A Hairpin-like Structure within an AU-rich mRNA-destabilizing Element Regulates trans -Factor Binding Selectivity and mRNA Decay Kinetics*

In mammals, rapid mRNA turnover directed by AU-rich elements (AREs) is mediated by selective association of cellular ARE-binding proteins. These trans -acting factors display overlapping RNA substrate specificities and may act to either stabilize or destabilize targeted transcripts; however, the mechanistic features of AREs that promote preferential binding of one trans -factor over another are not well understood. Here, we describe a hairpin-like structure adopted by the ARE from tumor necrosis factor (cid:1) (TNF (cid:1) ) mRNA that modulates its affinity for selected ARE-binding proteins. In particular, association of the mRNA-destabilizing factor p37 AUF1 was strongly inhibited by adoption of the higher order ARE structure, whereas binding of the inducible heat shock protein Hsp70 was less severely compromised. By contrast, association of the mRNA-stabilizing protein HuR was only minimally affected by changes in ARE folding. Consistent with the inverse relationship between p37 AUF1 binding affinity and the stability of ARE folding, mutations that stabilized the ARE hairpin also inhibited its ability to direct rapid mRNA turnover in transfected cells. Finally, phylogenetic analyses and structural modeling indicate that TNF (cid:1) mRNA sequences flanking the ARE are highly conserved and may stabilize the hairpin fold in vivo . Taken together, these data suggest that local higher order structures involving AREs may function as potent regu-lators of mRNA turnover in mammalian cells by modulating trans -factor binding selectivity. The of is highly the Eclipse fluorescence spectrophotometer (Varian Instruments) with a Peltier tempera- ture controller and in-cell temperature probe. Thermodynamic parameters describing the folding of RNA substrates were extracted from the partial derivative of E FRET ( (cid:7) T (cid:5) 4 °C) as a function of temperature. For a given structural transition, the reaction temperature yielding a local maximum in the derivative plot was defined as the apparent melting temperature ( T m ), whereas the temperature above T m yielding half- maximal derivative height was defined as T 2 . Using these parameters, the change in the van’t Hoff enthalpy ( (cid:7) H VH ) for a unimolecular RNA structural transition is given by Equation 6 (48). Taking (cid:7) H VH together with the experimentally derived T m thus allows the transition free energy ( (cid:7) G fold ) to be calculated as a function of reaction temperature ( T ) using Equation 7 Protein-RNA Binding Assays— Qualitative assessment of binding events between 32 P-labeled RNA substrates and recombinant proteins

The rate of mRNA turnover is highly variable among the cytoplasmic mRNA population and thus plays a significant role in regulating the steady-state concentrations of individual mRNA species available to program protein synthesis. In mammalian cells, different transcripts exhibit a range of decay kinetics spanning over 2 orders of magnitude, largely due to the presence of discrete cis-acting elements contained within each mRNA (1,2). Most mRNAs encoding cytokines, inflammatory mediators, and proto-oncogenes are inherently unstable, often exhibiting cytoplasmic half-lives of 1 h or less. Rapid turnover of these transcripts is principally due to the activity of AU-rich elements (AREs), 1 a broad family of mRNA-destabilizing sequences localized to the 3Ј-untranslated regions (3Ј-UTRs) of many labile mRNAs (3). The intrinsic lability of ARE-containing mRNAs enables their cytoplasmic concentrations to be rapidly modulated following acute changes in their synthetic rates (4,5). Additionally, modulation of ARE-directed mRNA decay pathways by selected signal transduction systems can regulate the cytoplasmic levels of some mRNAs independent of, or in concert with, changes in the synthetic rate (6 -8).
The ability of AREs to direct mRNA turnover is mediated by the activity of selected ARE-binding proteins (9 -11). To date, over 25 such factors have been identified, although the regulatory significance of most remains unknown. Some proteins, including AUF1 (12)(13)(14), tristetraprolin (TTP) (15,16), and KSRP (17,18), appear to promote ARE-directed mRNA turnover, whereas others, including members of the Hu family of RNA-binding proteins, function to prevent the degradation of some ARE-containing transcripts (14,19,20). Recent studies, however, have indicated that different AREs or subdomains thereof may be targeted by distinct populations of ARE-binding proteins. For example, whereas Hu proteins bind relatively promiscuously to RNA substrates in vitro, with little selectivity beyond a general preference for sequences rich in uridylate residues (21)(22)(23), some AU-rich sequences functioning as potent mRNA destabilizers are insensitive to HuR activity (24,25). By contrast, TTP binding to RNA is highly discriminatory, requiring an AUUUA or AUUUUA motif within a U-rich tract for high affinity interactions (26,27).
The mRNA-destabilizing factor AUF1 exists as a family of four protein isoforms generated by alternative pre-mRNA splicing and are named according to their apparent molecular weights (28). The p45 AUF1 and p42 AUF1 isoforms are principally nuclear, whereas p40 AUF1 and p37 AUF1 are generally present in both nuclei and cytoplasm (29 -31). To date, the p40 AUF1 and p37 AUF1 isoforms have been most closely associated with the regulation of ARE-directed mRNA turnover (12)(13)(14)31). In particular, extensive studies have documented p37 AUF1 interactions with a broad spectrum of AREs and related RNA sequences (12,(32)(33)(34)(35)(36); however, the specific features of RNA substrates that discriminate high from low affinity AUF1 targets remain unclear. p37 AUF1 binds U-rich RNA substrates by sequential association of protein dimers (37) and induces local changes in RNA conformation upon binding (38). However, p37 AUF1 binding to the ARE from tumor necrosis factor ␣ (TNF␣) mRNA is inhibited by the presence of Mg 2ϩ in vitro, concomitant with the adoption of a condensed but undefined RNA structure (33,38). Both inhibition of AUF1 binding and RNA structural condensation resulting from Mg 2ϩ treatment were specific for the ARE sequence, since the cation did not significantly alter the affinity of p37 AUF1 binding to polyuridylate substrates, nor did Mg 2ϩ induce substantial conformational changes in these RNA ligands (33).
Together, these data suggested that one or more higher order RNA structures involving the TNF␣ ARE could inhibit association of AUF1 and, by extension, that local higher order RNA structures involving AREs may serve as significant determinants directing their binding preferences for different transacting factors. In this study, we have used structural modeling, nuclease mapping, and fluorescence-based approaches to define a candidate folded structure for the TNF␣ ARE in solution. Quantitative assessment of the thermodynamic consequences of selected site-directed mutations within the TNF␣ ARE further supports the proposed folded model. Stabilization of the folded ARE structure potently inhibited binding of p37 AUF1 to this element in vitro. However, trans-factor binding was not universally responsive to the folded state of the TNF␣ ARE. The 70-kDa inducible heat shock protein, Hsp70, also associates with U-rich RNA substrates (39), but its ARE-binding activity was significantly less sensitive to the stability of RNA folding. More dramatically, binding of the mRNA-stabilizing factor HuR to the ARE substrate was virtually independent of ARE folding. The preference of folded ARE substrates for HuR over p37 AUF1 was consistent with the results of transfected cell studies, where ARE-directed mRNA turnover was inhibited by a mutation that stabilized ARE folding. Taken together, these data indicated that the potential of an ARE to form higher order structures can significantly influence its selectivity for trans-acting factors, thus contributing to modulation of mRNA decay kinetics in cells. Finally, phylogenetic sequence-and structure-based analyses indicate that TNF␣ mRNA 3Ј-UTR sequences flanking the ARE may serve to stabilize an extended folded structure in vivo and that this regulatory element probably exists in similarly folded RNA conformations throughout mammals. From these data, we propose that folded RNA structures involving the TNF␣ ARE constitute important determinants of RNA-binding activity for selected ARE-binding proteins, in some cases possibly as significant in magnitude and scope as RNA primary structural information.

EXPERIMENTAL PROCEDURES
RNA Substrates-Synthesis, deprotection of 2Ј-hydroxyl groups, and purification of RNA substrates were performed by Dharmacon Research or Integrated DNA Technologies. The oligoribonucleotide ARE (38) includes the core ARE sequence from TNF␣ mRNA (5Ј-GUGAU UAUUUAUUAUUUAUUUAUUAUUUAUUUAUUUAG-3Ј). Selected site-directed mutants are identified where applicable. 5Ј-linked cyanine-3 (Cy3) and 3Ј-linked fluorescein (Fl) groups are indicated by relevant prefixes and suffixes, respectively, where applicable. Lyophilized RNA pellets were reconstituted in 10 mM Tris⅐HCl (pH 8). RNA yields and fluorophore labeling efficiencies were quantified by absorbance spectroscopy, incorporating fractional contributions of fluorescent dyes to A 260 for RNA substrates containing Fl and/or Cy3 moieties as previously described (38,40). For fluorescence experiments, RNA substrate concentrations were based on the abundance of Fl-conjugated RNA, which typically exceeded 90% of the total synthesis. For experiments requiring 5Ј-32 P-labeled RNA, 5Ј-hydroxyl RNA substrates were radiolabeled using T4 polynucleotide kinase and [␥-32 P]ATP to specific activities of 3-5 ϫ 10 3 cpm/fmol as described (37).
Preparation of Recombinant Proteins-Standard DNA subcloning techniques were used to modify the bacterial expression plasmid pBAD/ HisB (Invitrogen) to permit expression of fusion proteins containing the N-terminal peptide sequence MSHHHHHHGT. The resulting plasmid was termed pBAD/HT. A cDNA encoding human p37 AUF1 was sub-cloned immediately downstream of this sequence tag to yield plasmid pB/HT-p37 AUF1 . All constructs were verified by restriction digests and automated DNA sequencing. The pB/HT-p37 AUF1 plasmid was transformed into Escherichia coli TOP10 cells, and expression of the encoded His 6 -p37 AUF1 recombinant protein was induced by arabinose. The His 6 -p37 AUF1 protein was purified from lysed cells using Ni 2ϩ -affinity chromatography as described (33). The His 6 -Hsp70 protein was similarly expressed from plasmid pBAD/HisC-hsp70 and purified using Ni 2ϩ affinity chromatography as described (39).
A cDNA encoding the open reading frame of HuR was amplified by reverse transcription-PCR using RNA purified from the human monocytic leukemia cell line THP-1 and then subcloned into pGEX-3X (Amersham Biosciences). Following sequence verification, a glutathione S-transferase (GST)-HuR chimeric protein was expressed in E. coli TOP10 cells by induction with isopropyl 1-thio-␤-D-galactopyranoside and purified by glutathione affinity chromatography using a GSTrap FF column (Amersham Biosciences) as recommended by the manufacturer. GST lacking the HuR moiety was similarly expressed from unmodified pGEX-3X and purified. All recombinant proteins were quantified using Coomassie Blue-stained SDS-PAGE gels against a titration of bovine serum albumin.
Bioinformatics, RNA Structural Modeling, and Nuclease Mapping-Sequence alignments were prepared using the ClustalW program at the San Diego Supercomputer Center Biology Workbench (available on the World Wide Web at www.workbench.sdsc.edu). Candidate local RNA secondary structures were identified using mFold version 2.3 (available on the World Wide Web at www.bioinfo.rpi.edu/ϳzukerm) (41,42).
Nuclease mapping experiments were performed using a modification of the RNase V 1 endonuclease protocol provided by Ambion. Briefly, 5Ј-32 P-labeled RNA substrates (25 fmol) were incubated in a final volume of 10 l containing 25 mM HEPES⅐HCl (pH 7.4), 50 mM KCl, and 4 g of yeast tRNA. Where indicated, MgCl 2 was included to 5 mM. In reactions lacking Mg 2ϩ , 0.5 mM EDTA was added to scavenge residual divalent cations. After equilibration at 25°C, RNase A (0.045 Kunitz U; Sigma) was added, and incubation continued for 5 min. Reactions were stopped, and RNA fragments were precipitated using the inactivation/ precipitation buffer (Ambion). Reaction products were fractionated on a 12% denaturing acrylamide gel, which was then dried and visualized using a PhosphorImager (Amersham Biosciences). Single base RNA ladders were prepared by hydroxide ion-mediated cleavage following protocols provided by Ambion.
FRET-based Assays of RNA Folding-Folding events involving RNA substrates were monitored by changes in the distance between 3Јlinked Fl donor and 5Ј-linked Cy3 acceptor moieties using fluorescence resonance energy transfer (FRET). The efficiency of FRET (E FRET ) is inversely related to the sixth power of the scalar distance (r) between a fluorescent donor and acceptor (Equation 1). R 0 is the Förster distance, defined as the distance between a donor-acceptor pair where FRET efficiency is 50% (43,44).
R 0 for the Fl-Cy3 donor-acceptor pair linked to single-stranded nucleic acids has been calculated as 55.7 Å (45). E FRET between donor and acceptor fluorophores may be resolved using Equation 2 (44,46), where F DA is the fluorescence of the donor in the presence of the acceptor, whereas F D is the fluorescence measured from parallel reactions containing RNA substrates labeled with the donor fluorophore (i.e. Fl) alone.
However, without quantitative acceptor labeling, E FRET will be underestimated due to the presence of RNA molecules labeled only with Fl within the (Cy3 ϩ Fl)-labeled RNA substrate population. To correct for this, donor emission from (Cy3 ϩ Fl)-labeled RNA substrates (F Cy-Fl ) was interpreted by Equation 3, where f DA is the efficiency of acceptor labeling, in this case Cy3 (typically 75-90%).
Incorporating this function into Equation 2 and substituting F D ϭ F Fl yields Equation 4.
The ability of Mg 2ϩ to stabilize folding of RNA substrates was quantified by measuring E FRET as a function of Mg 2ϩ concentration at con-stant temperature essentially as described (38), with E FRET calculated using Equation 4. Considering the conformation of the RNA substrate to be in equilibrium between an open, cation-free state and a folded, cation-associated state across a titration of Mg 2ϩ concentrations allows these data to be resolved by Equation 5, derived from a variant of the Hill model (47). The thermodynamic stability of RNA folding was evaluated by thermal denaturation in paired experiments containing either (Cy3 ϩ Fl) double-labeled (for F Cy-Fl ) or Fl single-labeled (for F Fl ) oligoribonucleotides. Samples contained 20 nM RNA in 10 mM Tris⅐HCl (pH 8), 50 mM KCl, 1 g/l heparin, and 0.1 g/l acetylated bovine serum albumin. Where indicated, Mg 2ϩ was added as MgCl 2 , whereas samples lacking Mg 2ϩ contained 0.5 mM EDTA. Samples were equilibrated in solution for 5 min at 10°C, and then fluorescence was measured ( ex ϭ 490 nm, em ϭ 518 nm, 5-nm band pass) in 0.5°C intervals as the temperature increased from 10 to 70°C at 1°C/min using a Cary Eclipse fluorescence spectrophotometer (Varian Instruments) with a Peltier temperature controller and in-cell temperature probe. Thermodynamic parameters describing the folding of RNA substrates were extracted from the partial derivative of E FRET (⌬T ϭ 4°C) as a function of temperature. For a given structural transition, the reaction temperature yielding a local maximum in the derivative plot was defined as the apparent melting temperature (T m ), whereas the temperature above T m yielding halfmaximal derivative height was defined as T 2 . Using these parameters, the change in the van't Hoff enthalpy (⌬H VH ) for a unimolecular RNA structural transition is given by Equation 6 (48).
Taking ⌬H VH together with the experimentally derived T m thus allows the transition free energy (⌬G fold ) to be calculated as a function of reaction temperature (T) using Equation 7 (48).
Protein-RNA Binding Assays-Qualitative assessment of binding events between 32 P-labeled RNA substrates and recombinant proteins was performed using gel mobility shift assays essentially as described (49), except that Mg 2ϩ was not included in binding reactions. Protein-RNA binding affinity was quantitatively measured by monitoring the changes in the fluorescence anisotropy of Fl-conjugated RNA substrates across a titration of protein essentially as described (33,37). Concomitant measurement of total fluorescence emission for each reaction demonstrated that association of either His 6 -p37 AUF1 , His 6 -Hsp70, or GST-HuR with Fl-labeled RNA substrates did not significantly affect their fluorescence quantum yield (data not shown).
AUF1 proteins interact with the TNF␣ ARE by sequential association of protein dimers yielding a tetrameric protein complex on the RNA substrate (P 4 R). Assembly of this complex as a function of protein dimer concentration ([P 2 ]) is thus resolved by tandem association equilibrium constants (33,37); K 1 describes the initial interaction between the RNA substrate and a protein dimer, whereas K 2 describes the interaction between the RNA⅐protein dimer complex (P 2 R) and a subsequent protein dimer. Under conditions of constant quantum yield and limiting RNA substrate (i.e. [RNA-Fl] total Ͻ 1/K 1 ), K 1 and K 2 are derived from the total measured anisotropy of the Fl-conjugated RNA substrate (A t ) as a function of protein concentration by Equation 8, where A R , A P2R , and A P4R represent the intrinsic anisotropy values of the free RNA, the RNA⅐protein dimer complex, and the RNA⅐protein tetramer complex, respectively. A R is measured directly from binding reactions lacking protein (n Ն 4), whereas all other constants are resolved by nonlinear regression using PRISM version 3.03 software (GraphPad). The formation of equimolar complexes between His 6 -Hsp70 and Fl-labeled TNF␣ ARE substrates was resolved using a binary association function (Equation 9) as described previously (39).
Quantitative assessment of binding between GST-HuR and Fl-labeled RNA substrates was similarly preformed using fluorescence anisotropy, but with two significant variations. First, preliminary on-rate analyses indicated significantly slower association kinetics for GST-HuR versus His 6 -p37 AUF1 or His 6 -Hsp70 binding to ARE substrates (data not shown). Accordingly, RNA-binding reactions containing GST-HuR were incubated for a minimum of 1 h prior to measurement of fluorescence anisotropy. Second, the cooperative nature of GST-HuR binding to ARE substrates (detailed under "Results") necessitated analysis by a variant of the Hill model given in Equation 10, where A PxR is the intrinsic anisotropy of the saturated Fl-RNA⅐GST-HuR x complex, and [P]1 ⁄2 is the concentration of protein giving half-maximal binding.
Measurement of mRNA Decay Kinetics-The cellular decay rates of ␤-globin-ARE reporter mRNAs were measured using doxycycline (Dox) time course assays, slightly modified from previous descriptions (50). Briefly, the plasmid pTRER␤-wt was constructed, encoding the rabbit ␤-globin gene under the control of a tetracycline-responsive promoter. ARE sequences were subcloned into the unique BglII restriction site located downstream of the translation termination codon and verified by sequencing. pTRER␤-wt and derivative plasmids were transfected using Superfect (Qiagen) into HeLa/Tet-Off cells (BD Biosciences) along with the control plasmid pSV␣1, encoding ␣-globin. 48 h post-transfection, transcription from the pTRER␤-derived plasmids was arrested by the addition of Dox (2 g/ml). At selected time points thereafter, RNA was harvested and analyzed for ␣-globin and ␤-globin mRNA levels using the Direct Protect RNase Protection Assay (Ambion). RNase protection assay reactions were programmed with 1 fmol each of 32 Plabeled antisense ␤-globin (complementary to portions of intron 2 and exon 3) and ␣-globin (complementary to exon 2) riboprobes, prepared by in vitro transcription from plasmid DNA templates to specific activities of 2-4 ϫ 10 4 cpm/fmol. Following normalization to ␣-globin mRNA, turnover rates of ␤-globin and ␤G-ARE mRNAs were calculated by nonlinear regression of the percentage of (␤-globin or ␤G-ARE) mRNA remaining as a function of time following Dox treatment, yielding first-order decay constants (k).

RESULTS
A Candidate Stem-Loop Folded Structure for the TNF␣ ARE-Previous data indicated that an RNA substrate containing the core TNF␣ ARE sequence adopts a unimolecular, condensed structure in a cation-dependent manner (33,38). Since folding of this substrate was preferentially facilitated by highly charged, spatially compact cations, we concluded that multivalent cations probably stabilize the condensed ARE structure by targeted counterion neutralization at regions of high negative charge density. To identify potential local RNA structures involving the core TNF␣ ARE, we first analyzed a 38-nucleotide RNA sequence containing this element (ARE(38)) using the mFold algorithm (41,42). At 25°C, mFold returned a hairpinlike structure punctuated by symmetrical U:U mismatches as the thermodynamically preferred RNA conformation (Fig. 1A). Suboptimal predicted structures were principally slip-shifted variants of this candidate fold.
This candidate ARE structure was first tested by nuclease mapping. RNase A, which preferentially cleaves 3Ј of singlestranded pyrimidine residues, completely digested the ARE (38) substrate in the absence of Mg 2ϩ under the conditions tested (Fig. 1B). By contrast, the addition of 5 mM Mg 2ϩ conferred significant nuclease resistance, consistent with protection of single-stranded regions. Control experiments using an unstructured RNA substrate verified that the catalytic activity of RNase A did not significantly differ in the presence or absence of Mg 2ϩ (data not shown). Whereas inclusion of Mg 2ϩ strongly decreased the sensitivity of the ARE(38) substrate to RNase A, specific cleavage sites observed in the presence of the cation localized to unpaired uridylate residues predicted by mFold, with the exception of U 31 (Fig. 1A, arrows). These data support Mg 2ϩ -dependent adoption of the folded ARE(38) structure predicted by mFold based on presentation of unpaired uridylate residues at predicted sites and general loss of accessibility for the single strand-specific nuclease, consistent with adoption of intramolecular base pairs in the presence of the cation.
Site-directed Mutations within the TNF␣ ARE Modulate the Stability of RNA Folding-Whereas the nuclease mapping data are largely consistent with the stem-loop ARE structure predicted by mFold, this model was further tested by creating single point mutations within the ARE (38) sequence that were predicted to alter the thermodynamic stability of its folded structure. The A 4 3 C mutation was expected to weaken the folded structure by abrogating formation of the A 4 -U 31 base pair (Fig. 1C). By contrast, the U 32 3 C substitution was predicted to strengthen the folded stem-loop, based on replacement of the noncanonical G 3 -U 32 interaction with a stronger G 3 -C 32 pair.
To determine the influence of these mutations on formation of the putative TNF␣ ARE hairpin, the Mg 2ϩ sensitivity and thermodynamic stability of folding events involving the ARE (38), ARE(A 4 3 C), and ARE(U 32 3 C) substrates were quantitatively assessed in vitro using FRET. These experiments monitored the difference in E FRET between 3Ј-linked Fl and 5Ј-linked Cy3 moieties of double-labeled RNA substrates resulting from the transition between a largely unfolded state (low E FRET ) and a folded state where the termini are in close proximity (high E FRET ) ( Fig. 2A). In the absence of monovalent salt, the addition of Mg 2ϩ resulted in a decrease in emission from the Fl moiety ( max ϭ 518 nm) of the Cy3-ARE(38)-Fl RNA substrate (Fig. 2B). This decrease was not a function of nonquantum quenching, since Fl emission from the corresponding substrate lacking the Cy3 acceptor (ARE(38)-Fl) was not significantly affected by Mg 2ϩ (Fig. 2C). Parallel measurements of donor emission from the Cy3-ARE(38)-Fl and ARE(38)-Fl substrates thus allowed calculation of E FRET by Equation 4 and demonstrated that the distance between the 5Ј-and 3Ј-termini of the Cy3-ARE(38)-Fl substrate decreased as a function of Mg 2ϩ concentration (Fig. 2D, solid circles). Resolution of these data by the Hill model (Equation 5) indicates that adoption of the folded RNA structure is cooperative with respect to Mg 2ϩ at 25°C (h ϭ 1.53 Ϯ 0.09) and exhibits half-maximal folding ([Mg 2ϩ ]1 ⁄2 ) at 73 Ϯ 3 M Mg 2ϩ , consistent with previous findings (38). Folding of the Cy3-ARE(A 4 3 C)-Fl substrate exhibited similar cooperative character with respect to Mg 2ϩ but required higher cation concentrations to stabilize the folded state. By contrast, folding of the Cy3-ARE(U 32 3 C)-Fl substrate was observed at lower concentrations of Mg 2ϩ and without discernible cooperativity (Table I). Differences in the concentration of Mg 2ϩ necessary to promote folding of each RNA substrate (i.e. [Mg 2ϩ ]1 ⁄2 for ARE(U 32 3 C) Ͻ ARE(38) Ͻ ARE(A 4 3 C)) are thus consistent with their predicted rank order of intrinsic folded stability in solution (i.e. ARE(U 32 Characterizing the Mg 2ϩ -stabilized structural condensation of ARE substrates in the absence of monovalent cations has provided useful information regarding the physical basis for local RNA remodeling by Mg 2ϩ (38)  sion of monovalent cations in subsequent experiments designed to elucidate the biochemical significance of ARE folding. First, nucleic acid interactions with mono-versus multivalent cations occur through diverse modes with varying affinities, which can promote distinct structural consequences (51). Second, the intracellular environment includes a broad spectrum of cations of differing valence and charge distribution. Finally, interactions between RNA substrates and cellular proteins are often sensitive to ionic strength (52,53). In particular, association of AUF1 with AU-rich RNA substrates is almost completely inhibited under hypotonic ([K ϩ ] Ͻ 25 mM) conditions (38), thus precluding omission of monovalent cations from any study of AUF1-RNA binding affinity. Based on these concerns, assessing the role of ARE folding in the regulation of trans-factor binding affinity required the establishment of parameters describing the stability of folded ARE structures in mixed cationic environments. Analysis of the Mg 2ϩ dependence of Cy3-ARE(38)-Fl folding in the presence of 50 mM KCl yielded three principal observations (Fig. 2E). First, resolution of these data   (51). Finally, as the concentration of Mg 2ϩ increases, E FRET of the Cy3-ARE(38)-Fl RNA substrate approaches a limiting value (E f ) of greater than 0.9. By Equation 1, this indicates that the average distance between the Fl and Cy3 moieties in the folded state is no more than 39 Å. Given that the average diameter of an A-form double helix is ϳ26 Å (54) and that the donor and acceptor dyes are tethered to the Cy3-ARE(38)-Fl substrate by 6-and 3-atom linkers, respectively, this calculated interfluorophore distance is consistent with the mFold model of the folded ARE(38) structure (Fig. 1A), where the 5Ј-and 3Ј-termini are closely spaced in solution.
Thermodynamic parameters describing the stability of Cy3-ARE(38)-Fl substrate folding in a background of 50 mM KCl were estimated by thermal denaturation. In the absence of Mg 2ϩ , the Cy3-ARE(38)-Fl substrate adopted a condensed conformation at low temperature denoted by high E FRET (Fig. 2F, dotted line). However, a decrease in E FRET as temperature increased indicated release of the folded structure. The change in van't Hoff enthalpy accompanying this transition was calculated from the partial derivative of E FRET as a function of temperature (Fig. 2G) as described under "Experimental Procedures," which in turn permitted solution of the apparent free energy of folding (⌬G 0 fold ) at 25°C using Equation 7. In the absence of Mg 2ϩ , folding of the Cy3-ARE(38)-Fl substrate is unfavorable at 25°C (⌬G 0 fold Ͼ 0), but the addition of Mg 2ϩ significantly shifts the equilibrium in favor of the folded state in a dose-dependent manner, based on both increases in the apparent melting temperature (T m ) and concomitant decreases in ⌬G 0 fold (Table II). For example, inclusion of 5 mM Mg 2ϩ (Fig.  2, F and G, solid line) increased T m of the Cy3-ARE(38)-Fl substrate by 11.5°C, yielding ⌬⌬G of Ϫ2.4 kcal/mol at 25°C relative to substrate folding in the absence of Mg 2ϩ . Thermal denaturation analyses of fluorescent RNA substrates containing the ARE(A 4 3 C) and ARE(U 32 3 C) sequences further validated the predicted roles of the mutated bases in stabilization of the ARE hairpin structure. For each concentration of Mg 2ϩ tested, values of ⌬G 0 fold indicated that the folded structure of the Cy3-ARE(A 4 3 C)-Fl substrate was significantly less stable than that formed with the Cy3-ARE(38)-Fl substrate, whereas folding of the Cy3-ARE(U 32 3 C)-Fl substrate was significantly more stable (Table II). Taken together, the increased concentration of Mg 2ϩ required to stabilize the folded Cy3-ARE(A 4 3 C)-Fl substrate relative to the Cy3-ARE(38)-Fl substrate (Table I), coupled with the decreased value of ⌬G 0 fold for the A 4 3 C mutant (Table II), is consistent with participation of A 4 in an A-U base pair contact in the wild type ARE structure. Similarly, the improved stability and diminished Mg 2ϩ requirement of the folded Cy3-ARE(U 32 3 C)-Fl substrate relative to the wild type ARE supports formation of a G-C base pair in the mutant RNA structure. Since the U 32 3 C substitution would weaken an RNA structure containing a U-A base pair involving U 32 , and only two guanosine residues are present upstream of U 32 in the ARE(38) sequence, these findings are consistent with a base pair interaction between U 32 and either G 1 or G 3 in cation-stabilized folding of the wild type TNF␣ ARE.
A Compensatory Double Mutation Highlights an Interaction between G 3 and U 32 during ARE Folding-In order to confirm G 3 as the interacting partner of U 32 in the folded ARE structure, a compensatory double mutation was made by changing G 3 3 U and U 32 3 C (Fig. 1C). Cation titration and thermal denaturation experiments of RNA substrates containing these substitutions yielded three principal observations. First, significantly higher concentrations of Mg 2ϩ were required to stabilize the folded state of the Cy3-ARE(G 3 3 U; U 32 3 C)-Fl substrate relative to the Cy3-ARE(U 32 3 C)-Fl substrate (Fig. 2D, Table I). Second, the cooperativity of ARE substrate folding with respect to Mg 2ϩ that was lost by the U 32 3 C substitution was largely restored in the double mutant (Table I). Finally, thermal denaturation experiments demonstrated that the improvement in folding energy conferred by the U 32 3 C substitution relative to the wild type ARE was completely lost by concurrent mutation of G 3 3 U at all Mg 2ϩ concentrations tested (Table II). Taken together, these data indicate that the G 3 3 U substitution abrogates the stabilizing influence of the U 32 3 C mutation on ARE folding, consistent with the formation of a noncanonical G 3 -U 32 base pair in the folded structure of the wild type ARE substrate.
Trans-Factor Binding to the ARE(38) RNA Substrate Is Selectively Influenced by the Stability of RNA Folding-The cation titration and thermal denaturation experiments demonstrated that the stability of the ARE(38) folded structure can be altered by changes in the concentration of Mg 2ϩ . Previously, we reported that binding of His 6 -p37 AUF1 to RNA substrates containing the TNF␣ ARE sequence could be inhibited by Mg 2ϩ (33). To quantitatively assess whether the stability of ARE folding was coupled to alterations in trans-factor binding potential, the association of selected ARE-binding proteins with the ARE(38)-Fl substrate was measured in the absence or presence of Mg 2ϩ . His 6 -p37 AUF1 binding to the ARE(38)-Fl RNA substrate was well described by the sequential dimer binding model (Equation 8) in both the absence and presence of Mg 2ϩ . The appropriateness of this model was validated by strong coefficients of determination (R 2 Ͼ 0.99) and random distributions of residuals about each regression solution (Fig. 3A). Comparison of the equilibrium association constants describing both stages of His 6 -p37 AUF1 tetramer assembly on this RNA substrate indicate that the initial dimer binding event (K 1 ) is inhibited 15fold in the presence of 5 mM Mg 2ϩ (Table III; His 6 -p37 AUF1 , cf. 0 mM versus 5 mM MgCl 2 ), whereas the second dimer binding event is also inhibited, although to a lesser degree (4 -5-fold). Association of His 6 -p37 AUF1 with the ARE(38)-Fl substrate in the presence of 1 mM Mg 2ϩ yielded intermediate values for each of these binding constants (Table III). a Solved from thermal denaturation analyses of (Cy3 ϩ Fl)-labeled RNA substrates as described under "Experimental Procedures" and in To determine whether the sensitivity of trans-factor binding to the folded stability of ARE substrates was limited to p37 AUF1 , similar experiments were performed using recombinant forms of the ARE-binding proteins Hsp70 and HuR. Association of His 6 -Hsp70 with the ARE(38)-Fl substrate was well resolved by a binary binding model (Fig. 3B), consistent with the 1:1 protein/RNA stoichiometry previously reported for this interaction (39), but was inhibited ϳ6-fold by the presence of 5 mM Mg 2ϩ (Table III). Although the affinity of His 6 -Hsp70 binding to this RNA substrate was decreased by Mg 2ϩ -stabilized ARE folding, the magnitude of this effect was less dramatic than for His 6 -p37 AUF1 .
The third ARE-binding factor, GST-HuR, formed two detectable complexes with the ARE(38)-Fl substrate in a protein concentration-dependent manner by gel mobility shift assay (Fig. 4A). Both gel mobility shift assay and fluorescence anisotropy analyses demonstrated that GST alone did not interact with the RNA substrate, indicating that the GST moiety does not make significant contributions to the RNA-binding activity of the GST-HuR chimera. Analyses of GST-HuR binding to the ARE(38)-Fl substrate by fluorescence anisotropy were consistent with a cooperative binding model (Equation 10), justified by high coefficients of determination (R 2 Ͼ 0.99) and random distributions of residuals both in the absence and presence of Mg 2ϩ (Fig. 4B). Unlike His 6 -p37 AUF1 and His 6 -Hsp70, however, the affinity of GST-HuR binding to the ARE(38)-Fl substrate was only minimally affected by the presence of Mg 2ϩ (Table  IV), indicating that association of GST-HuR is largely independent of the folded stability of the target ARE sequence. Contrasting these data with the differential sensitivities of His 6 -p37 AUF1 and His 6 -Hsp70 binding to Mg 2ϩ -stabilized ARE folding thus demonstrates that the stability of higher order structures involving AREs can exert a significant influence over their trans-factor binding preferences.
Stabilization of the Folded ARE Structure Inhibits AREdirected mRNA Turnover-The high affinities exhibited by both the mRNA-destabilizing factor, p37 AUF1 , and the mRNA stabilizing protein, HuR, for the unfolded TNF␣ ARE (Tables  III and IV; 0 mM Mg 2ϩ ) indicate that both proteins may effectively compete for binding to this RNA substrate. However, inhibition of p37 AUF1 binding to the folded TNF␣ ARE substrate without significant changes in HuR binding affinity (Tables III and IV; 5 mM Mg 2ϩ ) suggests that stabilization of the TNF␣ ARE hairpin structure perturbs this equilibrium in favor of the mRNA-stabilizing factor. It follows, therefore, that the ability of the TNF␣ ARE to direct rapid mRNA turnover in cis may be closely linked to the stability of ARE folding. To test this hypothesis, sequences encoding the ARE(38), ARE(A 4 3 C), ARE(U 32 3 C), and ARE(G 3 3 U; U 32 3 C) elements were sub-   58 4 a Association binding constants were solved from A t versus ͓protein͔ data sets (Fig. 3) using Equation 8 (for His 6 -p37 AUF1 ) or Equation 9 (for His 6 -Hsp70) and are expressed as mean Ϯ n Ϫ 1 for n Ն 3 or mean Ϯ spread for n ϭ 2.
b Dissociation constants solved as K d ϭ 1/K. cloned into the 3Ј-UTR of a ␤-globin reporter gene under the control of a tetracycline-responsive promoter (Fig. 5A). Following transfection into HeLa/Tet-Off cells, the cellular decay rates of these transcripts were assessed using Dox time course assays (Fig. 5B), followed by a solution of first order decay constants (Fig. 5C). Introduction of the ARE(38) sequence into the 3Ј-UTR of ␤-globin mRNA decreased the mRNA half-life by ϳ14-fold (Table V; cf. ␤G-wt versus ␤G-ARE (38)). Substitution of the more weakly folded ARE(A 4 3 C) element did not significantly alter mRNA decay kinetics relative to the ARE (38) insert. However, the chimeric ␤-globin transcript was stabilized 2.7-fold by introduction of the more strongly folded ARE(U 32 3 C) element relative to the wild type ARE sequence (Table V; cf. ␤G-ARE(U 32 3 C) versus ␤G-ARE (38)). Conceiv-ably, inhibition of ARE-directed mRNA turnover by the U 32 3 C substitution may result either from stabilization of ARE folding or by loss of a primary RNA structural determinant for transfactor binding. However, the U 32 3 C substitution was unable to slow ARE-directed mRNA decay kinetics in the presence of the compensatory G 3 3 U mutation (Table V; cf. ␤G-ARE(U 32 3 C) versus ␤G-ARE(G 3 3 U; U 32 3 C)), indicating that mutation of U 32 3 C inhibits mRNA decay through stabilization of ARE folding via interaction with G 3 (Fig. 1) and not simply by abrogation of trans-factor binding contacts through U 32 . Taken together, these data show that a mutation that stabilizes the folded structure of the TNF␣ ARE (Tables I and  II) also decreases the ability of this ARE to direct rapid mRNA turnover in cells, consistent with a model whereby the stability of ARE folding may influence the equilibrium between various trans-acting factors competing for this cis-acting element. Sequences Flanking the TNF␣ ARE May Contribute to a Conserved, Extended Secondary Structure-If structural presentation of the TNF␣ ARE is a significant determinant of trans-factor recognition in vivo, it follows that features directing the folding of this element would probably be subject to evolutionary pressure. To address this possibility, we first compared mammalian TNF␣ mRNA sequences spanning and flanking the ARE. By ClustalW alignment, the core ARE was completely conserved (Fig. 6A, capital letters), with the exception of an A3 G substitution at position 1178 of the murine sequence. The core ARE sequence is sufficient for high affinity binding of p37 AUF1 (33) (and Fig. 3A) and induction of rapid mRNA turnover in cis (55). However, sequences immediately up-and downstream of the core ARE also exhibited a high degree of conservation among mammals, raising the possibility that they may also contribute to the regulation of trans-factor recognition by this element. By mFold, these flanking sequences are predicted to form extensive intramolecular hybrids, which would position the core ARE sequence at the apex of an extended stem-loop (Fig. 6B). Sequence divergence between species is predicted to exert minimal influence over this general structure, based on three observations. First, many species-specific substitutions near the base of the stem are clustered within internal loops of the mFold model (Fig. 6B, human bases G 1136 , U 1216 , A 1222 , C 1223 , G 1225 , and G 1226 ). Second, substitution of the hybridized human residue A 1145 3 G does not abrogate base pair formation, and third, some species present clustered sequence variations that are predicted to restore folding stability in a compensatory manner. For example, in goat and sheep TNF␣ mRNA, the nucleotide corresponding to G 1161 in humans is altered to A, thus precluding formation of a base pair contact with C 1198 . However, these species exhibit concomitant substitutions of A 1199 3 U and G 1200 3 A. As such, hybrid energetics may be restored by bulging C 1198 from the helix, thus permitting pairing of bases A 1161 with U 1199 , U 1160 with A 1200 , and additionally U 1159 with A 1201 .
Predicted RNA structures involving the murine TNF␣ ARE exhibit some minor differences relative to the other mamma- The concentration of GST-HuR required for half-maximal RNA binding (͓GST-HuR͔1 ⁄2 ) and the Hill coefficient (h) were calculated from A t versus ͓GST-HuR͔ data sets (Fig. 4B) using Equation 10 and are expressed as the mean Ϯ n Ϫ 1 for n ϭ 4 or mean Ϯ spread for n ϭ 2. lian examples, largely due to the loss of the GUG sequence following position 1144 of the murine mRNA (Fig. 6C). Whereas mFold models of the murine TNF␣ ARE region returned a punctuated stem-loop structure similar to that predicted for the corresponding domains from other mammals, net hybrid formation was slip-shifted by four bases (Fig. 6, B and C, compare G 1132 of the human sequence with G 1115 of the murine sequence). As a result, the center of the core ARE sequence is positioned slightly 5Ј of the central loop. However, several suboptimal structures with similar energetics were also predicted, an example of which is depicted in Fig. 6D. In most cases, base pair contacts involving sequences flanking the core ARE were invariant; rather, slight differences in predicted folding energy were the result of alternative base pairing in the core ARE region.
The high degree of sequence conservation flanking the core ARE sequence from mammalian TNF␣ mRNAs suggests one or more important regulatory roles for these domains. Phylogenetic comparisons of structural models indicate that one such role may be to direct the conformational presentation of the core ARE. Given that even modest stabilization of ARE folding significantly inhibits the binding activity of p37 AUF1 and, to a lesser extent, Hsp70 ( Fig. 3 and Table III) while only minimally influencing HuR binding ( Fig. 4B and Table IV), these data support the hypothesis that local RNA folding potential may constitute an important determinant of ARE function or regulation in vivo, by presenting a novel mechanism for discrimination of trans-factor binding events.

DISCUSSION
From this work, several lines of evidence indicate that the TNF␣ ARE may adopt a stem-loop conformation in solution analogous to the model predicted by mFold (Fig. 1A). First, the ARE(38) substrate is partially protected from digestion by RNase A in the presence of Mg 2ϩ , consistent with occlusion of single-stranded regions in the presence of the cation (Fig. 1B). Second, nucleotides remaining sensitive to RNase A in the presence of Mg 2ϩ map almost exclusively to unpaired uridylate residues in the folded model. Third, FRET analyses indicate a limiting distance of 39 Å between fluorophores linked to the 5Ј and 3Ј termini of the ARE(38) substrate in the folded state (Fig.  2E). Fourth, single nucleotide substitutions predicted to stabilize (U 32 3 C) or destabilize (A 4 3 C) the ARE stem-loop structure, as well as a double mutant ARE (G 3 3 U; U 32 3 C) predicted to abrogate stabilization by U 32 3 C, all yielded these anticipated effects, demonstrated by (i) changes in the Mg 2ϩ concentrations required to stabilize RNA substrate folding (Table I) and (ii) changes in the thermodynamic stability of each RNA substrate (Table II). Finally, sequences spanning and flanking the TNF␣ ARE are highly conserved among mammals, with substitutions and covariance between species consistent with maintenance of the TNF␣ core ARE near the terminal loop of an extended hairpin structure (Fig. 6).
Additional details, however, raise the likelihood that the stem-loop model presented in Fig. 1A may be representative of a limited population of similar structures involving the TNF␣ ARE. For example, mFold also predicted a small number of suboptimal structures retaining overall stem-loop character but slip-shifted relative to the most stable candidate ( Fig. 6D and data not shown). Shifting base pair contacts by only 2 or 3 nucleotides within a subset of ARE hairpin structures would exert fairly minor influences on the average distance between the termini of the ARE(38) substrate in solution (and hence minimally influence E FRET ) but may  Table V. account for occasional exposure of U 31 to single-stranded nucleases (Fig. 1B). Furthermore, the mFold models do not account for potential energetic contributions from base stacking within symmetrical bulges or formation of non-Watson-Crick U-U base pairs (56,57). Finally, with the specificity of ARE folding driven by small numbers of contiguous A-U or G-U base pairs, it is likely that ARE stem-loop structures would exhibit rapid conformational dynamics in solution (58). Although the core ARE is flanked by sequences capable of forming a more stable hybrid within the context of the TNF␣ mRNA 3Ј-UTR, placement of the ARE near the top of the predicted stem-loop may yet permit dynamic conformational "breathing" within the core mRNA-destabilizing sequence, particularly given the paucity of contiguous base pairs in this region and the lack of G-C interactions.
In this study, stabilizing the folded state of RNA substrates containing the core TNF␣ ARE by addition of Mg 2ϩ inhibited p37 AUF1 binding and oligomerization activities (Table III) concomitant with occlusion of single-stranded RNA sequence determinants contributing to protein recognition. In this manner, accessibility of ARE target sites for trans-acting factors may be considered in terms of a partition function, where the opportunity for associative RNA-protein contacts is influenced by both the length of RNA required for a given RNA-protein interface and the stability and/or dynamics of folded RNA structures involving this site. By this model, the binding activity of FIG. 6. Conservation of sequences and predicted structures of the TNF␣ mRNA spanning the ARE. A, the human TNF␣ 3Ј-UTR sequence spanning and flanking the core ARE was aligned using ClustalW with corresponding sequences from goat, sheep, mouse, dolphin, and whale. Nucleotide numbers are relative to the TNF␣ mRNA translational initiation codon from each species, and the core ARE sequence is capitalized. Bases conserved with the human sequence are given by dots, dashes indicate bases deleted in individual species, and base substitutions between species are listed where applicable. Data base sequences were extracted from the following accession numbers: human, Homo sapiens, NM_000594; goat, Capra hircus, X148282; sheep, Ovis aries, X56756; mouse, Mus musculus, X02611; dolphin, Tursiops truncates, AB049358; beluga whale, Delphinapterus leucas, AF320323. B, the optimal mFold predicted structure for the human TNF␣ 3Ј-UTR sequence flanking and spanning the ARE. The core ARE sequence is boxed. Base substitutions are indicated for goat (1), sheep (2), dolphin (3), and whale (4). C, the optimal mFold predicted structure for the mouse TNF␣ 3Ј-UTR domain homologous to the human sequence in B. Base substitutions in the human sequence are given. del denotes a base that is deleted in the human sequence, whereas Ͼ indicates the location of a GUG trinucleotide found in the human TNF␣ mRNA. D, a suboptimal mFold prediction for the murine TNF␣ ARE hairpin structure. Base pairs involving sequences further 5Ј and 3Ј are identical to those shown in C. p37 AUF1 is likely to be highly sensitive to the RNA folding potential of ARE substrates, since p37 AUF1 requires relatively large (Ͼ20-nucleotide) AU-rich sequences for high affinity binding (32). By contrast, ARE-binding proteins capable of forming stable complexes with significantly smaller RNA substrates may be less sensitive to the folded stability of ARE substrates. For example, the neuronal HuR-related protein, HuD, binds a 13-nucleotide, AU-rich RNA substrate with K d Ͻ 10 nM (22), whereas the zinc finger domain of TTP forms complexes of similar affinity with RNA substrates as small as 9 nucleotides (26). The strong influence of ARE folding on the binding activity of His 6 -p37 AUF1 , coupled with the very modest inhibition of GST-HuR binding to the folded ARE substrate (this work) are supportive of this model. Furthermore, the inhibition of cellular ARE-directed mRNA turnover resulting from a modest increase in the stability of ARE folding (Fig. 5 and Table V) suggests that some ARE-binding proteins also show sensitivity to local RNA structure in the cellular environment. Given the growing number of ARE-binding proteins identified to date, local RNA structures involving these elements may thus constitute an important determinant of transfactor selectivity.
Current models of ARE-directed mRNA decay indicate that association of some ARE-binding factors serves to target nucleases (17,18) or other ancillary proteins (59) to the RNA substrate. However, given that AREs from some mRNAs extend beyond 120 nucleotides in length (3) and that flanking sequences may be conserved to a much greater degree than required solely for maintenance of local secondary structure (e.g. Fig. 6A), it is likely that assembly of multisubunit, transacting complexes may include multiple RNA-protein interactions near the ARE. Accordingly, alterations in local RNA topology could potentially influence ARE-directed mRNA turnover kinetics by promoting or restricting any number of RNA-protein binding events. A recent report showing that ARE-directed mRNA decay kinetics can be modulated by hybridization of antisense sequences adjacent to the ARE lends further support for this model (60). The relationship between local ARE structure and regulation of mRNA decay is complicated, however, by observations that ARE-binding proteins themselves are capable of remodeling local RNA structure. For example, the zinc finger domains of TTP and the related protein TIS11d both retain bound RNA substrates in elongated conformations (26,27). In contrast, the two N-terminal RNA recognition motifs of HuD induce a bend in associated ARE substrates (21). AUF1 proteins present a more complex case, in that both p37 AUF1 and p40 AUF1 structurally condense RNA substrates upon binding (38,61). However, in THP-1 monocytic leukemia cells, phorbol ester-induced stabilization of ARE-containing mRNAs is accompanied by loss of phosphate from Ser 83 and Ser 87 of polysome-associated p40 AUF1 (31). Although the TNF␣ ARE binds and induces oligomerization of the unphosphorylated and Ser 83 -, Ser 87 -, and (Ser 83 ϩ Ser 87 )-bis-phosphorylated forms of p40 AUF1 , only the bis-phosphorylated form retains the bound ARE substrate in an elongated conformation (61). As such, contributions of bis-phosphorylated p40 AUF1 to the rapid turnover of TNF␣ mRNA may function, at least in part, through promotion of proximal protein binding events by unfolding RNA sequences flanking the ARE. We speculate that this may be particularly significant in the case of the TNF␣ ARE, where AUF1-induced RNA remodeling at the core ARE sequence may weaken hybrids formed between the 5Ј and 3Ј extensions of the ARE (Fig. 6B), thus exposing adjacent U-rich RNA sequences for trans-factor recognition.
In summary, the interdependence of local ARE structure and trans-factor binding have the potential to modulate ARE-di-rected mRNA decay activity not only by selective recruitment of ARE-binding proteins but also by the ability of such factors to remodel flanking RNA structures. As a result, we speculate that gene-specific control of ARE-directed mRNA decay may ultimately be possible by modulating the stability or dynamics of local ARE folding events.