 bed volumes of cleavage buffer (HuR column buffer containing 50  20 UUUAUUAUUUAUUUAUUUAG ARE 18 UUAUUAUUUAUUUAUUUA ARE 16 AUUAUUUAUUUAUUUA ARE 15 UUAUUUAUUUAUUUA ARE 14 UAUUUAUUUAUUUA ARE 13 AUUUAUUUAUUUA ARE 11 UUAUUUAUUUA R␤ UGGCCAAUGCCCUGGCUCACAAAUACCACUG mM dithiothreitol (DTT)). The column was then capped, and DTT-induced self-cleavage of the intein tag was allowed to proceed for 40 h. Cleaved protein was eluted in HuR column buffer and concentrated using an Amicon Ultra-4 concentrator (10-kDa molecular mass cutoff). DTT remaining in the sample was removed by repeated concentration in HuR column buffer before a final centrifugation step (20,000 ϫ g for 15 min) to remove any particulate matter. Recombinant proteins were quantified using Coomassie Blue-stained SDS-polyacrylamide gels against a titration of bovine serum albumin. The purity of all recombinant proteins was judged to be Ͼ95% by Coomassie Blue-stained SDS-PAGE (see Fig. 1B, left panel). Western blot analyses using anti-HuR antibodies (Santa Cruz Biotechnology, Inc.) further verified that each recombinant protein was immunologically related to HuR (see Fig. 1B, right panel). UV absorbance spectra indicated no significant nucleic acid contamination of HuR⌬RRM3 or HuR⌬Hinge3 (supplemental Fig. S1A). By contrast, absorbance of the fulllength HuR protein was enhanced at wavelengths below 270 nm, characteristic of co-purifying nucleic acids. Comparisons with HuR spectra measured following addition of varying amounts of an RNA substrate containing a single HuR-binding site (ARE 16 ) indicated that the total number of nucleic acidbinding sites co-purifying with recombinant HuR constituted Ͻ0.1 mol eq (supplemental Fig. S1B).
Dynamic light scattering verified that recombinant protein preparations were free of high molecular mass complexes or aggregates. Typically, protein preparations returned peaks representing particle diameters of 5.1-7.5 nm, constituting Ͼ98% of the total sample volume (see Fig. 1C). Finally, the oligomerization status of purified recombinant proteins was monitored by gel filtration chromatography (see Fig. 1D). A HiPrep 16/60 column (GE Healthcare) packed with Sephacryl S-200 high resolution resin was equilibrated with HuR column buffer before loading protein samples (50 -100 g). Chromatography was performed at 30 ml/h with protein elution monitored by absorbance at 280 nm. Column void volume was determined using blue dextran. The column was calibrated by monitoring elution of the following protein standards (Sigma) at A 280 : alcohol dehydrogenase (150 kDa), bovine serum albumin (66 kDa), carbonic anhydrase (29 kDa), and cytochrome c (12.4 kDa). The apparent molecular masses calculated for each of the HuR, HuR⌬RRM3, and HuR⌬Hinge3 proteins resolved to within 10% of their predicted molecular masses calculated from amino acid sequence (supplemental Table SI), indicating that each recombinant HuR protein is a monomer in solution.
Analyses of Protein Folding Stability by Chemical Denaturation-For all recombinant proteins, the thermodynamic stability of protein folding was estimated by equilibrium denaturation in guanidine HCl (GdnHCl). Protein samples (2 M) were incubated at room temperature in RNA binding buffer (10 mM Tris-HCl (pH 8) containing 50 mM KCl, 2 mM DTT, and 0.5 mM EDTA) in the absence or presence of varying concentrations of GdnHCl (60-l total volume). After 60 min, the extent of protein unfolding was assessed by measurement of protein fluorescence ( ex ϭ 270 nm and em ϭ 290 -400 nm; 10 nm bandwidth) at 21°C using a Cary Eclipse spectrofluorometer equipped with a microcell. Because the HuR⌬RRM3 and HuR⌬Hinge3 proteins do not contain tryptophan, the excitation wavelength was blue-shifted to 270 nm to preferentially excite tyrosine residues (33). Protein denaturation was considered as a two-state, GdnHCl-dependent transition between native and unfolded conformations, yielding distinct fluorescence emission intensities, F native and F unfolded , respectively. Thermodynamic parameters describing the denaturant-induced protein unfolding transition were estimated from the change in protein fluorescence measured at 305 nm (F 305 ) as a function of GdnHCl concentration using Equations 1 and 2 adapted from the linear extrapolation method of Santoro and Bolen (34) as modified by Manyusa and Whitford (35).
Here, ⌬G u represents the free energy of protein denaturation at each concentration of GdnHCl. ⌬G uw is the extrapolated free energy of protein unfolding in the absence of denaturant, and m eq relates the sensitivity of ⌬G u to GdnHCl concentration. All parameters were resolved from F 305 versus [GdnHCl] plots by nonlinear regression using GraphPad Prism Version 3.03.
Protein⅐RNA Binding Assays-EMSAs were used to qualitatively assess the formation of complexes between recombinant HuR proteins and 32 P-labeled RNA substrates. EMSAs were performed essentially as described (30) but with EDTA (0.5 mM) instead of MgCl 2 in the binding buffer. Quantitative measurement of protein⅐RNA binding equilibria was performed using fluorescence anisotropy essentially as described (28,36). Reactions containing limiting concentrations of 5Ј-Fl-labeled RNA substrates (0.2 nM) and varying concentrations of protein were assembled in 10 mM Tris-HCl (pH 8) containing 50 mM KCl, 2 mM DTT, 0.5 mM EDTA, and 0.1 g/l acetylated bovine serum albumin (100-l final volume). Alternative monovalent salt or RNA substrate concentrations are indicated where applicable. Heparin (1 g/l) was also included in all binding reactions to suppress nonspecific interactions between proteins and RNA substrates (30). In the absence of this polyanionic competitor, HuR displayed modest binding to irrelevant RNA substrates (data not shown). Following incubation at 25°C for 60 min, fluorescence anisotropy of the Fl-RNA substrates was measured using a Beacon 2000 fluorescence polarization system (PanVera Corp.) equipped with fluorescein excitation (490 nm) and emission (535 nm) filters. Total fluorescence emission from each reaction was concomitantly measured to verify that the fluorescence quantum yields of RNA substrates were not significantly affected by protein binding events. Equilibrium binding constants describing formation of protein⅐RNA complexes were resolved from plots of total measured anisotropy versus protein concentration by nonlinear regression using Prism and employing analytical functions describing one or more of the binding models defined below (equations derived from Refs. 36 and 37). The appropriateness of each binding model was evaluated by the coefficient of determination (R 2 ) from individual binding experiments and analysis of residual plot non-randomness to detect any bias for data subsets (Prism). When multiple models were considered for a common data set, pairwise comparisons of sum-of-squares deviations were performed using the F test (Prism), with differences exhibiting p Ͻ 0.05 considered significant.
Reactions best described by cooperative binding were resolved by a variant of the Hill model. Under conditions of constant fluorescence quantum yield and limiting RNA substrate, measured anisotropy (A t ) varies with total protein concentration ([P]) as a function of the protein concentration yielding half-maximal binding ([P]1 ⁄ 2 ) and the Hill coefficient (h) by Equation 3.
Here, A R represents the intrinsic anisotropy of the free Fl-RNA substrate measured directly from sample reactions assembled without protein (n Ն 4), and A PxR represents the intrinsic anisotropy of the saturated HuR⅐Fl-RNA complex. In some cases, assembly of protein⅐RNA complexes involving two distinct protein binding events without cooperativity was indicated. Here, a sequential binding model given by Equation 4 resolves explicit equilibrium association binding constants for each stage of complex assembly provided that (i) the affinity of the first binding step (K 1 ) is at least 5-fold greater than that of the second (K 2 ), (ii) the concentration of the Fl-RNA substrate is significantly lower than 1/K 1 , and (iii) the fluorescence quantum yield of the Fl-RNA substrate is not significantly influenced by protein binding (36,37). Additional resolved parameters are the intrinsic anisotropy values of the Fl-RNA substrate bound to one (A PR ) or two (A P2R ) protein molecules.
If a single or multiple identical binding events are indicated, Equation 4 may be simplified by substitution of K 2 ϭ 0 to yield Equation 5.
In cases in which the apparent ARE-binding affinity was very high (1/K Ͻ 1 nM), data sets were also resolved by Equation 6. This function describes a quadratic binding algorithm incorporating protein depletion and includes terms for the total concentrations of RNA ([R] T ) and protein ([P] T ).
Where indicated, the free energy of protein binding was extracted from equilibrium association constants as ⌬G ϭ ϪRT ln(K), where R ϭ 1.987 ϫ 10 Ϫ3 kcal mol Ϫ1 K Ϫ1 .
HuR Binding Density Analysis-The size of the HuR-binding site on the Fl-ARE 38 substrate was calculated using the modelindependent ligand binding density analysis approach of Lohman and Bujalowski (38) and Bujalowski and Jezewska (39). Briefly, quantitative estimates of the average binding density (⌺, defined as the average number of HuR proteins bound per Fl-labeled RNA) and free HuR protein concentration ([P] f ) were extracted from the measured fluorescence anisotropy of the Fl-ARE 38 substrate across a titration of HuR. The analysis holds that each possible HuR complex with the Fl-ARE 38 RNA affects the experimentally observable net change in anisotropy (⌬A ϭ A x Ϫ A R , where A R is the intrinsic anisotropy of the Fl-ARE 38 substrate in the absence of protein and A x is the anisotropy in the presence of x nM HuR). As such, each HuR⅐Fl-ARE 38 complex displays an intrinsic anisotropy (⌬A i ) that contributes to the population weighted ensemble average (⌬A obs

Wild-type HuR Forms a Cooperative Oligomeric
Complex on the TNF␣ ARE 38 RNA Substrate-Previously, we (40) and others (18,41) employed glutathione S-transferase (GST)-tagged versions of HuR to characterize its RNA-binding properties. However, several features of this protein raised concerns that its RNA-binding activ- JULY 20, 2007 • VOLUME 282 • NUMBER 29 ity might not accurately reflect functions of cellular HuR. GST-HuR binding to an ARE substrate exhibited significant cooperativity in fluorescence anisotropy-based binding assays (40), which had not been reported previously for HuR or any other Hu family member. Given that GST domains are capable of self-association (42), it was also possible that free energy from GST-GST contacts could be responsible for the apparent cooperativity of GST-HuR binding to an ARE substrate by enhancing the second protein binding step. Conversely, the GST tag could sterically occlude adjacent binding sites on some RNA substrates because this moiety contributes 26 kDa to the chimeric protein.

Assembly of HuR Oligomers on RNA
To address these issues, we adapted a procedure by Meisner et al. (31) to produce untagged versions of full-length HuR or deletion variants lacking the C-terminal RRM domain but either retaining (HuR⌬RRM3) or lacking (HuR⌬Hinge3) the hinge domain (Fig. 1A). As determined by EMSA, the untagged wild-type HuR protein bound specifically to the ARE 38 substrate because no HuR⅐RNA complexes were formed with the R␤ substrate, which lacks an ARE sequence ( Fig. 2A). Interestingly, the association of HuR with the ARE 38 substrate yielded three distinct complexes at protein concentrations between 2 and 5 nM. Because HuR exists as a monomer in solution, the three binding events between HuR and ARE 38 likely indicate three separate monomers binding to the RNA substrate.
Weakly resolved bands migrating slightly ahead of the largest HuR⅐ARE 38 complex at 20 nM protein may indicate additional protein binding events, but more likely result from complex dissociation during electrophoresis. Alternatively, this band may represent protein binding to a small proportion of degraded RNA (43). Although only two complexes were detected in EMSAs using GST-HuR (40), we cannot rule out the possibility that the size of the GST tag may occlude a third binding site on the RNA or that additional GST-HuR⅐ARE complexes are too dynamic to be retained during gel loading and/or electrophoresis.
Fluorescence anisotropy-based binding assays were used to quantitatively characterize the association between HuR and the Fl-ARE 38 RNA substrate. Total fluorescence emission from the Fl-ARE 38 substrate did not significantly vary as a function of HuR concentration (Fig.  2B, upper panel), indicating that HuR binding does not influence the quantum yield of the fluorophore and validating assumptions required to derive analytical solutions for binding models (37). The increase in fluorescence anisotropy of the Fl-ARE 38 substrate as a function of HuR concentration was best resolved by the cooperative binding model described by Equation 3 (Fig. 2B, middle panel) based on random distribution of residuals (lower panel) and high coefficients of determination (R 2 Ͼ 0.99 across each of six independent experiments). By comparison, these data were not resolved by Equation 5, which could describe a single-site binding model or multiple identical but non-interacting HuR-binding sites on the Fl-ARE 38 substrate (Fig. 2B, middle panel, dotted line). The inappropriateness of non-cooperative models was indicated by significant residual non-randomness (p ϭ 0.0027) and by significant increases in the sum-of-squares deviation when compared with the cooperative binding model using the F test (p Ͻ 0.0001). Cooperative HuR binding to the ARE 38 substrate is also qualitatively supported by EMSA data (Fig. 2A) because the two fastest migrating HuR⅐RNA complexes accumulated concomitantly as a function of protein concentration rather than sequentially. Assembly of HuR oligomers on the Fl-ARE 38 substrate is a positively cooperative process because resolution of anisotropy data to Equation 3 yielded a Hill coefficient significantly greater than unity ( Table 2). The concentration of HuR conferring 50% binding saturation under these conditions was 2.0 Ϯ 0.3 nM, indicating that HuR targets the TNF␣ ARE with an affinity similar to that of the ARE-binding protein p37 AUF1 and the RNA-binding domain of tristetraprolin (40,44). Finally and consistent with the EMSA data, the fluorescence anisotropy assays also support specific binding of HuR to AU-rich RNA substrates because the presence of HuR did not significantly influence the anisotropy of a fluorescent RNA substrate lacking AU-rich sequences (Fl-R␤) (Fig. 2B, middle panel, open  circles).
ARE Substrate Length Requirements for Cooperative Assembly of HuR Oligomers-Model-independent ligand binding density analysis (38, 39) was used to determine the occluded site size of HuR binding to RNA substrates. By this method, the site size is calculated from an estimate of the maximal number of HuR proteins associated with a given RNA substrate at binding saturation. Association of HuR with the Fl-ARE 38 substrate was monitored by the fluorescence anisotropy-based binding assay across a range of RNA concentrations. As the RNA substrate concentration ([R] T ) was increased, greater concentrations of total HuR ([P] T ) were required to obtain equivalent net changes in measured anisotropy (⌬A obs ) (Fig. 3A). The population weighted average binding density (⌺) was calculated from the linear dependence of interpolated [P] T /[R] T pairs obtained at constant ⌬A obs as described under "Experimental Procedures." ⌺ was resolved across 80 distinct values of ⌬A obs spanning the lower 75% of the anisotropy isotherms to minimize the inherent inaccuracies in the derivation of this function near saturating protein concentrations reported by Bujalowski and Jezewska (45). Linear extrapolation of ⌺ to the maximal ⌬A obs (⌬A max ) of 0.119 Ϯ 0.003 (Fig. 3B) Fig. S2).
As an orthogonal approach to define minimal RNA substrate requirements for cooperative HuR binding, an extensive series of ARE 38 truncation mutants were employed for both qualitative and quantitative in vitro binding studies (Table 1). By EMSA, multiple HuR binding events were not observed on RNA substrates Ͻ18 nucleotides in length. Whereas two  ). All resolved constants are expressed as the means Ϯ nϪ1 for n Ն three or the means Ϯ spread for n ϭ two independent experiments. c ND, not determined.
HuR⅐RNA complexes were readily assembled on the ARE 20 and ARE 18 substrates, consistent with the HuR site size of eight to nine nucleotides resolved by binding density analysis, only a single protein⅐RNA complex was detectable with either the ARE 16 (Fig. 4A) or ARE 13 substrate (data not shown). Results from fluorescence anisotropy experiments were also consistent with the loss of HuR oligomer formation on ARE substrates shorter than 18 nucleotides. HuR binding to the Fl-ARE 20 and Fl-ARE 18 substrates was well resolved by the cooperative binding model of Equation 3 (Fig. 4B, left and middle panels, solid  lines), exhibiting Hill coefficients significantly greater than unity ( Table 2). Comparative regression solutions using singlesite or multiple identical binding site models incorporating protein depletion (Equation 6) were less favorable (Fig. 4B, left  and middle panels, dotted lines). For the Fl-ARE 20 substrate, the non-cooperative quadratic solution yielded a random distribution of residuals (p ϭ 0.2487); however, the sum-of-squares deviation was significantly improved using the cooperative binding model (p Ͻ 0.0011 by F test). For HuR binding to the Fl-ARE 18 substrate, the non-cooperative quadratic model was even less favorable because regression solutions using Equation 6 exhibited both nonrandom residuals (p ϭ 0.0201) and significantly poorer sum-of-squares deviation relative to the cooperative binding model (p ϭ 0.0002 by F test). By contrast, no cooperativity was detected in HuR binding to any Fl-ARE substrate of 16 nucleotides or less because resolved Hill coefficients did not significantly differ from 1 ( Table 2), and regression to cooperative binding models did not significantly improve sumof-squares deviations. Finally, interactions with HuR did not influence the fluorescence quantum yield of any Fl-labeled RNA substrate tested (data not shown), thus validating use of the binding algorithms described by Equations 3-6.
The concurrence of a single HuR⅐RNA complex containing either the ARE 16 or ARE 13 substrate by EMSA (Fig. 4A, right panel) (data not shown), together with a confident single-site binding solution by anisotropy-based assays (Fig. 4B, right panel; and Table 2), indicates that HuR forms 1:1 complexes with ARE substrates Յ16 nucleotides in length. Truncation of a single nucleotide (Fl-ARE 15 ) did not alter HuR binding activity, whereas removal of one or two additional bases (Fl-ARE 14 and Fl-ARE 13 ) only modestly diminished the affinity of HuR binding (Table 2). By contrast, shortening the ARE substrate to 11 nucleotides (Fl-ARE 11 ) significantly decreased HuR binding activity. Together, these data demonstrate that wild-type HuR optimally recognizes ARE substrates of ϳ15 nucleotides in length, but that slightly larger sequences are required to promote cooperative formation of HuR oligomers. A potential mechanism reconciling differences in the RNA substrate requirements for initial versus subsequent cooperative HuR binding events is considered under "Discussion." The RRM3 and Hinge Domains Contribute to the Affinity and Mechanism of ARE Substrate Binding by HuR-Published structures of peptide⅐RNA complexes containing RRM1 and RRM2 of HuD or HuC indicate that these domains are sufficient for specific recognition of ARE substrates (21,22). However, these structures indicate peptide contact with only eight to nine contiguous RNA bases, whereas data reported in this work show that association of a single HuR monomer is significantly enhanced for longer RNA substrates ( Table 2). These observations suggest that peptide sequences outside of RRM1 and RRM2 may make substantive contributions to the stability of HuR⅐ARE complexes. To test whether additional C-terminal domains of the HuR protein might contribute to its ARE-binding activity, we generated HuR truncation mutant proteins that lacked the RRM3 domain alone (HuR⌬RRM3) or the hinge and RRM3 domains together (HuR⌬Hinge3) (Fig. 1) and then tested their ability to bind ARE substrates. EMSA experiments revealed that the recombinant HuR⌬RRM3 protein retained the ability to form multisubunit complexes with the ARE 38 substrate, consistently yielding at least two distinct protein⅐ARE complexes (Fig. 5A, left panel). By contrast, binding of the HuR⌬Hinge3 protein to the ARE 38 substrate generated only a single protein⅐RNA complex by EMSA (Fig. 5A, right panel). RNA binding by either HuR deletion mutant remained dependent on the AU-rich sequence because neither protein formed complexes on the R␤ substrate (data not shown). Anisotropy-based analyses of HuR⌬RRM3 binding to the Fl-ARE 38 substrate were well resolved by the sequential two-step binding model described by Equation 4 (Fig. 5B, left panel, solid line), consistent with the formation of two or more protein⅐RNA complexes indicated by the EMSA experiment. The utility of this model was further supported by random distribution of residuals and a high coefficient of determination (R 2 Ͼ 0.99). By contrast, residual non-randomness (p ϭ 0.00025) and pairwise model comparisons using the F test (p Ͻ 0.0001) indicated that a single-site binding model (Equation 5) was clearly inappropriate for these data (Fig. 5B, left panel, dotted line). Using the two-step binding model, the apparent affinity of the initial interaction between HuR⌬RRM3 and the Fl-ARE 38 substrate (K d1 ϭ 1.7 nM) ( Table 3) was similar to that between the full-length HuR protein and optimal single-site substrates Fl-ARE 15 and Fl-ARE 16 (K d ϭ 2.5 nM) ( Table 2). Analyses of HuR⌬RRM3 binding to truncated Fl-ARE substrates indicated that the affinity of this initial protein binding event was similar to that of wild-type HuR for comparably sized ARE substrates, with apparent K d values Յ10 nM for all substrates of 13 nucleotides or longer (Table 3). Whereas the EMSA experiment suggested that higher protein concentrations were required to drive the first HuR⌬RRM3⅐ARE 38 binding step, it is possible that this initial protein⅐RNA complex is dynamic in nature and does not remain associated during gel loading and/or prolonged electrophoresis. A similar discrepancy between binding affinities resolved by EMSA versus anisotropy was observed for p37 AUF1 truncation mutants, which form highly dynamic complexes with RNA substrates (36). Removal of RRM3 from the HuD protein also accelerates the kinetics of RNA binding and dissociation in surface plasmon resonance experiments (20). However, although anisotropy-based experiments indicated that initial contact between HuR⌬RRM3 and the Fl-ARE 38 substrate was not impaired by the absence of RRM3, the apparent affinity of the second protein binding event was inhibited by a factor of nearly 30 (Table 3, cf. K d1 versus K d2 ). This second HuR⌬RRM3 binding event was also sensitive to RNA substrate size because it was not detectable by either EMSA (data not shown) or anisotropy (Table 3) with ARE substrates Յ20 nucleotides in length. Together, these data show that HuR⌬RRM3 remains capable of high affinity ARE recognition. However, subsequent formation of multisubunit protein complexes on ARE substrates is significantly impaired by loss of RRM3, indicating that this domain is essential for the cooperative assembly of protein oligomers observed with full-length HuR.
The formation of a single complex between the HuR⌬Hinge3 protein and the ARE 38 substrate suggested by EMSA experiments was also reflected in quantitative anisotropy-based binding assays because the data were well resolved by the single-site binding model described by Equation 5 (Fig. 5B, right panel). However, the apparent affinity of HuR⌬Hinge3 for the Fl-ARE 38 substrate was reduced ϳ60-fold relative to the HuR⌬RRM3 protein (Table 3). Also, no significant differences in the ARE-binding affinity of the HuR⌬Hinge3 protein were observed for RNA substrates between 13 and 38 nucleotides in length. Comparing the apparent ARE-binding affinities of the HuR⌬Hinge3 and HuR⌬RRM3 proteins indicated that the hinge domain contributes significant binding energy (⌬⌬G ϭ Ϫ2.4 kcal/mol) to the protein⅐RNA complex.
The Thermodynamic Stability of Protein Folding Is Not Impaired by Removal of the RRM3 and Hinge Domains-A trivial explanation for the defects in ARE-binding affinity and cooperative oligomer assembly exhibited by the HuR⌬Hinge3 and HuR⌬RRM3 proteins, respectively, is that HuR proteins do not fold properly in the absence of the hinge and/or RRM3 domain. To test this possibility, the folding stability of each recombinant protein was measured in chemical denaturation experiments. Here, denaturant-induced release of protein structure was monitored by changes in the fluorescence of tyrosine residues as each protein transitioned from folded to unfolded conformations across a titration of GdnHCl (Fig. 6A). The fluorescence of free tyrosine was not sensitive to GdnHCl (supplemental Fig. S3). Thermodynamic parameters describing the folding stability of each protein were then calculated from the GdnHCl concentration dependence of tyrosine fluorescence using Equations 1 and 2 (Fig. 6, B-D). Resolved constants describing the extrapolated free energy of protein unfolding in the absence of denaturant (⌬G uw ) and the sensitivity of each protein to GdnHCl (m eq ) are listed in Table 4. No statistically significant differences were observed among the values of ⌬G uw or m eq for the three recombinant proteins, indicating that loss of either the RRM3 and/or hinge domain does not compromise the overall stability of HuR protein folding.

The Basic Hinge Region Does Not Stabilize Protein⅐ARE Complexes via Enhanced Electrostatic Contacts-
Reversible transport of HuR between the nucleus and cytoplasm is mediated by a nuclear localization sequence element within the hinge region (25). However, loss of the hinge region also results in a significant decrease in ARE-binding affinity that is not coupled to global protein folding defects (detailed above), indicating that this domain might have additional roles in HuR function beyond housing the nuclear localization sequence. One mechanism by which the hinge domain might contribute to ARE-binding affinity was indicated by a significant decrease in the calculated isoelectric point of HuR⌬Hinge3 relative to HuR⌬RRM3 (supplemental Table SI). The hinge domain of HuR includes five basic amino acid residues that are largely conserved among the Hu protein family (Fig. 7A), which could potentially enhance the stability of Hu protein⅐ARE complexes by formation of intra-or intermolecular ion pairs. Such ionic interactions frequently contribute to the stability of complexes between RNA-binding proteins and their polyanionic substrates (46,47). To test this hypothesis, we measured the apparent binding affinities of the HuR⌬RRM3 and HuR⌬Hinge3 proteins for the Fl-ARE 20 RNA substrate across varying concentrations of monovalent cation (K ϩ ) using the fluorescence anisotropy assay. The Fl-ARE 20 substrate was selected for these experiments to avoid cation-dependent effects on the folding of larger ARE substrates (48) and potential influences of ARE folding on protein binding affinity (40). Apparent binding constants describing HuR⌬RRM3⅐Fl-ARE 20 and HuR⌬Hinge3⅐Fl-ARE 20 complex assembly at each cation concentration were plotted as log(K app ) versus Ϫlog[K ϩ ] and resolved by linear regression (Fig. 7B). The net number of ion pairs formed during assembly of a protein⅐RNA complex is reflected in the slope of this line (49). If the basic hinge domain contributes to the thermodynamic stability of protein⅐ARE complexes by formation of unique ionic interactions, removal of this region should result in a decrease in the net number of ion pairs formed. However, no statistically significant difference was observed in the affinity of HuR⌬RRM3 versus HuR⌬Hinge3 binding to the Fl-ARE 20 substrate as a function of monovalent cation concentration

DISCUSSION
The HuR basic hinge domain contains the nuclear localization sequence element required for nucleocytoplasmic shuttling by this protein (25). However, the data presented in this study also indicate that this domain contributes to the stability of HuR complexes with ARE substrates (Table 3). This enhancement of ARE-binding activity was significantly greater than that observed in comparable experiments with HuD truncation mutants, where inclusion of the hinge region yielded only a 2-fold improvement in binding to a TNF␣ ARE-derived RNA substrate (20). However, two factors may account for the apparent differences in contributions of hinge domains to the ARE-binding activities of HuR versus HuD. First, the HuD RRM1 ϩ RRM2 protein included the first 13 residues of the hinge domain, which may have contributed to its strong AREbinding activity (K d ϭ 5.4 nM by surface plasmon resonance analysis) (20). Second, the hinge domain is the least conserved region in the Hu protein family with the possible exception of the extreme N-terminal end. The HuR hinge domain shares only 47% amino acid sequence identity with other family members (Fig. 7A) and also lacks ϳ10 -30 residues retained within the other three proteins. It is conceivable that sequence characteristics unique to the HuR hinge domain are required for enhancement of ARE-binding activity.
The HuR hinge domain does not include sequence motifs characteristic of any known RNA-binding domain. However, based on several findings reported in this study, we predict that the hinge domain stabilizes HuR⅐ARE ribonucleoprotein complexes through formation of specific protein-RNA contacts. First, equilibrium binding experiments demonstrated that removal of the hinge and RRM3 domains (HuR⌬Hinge3) significantly decreased binding affinity for ARE substrates relative to the HuR⌬RRM3 protein (Table 3). Second, enhanced ARE binding in the presence of the hinge domain required additional RNA sequences. Binding of the HuR⌬RRM3 protein was significantly enhanced for ARE substrates Ն14 nucleotides in length, whereas the ARE-binding activity of HuR⌬Hinge3 was not improved by increasing the length of the RNA substrate from 13 to 38 bases (Table 3). Third, GdnHCl denaturation studies demonstrated that the hinge domain did not enhance the thermodynamic stability of protein folding (Table 4), indicating that the conformations of RRM1 and RRM2 are likely similar whether alone or in the context of the full-length protein.
Finally, inclusion of the HuR hinge domain did not stabilize protein⅐ARE complexes by increasing the net number of ion pairs formed during complex assembly (Fig. 7B). Ionic interactions can make significant contributions to the free energy of  (GdnHCl). B-D, fluorescence emission at 305 nm from HuR, HuR⌬RRM3, and HuR⌬Hinge3, respectively, is plotted as a function of GdnHCl concentration. Thermodynamic parameters describing the folding stability of each protein were resolved by nonlinear regression using Equations 1 and 2 as indicated under "Experimental Procedures" and are quoted in Table 4.
protein-nucleic acid interactions in an RNA/DNA sequenceindependent manner (49,50). However, enhanced ARE binding by HuR⌬RRM3 relative to the HuR⌬Hinge3 protein without formation of new ion pairs suggests that the hinge domain can make specific but hitherto undefined non-ionic contacts with RNA substrates. Coincident RNA binding and nuclear localization functions within the basic hinge domain also raise the intriguing possibility that these events may be mutually exclusive. Future studies will address whether association with RNA precludes nuclear import and, by extension, whether HuR relocalization to the nucleus is a default consequence of dissociation from cytoplasmic RNA substrates.
Several observations indicate that full-length HuR cooperatively forms multimers on ARE substrates Ն18 nucleotides in length. (i) Multiple HuR⅐ARE complexes were detected by EMSA ( Figs. 2A and 4A); (ii) anisotropy-based binding assays resolved Hill coefficients significantly greater than unity ( Table  2); and (iii) apparent binding affinity was significantly improved relative to ARE substrates supporting only a single HuR binding event ( Table 2, cf. Fl-ARE 20 and Fl-ARE 18 versus Fl-ARE 16 and Fl-ARE 15 ). Cooperative assembly of HuR oligomers on AREbased RNA substrates also required both the third RRM of the protein and some RNA sequence flanking the first high affinity HuR-binding site. It is currently unclear whether HuR binding cooperativity is mediated by protein-protein interactions, likely involving RRM3, and/or by proteininduced remodeling of local RNA structure. However, although optimal binding of a single HuR monomer required ϳ15 nucleotides (Table 2), both binding density (Fig.  3) and EMSA (Fig. 4A) experiments support an HuR-binding site size of eight to nine nucleotides in the cooperative oligomeric complex. This site size is similar to the binding interface of the tandem RRM1 ϩ RRM2 domains of HuC and HuD (21,22), suggesting that RNA contacts involving the hinge and/or RRM3 domain may be released before or during association of the second HuR monomer. Data from this study also indicate that cooperative HuR binding may be limited to pairs of HuR monomers. Binding density analysis indicated that as many as four HuR binding events were possible on the ARE 38 substrate, three of which were resolved by EMSA ( Fig. 2A). However, the decreased [HuR]1 ⁄ 2 observed with the Fl-ARE 38 substrate versus Fl-ARE 18 and Fl-ARE 20 suggests that additional binding steps on the longest RNA substrate may occur with lower affinity because the resolved [HuR]1 ⁄ 2 value is an aggregate parameter describing all protein binding events contributing to formation of saturated HuR⅐ARE complexes. The EMSA showing the concentration dependence of HuR binding to the ARE 38 substrate ( Fig. 2A) is also consistent with this idea because formation of the third detectable HuR⅐ARE complex was observed only at substantially higher protein concentrations.
It is intriguing that the mechanism of ARE substrate binding employed by the mRNA-stabilizing trans-factor HuR is significantly different from that utilized by the cytoplasmic isoforms of AUF1, which may contribute to the destabilization of selected mRNA targets. In this work, we have shown that HuR forms oligomers on the TNF␣ ARE 38 substrate in a cooperative fashion. By contrast, although the p37 AUF1 and p40 AUF1 proteins bind the same ARE sequence via a sequential two-step binding mechanism, the second binding event typically displays 20 -50-fold lower affinity than the first (28,51). Additional studies have indicated that HuR and AUF1 are coexpressed in many cell types and that they associate with many of the same cellular mRNAs (27,52,53), suggesting that the equilibrium between HuR versus AUF1 binding may dictate the catabolic fate of specific ARE-containing transcripts. Supporting this model are observations that many cellular stresses increase cytoplasmic HuR concentrations by subcellular relocalization from the nucleus, concomitant with stabilization of HuR-tar- geted mRNAs (26,27,54,55). The cooperative ARE-binding activity of HuR would facilitate this process because modest changes in cytoplasmic HuR concentrations could thus dramatically influence the fractional binding density of this protein on ARE-containing mRNAs. Finally, loss of binding cooperativity in HuR proteins lacking RRM3 may explain discrepancies in the mRNA-stabilizing activity of these mutants (13,24) because significantly higher concentrations of HuR⌬RRM3 proteins would be required to effectively displace competing mRNA-destabilizing factors from mRNA substrates. Overexpression experiments that fail to achieve this threshold cytoplasmic HuR⌬RRM3 concentration would thus be less likely to stabilize these targeted transcripts.