Folding of A+U-rich RNA Elements Modulates AUF1 Binding

In mammals, A+U-rich elements (AREs) are potentcis-acting determinants of rapid cytoplasmic mRNA turnover. Recognition of these sequences by AUF1 is associated with acceleration of mRNA decay, likely involving recruitment or assembly of multi-subunit trans-acting complexes. Previously, we demonstrated that AUF1 deletion mutants formed tetramers on U-rich RNA substrates by sequential addition of protein dimers (Wilson, G. M., Sun, Y., Lu, H., and Brewer, G. (1999)J. Biol. Chem. 274, 33374–33381). Here, we show that binding of the full-length p37 isoform of AUF1 to these RNAs proceedsvia a similar mechanism, allowing delineation of equilibrium binding constants for both stages of tetramer assembly. However, association of AUF1 with the ARE from tumor necrosis factor (TNFα) mRNA was significantly inhibited by magnesium ions. Further fluorescence and hydrodynamic experiments indicated that Mg2+ induced or stabilized a conformational change in the TNFα ARE. Based on the solution of parameters describing both the protein-RNA and Mg2+-RNA equilibria, we present a dynamic, global equilibrium binding model describing the relationship between Mg2+ and AUF1 binding to the TNFα ARE. These studies provide the first evidence that some AREs may adopt higher order RNA structures that regulate their interaction withtrans-acting factors and indicate that mRNA structural remodeling has the potential to modulate the turnover rates of some ARE-containing mRNAs.

In eukaryotes, the cytoplasmic concentration of an mRNA is a critical determinant of its potential for translation and hence the production rate of the encoded gene product. As such, the rate of cytoplasmic mRNA decay influences both the timing and level of expression of many gene products, involving either defined constitutive mRNA decay rates or modulation of mRNA turnover rates in response to external stimuli (reviewed in Refs. 1 and 2).
Many mammalian mRNAs encoding oncoproteins, cytokines/ lymphokines, inflammatory mediators, and G protein-coupled receptors are unstable, owing to the presence of AϩU-rich elements (AREs) 1 in their 3Ј-untranslated regions (3)(4)(5)(6). AREs comprise a diverse family of U-rich RNA sequences varying in length from 40 to 150 bases and frequently containing one or more motifs of the form AUUUA, which may be overlapping or dispersed (7). In general, mRNA turnover mediated by AREs is characterized by rapid shortening of the poly(A) tail followed by degradation of the mRNA body (8 -10). Although several proteins have been identified that may bind AϩU-rich RNA sequences (reviewed in Ref. 11), the mechanisms linking association of trans-acting factors with accelerated mRNA decay remain largely unknown.
The ARE-binding protein that has been most extensively characterized is AUF1. The AUF1 gene encodes a family of four protein isoforms generated by alternative pre-mRNA splicing that are denoted by their apparent molecular weights as p37 AUF1 , p40 AUF1 , p42 AUF1 , and p45 AUF1 (12). Normally, the p42 and p45 isoforms appear exclusively nuclear (13,14), likely owing to an isoform-specific protein sequence that binds scaffold attachment factor-B, a protein associated with the nuclear matrix (14). However, several lines of evidence demonstrate that association of cytoplasmic p37 AUF1 and/or p40 AUF1 with an ARE is associated with acceleration of mRNA turnover. First, a protein complex containing cytoplasmic p37 AUF1 and p40 AUF1 is sufficient to destabilize polysomal c-myc mRNA in vitro (15). Second, AUF1 associates directly with AϩU-rich RNA sequences, and the affinity of recombinant p37 AUF1 for an ARE closely correlates with the potential of the ARE to destabilize an mRNA in cis (16). Third, inhibition of the p38 mitogenactivated protein kinase pathway stabilizes ARE-containing mRNAs (17)(18)(19), concomitant with the loss of an ARE binding activity containing AUF1 (19). Fourth, the efficiency of AREdirected mRNA turnover is compromised in cells expressing low levels of endogenous p37 AUF1 and p40 AUF1 (20,21) or following sequestration of AUF1 by treatment with hemin (22). Conversely, an increase in AUF1 protein levels during congestive heart failure accompanies a decrease in levels of ␤ 1 -adrenergic receptor mRNA (23,24). The ␤ 1 -adrenergic receptor transcript also contains an ARE in its 3Ј-untranslated region that associates with AUF1 (24).
In previous studies, we demonstrated that the association of recombinant AUF1 with an ARE in vitro proceeded by sequential binding of AUF1 dimers involving protein-protein and protein-RNA interactions (25), generating an oligomeric AUF1 complex on the ARE (25,26). Based on observations that cellular AUF1 is associated with other cellular factors (13) includ-ing the translation initiation factor eIF4G, poly(A)-binding protein, and the heat shock proteins Hsp70 and Hsc70 (27), we proposed that the oligomerization of AUF1 on an ARE nucleates the assembly of a trans-acting, mRNA-destabilizing complex on the RNA (11,25). As such, the potential of an ARE to associate with AUF1 and promote protein oligomerization may be a critical determinant of mRNA turnover efficiency. In this work, we have continued to define molecular mechanisms contributing to the recognition of AREs by AUF1. Using a fluorescence anisotropy-based assay for RNA-protein binding, we observed that the association of AUF1 to the ARE from tumor necrosis factor ␣ (TNF␣) mRNA was inhibited by magnesium ions, whereas AUF1 binding to a similarly sized polyuridylate sequence was largely unaffected. We present evidence that the TNF␣ ARE exhibits a conformational change upon association with Mg 2ϩ with the net result of restricting RNA flexibility and shortening the distance between the 5Ј-and 3Ј-termini of the RNA substrate. Based on independent assessments of AUF1-ARE and Mg 2ϩ -ARE binding equilibria, we have constructed a model for inhibition of AUF1 binding and oligomerization to the TNF␣ ARE in the presence of Mg 2ϩ . To our knowledge, this represents the first indication that higher order RNA structures may regulate the association of trans-acting factors with an ARE.

EXPERIMENTAL PROCEDURES
RNA Substrates-The sequences and fluorophore positions of RNA substrates (2Ј-hydroxyl) used in this study are listed in Table I. All substrates containing 5Ј-fluorescein (Fl) or 5Ј-cyanine 3 (Cy3) groups were synthesized by Dharmacon Research (Boulder, CO), then 2Ј-Odeprotected according to the manufacturer's instructions (28) and evaporated to dryness in a Speed-Vac. Deprotected RNA substrates were dissolved in 10 mM Tris⅐HCl (pH 8.0) and quantified by A 260 , with estimates of RNA extinction coefficients calculated as described (29). For fluorescein-conjugated RNA probes, A 260 was corrected by quantitation of the fluorescein moiety at 493 nm with ⑀ 493, Fl ϭ 74,600 M Ϫ1 ⅐cm Ϫ1 and ⑀ 260, Fl ϭ 26,000 M Ϫ1 ⅐cm Ϫ1 as described (30). For Cy3conjugated RNAs, A 260 was similarly corrected by quantitation of the Cy3 moiety at 552 nm using ⑀ 552, Cy3 ϭ 150,000 M Ϫ1 ⅐cm Ϫ1 and ⑀ 260, Cy3 ϭ 12,000 M Ϫ1 ⅐cm Ϫ1 (extinction coefficients for Cy3 provided by Amersham Pharmacia Biotech).
The RNA substrates TNF␣ ARE and P-TNF3Ј were synthesized lacking fluorescent labels (Dharmacon) and then 2Ј-O-deprotected and quantified by A 260 as described above. P-TNF3Ј contains a phosphate group conjugated at the 5Ј-end. Fluorescein labels were added to the 3Ј-ends of these RNAs by periodate oxidation and conjugation with Alexa Fluor 488 hydrazide (Molecular Probes, Eugene, OR) as described (31) to generate TNF␣ ARE-Fl and P-TNF3Ј-Fl. Unconjugated fluorophore was removed from the labeled RNA preparations by passage through two Quick Spin G-25 spin columns (Roche Molecular Biologicals). The yields of RNA and conjugated fluorophore were monitored by absorbance as described above, with fluorophore coupling efficiencies typically Ͼ95%.
The double-labeled RNA oligonucleotide Cy-TNF-Fl was constructed by ligation of RNA substrates Cy-TNF5Ј and P-TNF3Ј-Fl using T4 DNA ligase and a bridging antisense DNA oligonucleotide as described (32). Following ligation, the reaction was heated to 95°C for 5 min and quickly cooled on ice to denature DNA-RNA hybrids. The bridging DNA oligonucleotide was then digested with RQ1-DNaseI. Ligated RNA molecules were purified by denaturing polyacrylamide gel electrophoresis and were recovered as described (33). The yield of Cy-TNF-Fl RNA was estimated by measurement of total fluorescence intensity of the Cy3 moiety ( ex ϭ 535 nm; em ϭ 580 nm) and comparison with a standard curve of Cy-TNF␣ ARE RNA. This assumes that the fluorescence quantum yield of the 5Ј-Cy3 group was not significantly affected by the addition of the fluorescein moiety to the 3Ј-end of the RNA, because placement of the fluorophores on opposite ends of the RNA substrate and the overall hydrophilicity of RNA make it unlikely that the 3Ј-Fl could substantially alter the chemical environment of the 5Ј-Cy group.
Preparation of Recombinant His 6 -p37 AUF1 -The complete coding sequence of p37 AUF1 was excised from pTrcHisB-p37 AUF1 (16) by digestion with Acc65I ϩ HindIII and subcloned into similarly digested pBAD/ HisB (InVitrogen, Carlsbad, CA) to generate pBAD/HisB-p37 AUF1 . Recombinant His 6 -p37 AUF1 was expressed from this plasmid in Esche-richia coli TOP10 cells induced with 0.0002% arabinose for 4 h. Cells were disrupted by four rounds of sonication as described (34). However, inclusion of polyethylene glycol-6000 in the lysis buffer (5% final) abrogated the need for freeze-thaw cycles. Purification of His 6 -p37 AUF1 was performed by Ni 2ϩ affinity chromatography using a HiTrap Chelating affinity column (Amersham Pharmacia Biotech). The column (1 ml) was prerinsed at 120 ml/h with 8 column volumes of deionized water, charged with 6 volumes of 20 mM NiCl 2 , rinsed with 8 volumes of deionized water and 6 volumes of native imidazole elution buffer (10 mM sodium phosphate, 500 mM NaCl, 300 mM imidazole, pH 6.3), and equilibrated with 8 volumes buffer 1 (50 mM sodium phosphate, 500 mM NaCl, 20 mM imidazole, pH 8.0). The cleared bacterial cell lysate was applied at 60 ml/h and washed with buffer 1 until A 280 reached base line. His 6 proteins were eluted in native imidazole elution buffer at 30 ml/h with fractions monitored by A 280 . Peak fractions were pooled, buffer exchanged into 10 mM Tris⅐HCl (pH 7.5) using HiTrap Desalting columns (Amersham Pharmacia Biotech) as recommended by the manufacturer, and concentrated using a Centricon YM-30 concentrator (Millipore, Bedford, MA). Protein concentrations were determined by comparison of Coomassie Blue-stained SDS-polyacrylamide gels containing recombinant His 6 -p37 AUF1 and a titration of bovine serum albumin. Quantitation of band intensities was performed using the Kodak EDAS 120 gel documentation system and accompanying onedimensional image analysis software (Eastman Kodak Co.).
Evaluation of RNA-Protein and RNA-Cation Binding Equilibria by Fluorescence Anisotropy-Association of recombinant His 6 -p37 AUF1 with fluorescent RNA substrates was monitored by changes in the anisotropy of the fluorescein moiety following protein binding. Anisotropy is related to the rotational relaxation time of the fluorophore in solution (35,36), which, for a linear polymer like RNA, is dependent on gross molecular volume and/or intramolecular segmental motion under conditions of constant temperature and viscosity (35). Fluorescence anisotropy measurements were made using the Beacon 2000 Variable Temperature Fluorescence Polarization System (Panvera, Madison, WI) equipped with fluorescein excitation (490 nm) and emission (535 nm) filters. Binding reactions were performed with a range of His 6 -p37 AUF1 protein concentrations and 0.2 nM fluorescein-labeled RNA in a final volume of 100 l containing 10 mM Tris⅐HCl (pH 8.0), 100 mM potassium acetate, 2 mM dithiolthreitol, 0.1 mM spermine, and 0.1 g/l acetylated bovine serum albumin. Heparin (1 g/l) was included in RNA-protein binding reactions to inhibit nonspecific RNA binding activity. This concentration of heparin effectively prevented association of His 6 -p37 AUF1 (up to 250 nM protein dimer) with a fluorescent RNA substrate encoding a fragment of the ␤-globin coding region (Ref. 25 and data not shown). Magnesium acetate was used as a source of Mg 2ϩ ions where indicated. Samples lacking Mg 2ϩ contained 0.5 mM EDTA unless otherwise noted. For equilibrium binding experiments, the polarimeter was operated in static mode, with each sample read as blank prior to addition of fluorescent RNA substrates. Following probe addition, samples were incubated for 1 min at 25°C before anisotropy was measured. Preliminary on-rate analyses demonstrated that anisotropic equilibrium was reached within 10 -20 s at this temperature for all binding equilibria described in this report (data not shown). Data points represent the mean of 10 anisotropy measurements for each binding reaction.
For evaluation of Mg 2ϩ binding events involving RNA in the absence of protein, Mg 2ϩ titrations were monitored by fluorescence ansiotropy as described above but in buffers lacking heparin. Parallel titrations including heparin (1 g/l) yielded consistent equilibrium constants for data sets between 0 and 5 mM Mg 2ϩ but with small increases in absolute anisotropy values (data not shown). At higher Mg 2ϩ concentrations, however, more significant nonspecific increases in anisotropy were observed in the presence of heparin, possibly because of increases in sample viscosity resulting from the formation of heparin⅐Mg 2ϩ aggregates.
Concomitant with measurement of anisotropy values, total fluorescence intensity was monitored for each sample to detect changes in the fluorescence quantum yield of the fluorescein moiety as described (25). For all experiments described in this work, total fluorescence intensity did not vary significantly as a result of protein or cation binding (see Fig. 1, A and B, and data not shown). Accordingly, the total measured fluorescence anisotropy (A t ) of a mixture of fluorescent species was interpreted by simple additivity using Equation 1.
A i represents the intrinsic anisotropy of each fluorescent species and f i its fractional concentration (35,37,38). Using C-terminal deletion mutants of AUF1, a previous study demonstrated that AUF1 association with short U-rich RNA substrates was well described by sequential binding of AUF1 dimers (25) Here, A R , A P 2 R , and A P 4 R represent the intrinsic anisotropy values of the free RNA (R), AUF1 dimer-bound RNA (P 2 R), and AUF1 tetramerbound RNA (P 4 R), respectively. K 1 is the constant describing the R ϩ P 2 º P 2 R equilibrium, whereas K 2 describes the P 2 Rϩ P 2 º P 4 R equilibrium (see Fig. 1E). Additional algorithms employed in analysis of binding equilibria by fluorescence anisotropy are described in the text where applicable. The application of binding algorithms to experimental data was performed by nonlinear regression using PRISM, version 2.0 (GraphPad, San Diego, CA). The validity of all mathematical models was evaluated by the coefficient of determination (R 2 ) and analysis of residual plot nonrandomness to detect any bias for data subsets (PRISM). Where necessary, pair-wise comparisons of sum-of-squares deviations between mathematical models were performed using the F test (PRISM), with differences exhibiting p Ͻ 0.05 considered significant.
Analysis of RNA Folding by Fluorescence Resonance Energy Transfer-Changes in the distance between the 5Ј-and 3Ј-termini of doublelabeled RNA substrates were monitored by fluorescence resonance energy transfer (FRET), using a modification of the method of Walter et al. (39). Reactions were assembled as described above for evaluation of cation binding events involving RNA in the absence of protein. All measurements of fluorescence intensity were made using the Beacon 2000 fluorescence polarimeter (Panvera) operating in direct fluorescence mode. Prior to addition of fluorescent RNA substrates, background fluorescence emission ( ex ϭ 490 nm) was measured for each sample at 535 and 580 nm, given by B 535 and B 580 , respectively. Following probe addition, samples were incubated for 1 min at 25°C prior to measurement of total fluorescence intensity at the same wavelengths, giving F 535 and F 580 . FRET efficiency in this system was approximated by the ratio Q ϭ (F 580 Ϫ B 580 )/(F 535 Ϫ B 535 ). For analysis of cation-RNA binding equilibria, Q was normalized to an approximation of the FRET efficiency in the absence of cation (Q 0 ), to give the relative FRET efficiency (Q Ϫ Q 0 )/Q 0 .
Gel Filtration Chromatography-For size fractionation of RNA substrates, a 0.9 ϫ 30-cm column loaded with Superdex 75 (Amersham Pharmacia Biotech) was washed with 0.1 M NaOH for 2 h prior to equilibration with diethylpyrocarbonate-treated gel filtration buffer (10 mM Tris⅐HCl, 100 mM potassium acetate, 2 mM dithiolthreitol, pH 8.0) containing or lacking 5 mM magnesium acetate. RNA samples (1 nmol) were loaded in 0.2 ml of gel filtration buffer containing 2% dextrose and either 0.5 mM EDTA or 5 mM magnesium acetate. Elution of RNA was monitored by A 254 . The column was calibrated by monitoring the elution of protein standards (Sigma) by A 280 : soybean trypsin inhibitor (20 kDa), cytochrome c (12.4 kDa), and aprotinin (6.4 kDa). Column void volume was determined using blue dextran.

RESULTS
Association of Recombinant AUF1 Protein with the TNF␣ ARE Is Inhibited by Mg 2ϩ -Previously, we demonstrated that a recombinant p37 AUF1 deletion mutant lacking the C-terminal 29 amino acid residues (His 6 -p37 AUF1 -(1-257)) interacts with the core ARE sequence from TNF␣ mRNA, encoded by the RNA substrate Fl-TNF␣ ARE (Table I), by monitoring changes in the anisotropy of the fluorescein moiety of the RNA resulting from protein binding (25). This RNA sequence is a potent mRNA-destabilizing element (40) and contributes to the extreme instability of TNF␣ mRNA in vivo (41,42). Like the full-length His 6 -p37 AUF1 , His 6 -p37 AUF1 (1-257) contains an Nterminal dimerization domain (26) and exists as a dimer in solution (25). In the presence of an RNA substrate containing the TNF␣ ARE or the polyuridylate sequence contained within Fl-U 32 (Table I), His 6 -p37 AUF1 -(1-257) dimers associate sequentially to form a tetrameric protein complex on the RNA (25). Neither the dimerization (26) nor the RNA-binding activities (25) of AUF1 proteins are affected by the presence of the N-terminal His 6 tag.
The association of recombinant His 6 -p37 AUF1 with the RNA substrates Fl-TNF␣ ARE and Fl-U 32 was analyzed by measurement of the total fluorescence intensity and fluorescence anisotropy of binding reactions containing the fluorescent RNA substrates and titrations of recombinant protein (Fig. 1). For both the Fl-TNF␣ ARE and Fl-U 32 substrates, increasing the concentration of His 6 -p37 AUF1 had little effect on total fluorescence intensity (Fig. 1, A and B), indicating that the fluorescence quantum yield was not significantly affected by the presence of His 6 -p37 AUF1 . As such, the contribution of each fluorescent species (RNA or RNA⅐protein complexes) to the total measured anisotropy (A t ) could be defined as the product of its intrinsic anisotropy (A i ) and its fractional concentration (f i ), described in Equation 1 (see "Experimental Procedures"). Adapting this algorithm to the RNA-dependent formation of AUF1 tetramers by sequential association of protein dimers ( Fig. 1E) yields Equation 2 (25). In the absence of Mg 2ϩ , the association of His 6 -p37 AUF1 with both Fl-TNF␣ ARE and Fl-U 32 were well described by sequential dimer binding ( Fig. 1, C and D, solid circles). This was affirmed by strong coefficients of determination (R 2 ϭ 0.9967 for Fl-TNF␣ ARE; R 2 ϭ 0.9941 for Fl-U 32 ) and random positioning of residuals across the entire range of protein concentrations tested. The presence of the second binding phase, defined by K 2 (Fig. 1E), was further indicated by poor regression of each data set using a binary binding model described by Equation 3 (data not shown).
where A R-free and A R-bound represent the intrinsic anisotropy values of a fluorescent RNA substrate in the unbound and protein-associated states, respectively, defined by a single association constant K. Comparisons between the binary and sequential dimer binding models using the F test supported the two-stage formation of protein tetramers described by Equa-

Cy-TNF5Ј
Cy-GUGAUUAUUUAUUAUUUA Fl-GAUCUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUA a "Fl" and "Cy" indicate the positions of the fluorescein and cyanine 3 moieties, respectively, for each RNA substrate. b "P" indicates the presence of a 5Ј-phosphate group on the substrates P-TNF3Ј and P-TNF3Ј-Fl.  Table II, along with solutions for equilibrium constants K 1 and K 2 . In each case, the intrinsic anisotropy values increased with protein load (A R Ͻ A P2R Ͻ A P4R ), consistent with the restriction of fluorophore mobility by increasing molecular volume (35,36). In addition, the association of an AUF1 dimer with a free RNA substrate, given by K 1 , exhibited much higher affinity than the association of a subsequent protein dimer with a P 2 R complex, given by K 2 , for both RNA substrates. The inclusion of 5 mM Mg 2ϩ in the RNA-protein binding reactions elicited different consequences in reactions containing the Fl-TNF␣ ARE versus the Fl-U 32 substrate. In binding reactions containing Fl-TNF␣ ARE, Mg 2ϩ repressed changes in anisotropy as a function of protein concentration, indicating that AUF1 binding to this RNA substrate was inhibited by the presence of the cation (Fig. 1C, open circles). By contrast, the presence of 5 mM Mg 2ϩ induced little change in the anisotropy of binding reactions containing Fl-U 32 (Fig. 1D, open circles). Because all binding reactions were performed in a background of 100 mM potassium acetate, the observed changes in anisotropy were unlikely to be due solely to the relatively small change in ionic strength caused by inclusion of 5 mM magnesium acetate. were monitored as a function of AUF1 concentration to detect changes in the fluorescence quantum yield of either probe in response to protein binding. Fluorescence anisotropy was also measured for each binding reaction (C and D). Anisotropy data sets for samples lacking Mg 2ϩ were resolved by nonlinear least squares regression using Equation 2 with A R held constant (Fl-TNF␣ ARE, A R ϭ 0.031; Fl-U 32 , A R ϭ 0.029) and A P2R , A P4R , K 1 , and K 2 unfixed (solid lines). Residual plots (bottom panels) were prepared by subtraction of the regression-derived anisotropy value (A c ) from the measured anisotropy value (A t ) for each data point to detect any bias for subsets of experimental data. Equation 2 was derived from a model for association of AUF1 with U-rich RNA substrates by sequential binding of AUF1 dimers (E), under conditions of constant fluorescence quantum yield.

TABLE II
Solution of anisotropy and equilibrium binding constants for His 6 -p37 AUF1 association with RNA substrates (-Mg 2ϩ ) Solutions for the intrinsic anisotropy of the free RNA substrates (A R ) are based on measurements of anisotropy in the absence of added protein (n ϭ 9 for Fl-TNF␣ ARE; n ϭ 4 for Fl-U 32 ). The remaining anisotropy and equilibrium binding constants (defined in the text) were derived from triplicate AUF1 titration experiments performed in the absence of Mg 2ϩ as shown in Fig. 1 In addition to variations in AUF1 binding, however, Mg 2ϩ also significantly increased the intrinsic anisotropy of the Fl-TNF␣ ARE RNA substrate in the absence of AUF1 (cf. 0.031 Ϯ 0.002, n ϭ 9 in 0.5 mM EDTA versus 0.047 Ϯ 0.001, n ϭ 6 in 5 mM Mg 2ϩ ). The anisotropy of unconjugated fluorescein was completely independent of Mg 2ϩ concentration (cf. 0.0095 Ϯ 0.0003, n ϭ 4 in 0.5 mM EDTA versus 0.0092 Ϯ 0.0003, n ϭ 3 in 5 mM Mg 2ϩ ), indicating that the increase in anisotropy of the Fl-TNF␣ ARE substrate is dependent on the presence of the RNA moiety. Although Mg 2ϩ also increased the intrinsic anisotropy of Fl-U 32 , the effect was less dramatic (cf. 0.029 Ϯ 0.002, n ϭ 4 in 0.5 mM EDTA versus 0.036 Ϯ 0.002, n ϭ 4 in 5 mM Mg 2ϩ ). The large increase in the intrinsic anisotropy of Fl-TNF␣ ARE relative to Fl-U 32 in the presence of Mg 2ϩ demonstrates that restriction of fluorescein mobility in these substrates by Mg 2ϩ is also sequence-specific for the RNA. Because the anisotropy of a fluorescent RNA is dependent on both its molecular volume and segmental motion when temperature and viscosity are held constant (35,36), these data suggest either that the Fl-TNF␣ ARE substrate may form multimers in the presence of Mg 2ϩ , or that RNA structures presenting limited flexibility near the 5Ј-end may be induced or stabilized by the cation. The next series of experiments was designed to distinguish between these possibilities.
Mg 2ϩ Induces or Stabilizes Unimolecular Structural Changes in the TNF␣ ARE-Several experiments were performed to characterize structural events that might contribute to the Mg 2ϩ -induced decrease in fluorescein mobility observed with the Fl-TNF␣ ARE RNA substrate. First, gel filtration chromatography of RNA substrates through a Superdex 75 column demonstrated that the apparent molecular mass of Fl-TNF␣ ARE was not significantly altered in the presence of 5 mM Mg 2ϩ (Fig. 2). In both the presence and absence of Mg 2ϩ , Fl-TNF␣ ARE eluted at ϳ13 to 14 kDa, close to its calculated molecular mass of 12.5 kDa. By contrast, hybridization of Fl-TNF␣ ARE to an antisense DNA oligonucleotide dramatically decreased the elution volume, presenting an apparent molecular mass of Ͼ40 kDa. This experiment indicated that Mg 2ϩ did not significantly affect the molecular volume of the Fl-TNF␣ ARE substrate, making it unlikely that the Mg 2ϩ -induced increase in fluorescein anisotropy was due to increased molecular volume mediated by RNA multimerization.
In the absence of changes in molecular volume, an increase in the anisotropy of the fluorescein moiety of Fl-TNF␣ ARE may thus be reflective of restricted intramolecular segmental motion in the presence of Mg 2ϩ . To test this hypothesis, a potential interaction between Mg 2ϩ and the RNA was considered using a variation of a general binding scheme (43) where A R and A RЈ represent the intrinsic anisotropy values of the free and cation-complexed fluorescein-labeled RNA substrates, respectively. This model was first tested by measuring anisotropy values across a titration of RNA concentrations, in either the absence or the presence of Mg 2ϩ (Fig. 3). Varying the concentration of TNF␣ ARE RNA 5000-fold had no significant effect on the anisotropy (A t ) of the Fl-TNF␣ ARE substrate in reactions containing 0.5 mM EDTA, 1 mM Mg 2ϩ , or 5 mM Mg 2ϩ , indicating that the Mg 2ϩ -induced changes in anisotropy were independent of RNA concentration. By Equation 5, this only occurs when the binding reaction is unimolecular with respect to RNA (i.e. y ϭ 1), further supporting the idea that Mg 2ϩ induces or stabilizes structural conformations of TNF␣ ARE without forming RNA multimers.
Consequences of Mg 2ϩ interaction with an RNA substrate containing the TNF␣ ARE were also assessed by monitoring changes in the mean distance between the 5Ј-and 3Ј-termini of the RNA (5Ј-3Ј distance) using FRET. The efficiency of FRET between a fluorescent donor and acceptor increases as the scalar distance between them decreases (44). In one experiment, FRET efficiency was estimated in the absence of Mg 2ϩ for probe combinations selected so that the FRET donor (Fl) and acceptor (Cy) pairs were present either on different molecules (Cy-TNF␣ ARE ϩ TNF␣ ARE-Fl), on a single molecule (Cy-TNF-Fl), or on a single molecule that was maintained in a rigid form by hybridization to an unlabeled complimentary DNA oligonucleotide (Cy-TNF-Fl:DNA duplex) (Fig. 4A). FRET efficiency was significantly higher for the double-labeled RNA substrate (Cy-TNF-Fl) relative to an equimolar mixture of singly labeled substrates (Cy-TNF␣ ARE ϩ TNF␣ ARE-Fl), indicating that the intramolecular 5Ј-3Ј distance of the probe Cy-TNF-Fl was well within the sensitivity of this assay and that RNAs containing the TNF␣ ARE are unlikely to multimerize in the absence of Mg 2ϩ . However, annealing the Cy-TNF-Fl RNA substrate to a complimentary DNA molecule largely abrogated this increase in FRET efficiency. Because the Cy-TNF-Fl RNA substrate is retained as an elongated double helix in this hybrid, the distance between the 5Ј-and 3Ј-termini of the RNA is maximized, thus minimizing intramolecular FRET. This further indicated that the 5Ј-3Ј distance of the single-stranded Cy-TNF-Fl probe was shorter compared with a fully extended RNA molecule, supporting the proposition that the TNF␣ ARE RNA substrates are inherently flexible in the absence of Mg 2ϩ . Molecules exhibiting long tumbling rotations, such as doublestranded nucleic acids, may be subject to FRET artifacts because of linear polarization of the molecule (44). However, these artifacts are likely to be minimized in this case because each fluorophore is linked to the RNA substrate by multiple single bonds (seven for the 5Ј-Cy3; two for the 3Ј-Fl), thus increasing the segmental rotational freedom of each fluorophore independent of RNA-DNA hybridization. These experiments demonstrated that the FRET assay could be used to detect changes in the distance between the 5Ј-and 3Ј-termini of an RNA substrate containing the TNF␣ ARE and thus would be useful in evaluating changes in RNA conformation.
With the efficacy of this assay system established, changes in the relative FRET efficiency of each combination of RNA substrates were evaluated across a titration of Mg 2ϩ concentrations. FRET efficiency of the bimolecular donor-acceptor pair Cy-TNF␣ ARE ϩ TNF␣ ARE-Fl was independent of Mg 2ϩ concentration (Fig. 4B, solid triangles), further confirming that Mg 2ϩ -induced RNA structural changes involving the TNF␣ ARE do not involve the formation of RNA multimers. By contrast, Mg 2ϩ significantly enhanced the FRET efficiency of the unimolecular Cy-TNF-Fl probe in a dose-dependent manner (Fig. 4B, solid circles), demonstrating that this RNA substrate exhibits a cation-dependent change in structure with the net effect of shortening the distance between the 5Ј-and 3Ј-termini of the RNA. Annealing the Cy-TNF-Fl substrate to a complimentary DNA oligonucleotide abrogated the Mg 2ϩ -induced increase in FRET efficiency (Fig. 4B, open circles), suggesting that these structural changes require base-specific contacts and/or that the RNA substrate be inherently flexible. Taken together, the gel filtration experiments, the RNA titration experiments, and the FRET studies demonstrate that Mg 2ϩ induces or stabilizes structural changes in the TNF␣ ARE and that the structural changes are unimolecular with respect to RNA.
Equilibrium binding of Mg 2ϩ with the TNF␣ ARE-Because Mg 2ϩ -induced folding of the Fl-TNF␣ ARE substrate does not appear to involve the formation of RNA multimers, Equation 5 was simplified by the solution of y ϭ 1 to yield Equation 6.
To assess the stoichiometric contribution of Mg 2ϩ (x) to this binding event, experiments were performed in which the anisotropy of Fl-TNF␣ ARE was monitored across a titration of Mg 2ϩ concentrations (Fig. 5, solid circles). The Mg 2ϩ -induced increase in the anisotropy of this RNA substrate was well resolved by Equation 6 (Fig. 5, solid line). mal RNA folding was then solved as K f ϭ 7.0 Ϯ 0.1 ϫ 10 2 M Ϫ1 (n ϭ 3). The association of Mg 2ϩ with Fl-TNF␣ ARE was also very dynamic, because restoration of anisotropic equilibrium following addition of free Mg 2ϩ or chelation of Mg 2ϩ with excess EDTA was complete within 10 -20 s (data not shown). As expected, the Fl-U 32 substrate displayed only modest changes in fluorescence anisotropy as the concentration of Mg 2ϩ increased (Fig. 5, open circles), further confirming that the Mg 2ϩinduced increase in the anisotropy of the Fl-TNF␣ ARE substrate is dependent on the presence of the TNF␣ ARE sequence.
Characterization of TNF␣ ARE Structural Changes Induced or Stabilized by Mg 2ϩ -Additional experiments were performed to elucidate details of RNA structures involving the TNF␣ ARE that were induced or stabilized by Mg 2ϩ . Anisotropy experiments using the RNA substrate Fl-TNF␣ ARE-2 (Table I) were used to determine whether sequences outside of the AϩU residues within Fl-TNF␣ ARE were required for RNA structural changes in the presence of Mg 2ϩ . Similar changes in the anisotropy of both Fl-TNF␣ ARE-2 and Fl-TNF␣ ARE substrates were observed across titrations of Mg 2ϩ (data not shown), indicating that AϩU residues within the TNF␣ ARE are sufficient for adoption of Mg 2ϩ -induced RNA structures.
Next, a binary regression model analogous to Equation 6 was applied to the changes in FRET efficiency of Cy-TNF-Fl as a function of [Mg 2ϩ ] (Fig. 4B, solid circles), to determine whether Mg 2ϩ -induced changes in 5Ј-3Ј distance, measured by FRET, and RNA flexibility, measured by anisotropy, exhibited similar sensitivity with respect to the cation. Regression of this data set using the binary model (Fig. 4B, solid line) yielded an equilibrium constant K ϭ 6 Ϯ 1 ϫ 10 2 M Ϫ1 (n ϭ 2) that did not differ significantly from K f (7.0 Ϯ 0.1 ϫ 10 2 M Ϫ1 ), the anisotropic equilibrium constant described above. These data suggest that shortening of the 5Ј-3Ј distance of the TNF␣ ARE and restriction of its flexibility near the 5Ј-end may be consequences of the same Mg 2ϩ -induced or -stabilized RNA folding event.
To evaluate whether restraint of RNA mobility was limited to the 5Ј-end of the TNF␣ ARE, anisotropy of the 3Ј-fluorescein labeled RNA substrate TNF␣ ARE-Fl (Table I) was also monitored across a titration of Mg 2ϩ . Although this RNA substrate presented a value for K f comparable with Fl-TNF␣ ARE, (7 Ϯ 2 ϫ 10 2 M Ϫ1 , n ϭ 2), a significant decrease in the overall change in fluorescence anisotropy, given by ⌬A ϭ A RЈ Ϫ A R , was observed (cf. 0.009 Ϯ 0.002, n ϭ 2 for TNF␣ ARE-Fl versus 0.021 Ϯ 0.003, n ϭ 3 for Fl-TNF␣ ARE). The decreased value of ⌬A for TNF␣ ARE-Fl compared with Fl-TNF␣ ARE suggests that, although both ends of the RNA are affected by structural changes following association of Mg 2ϩ , the flexibility of the RNA appears more strongly restricted close to the 5Ј-end.
Besides Mg 2ϩ , other divalent cations are also known to stabilize higher order RNA structures (43,45,46). In anisotropy assays with Fl-TNF␣ ARE, both Ca 2ϩ (Fig. 6A) and Mn 2ϩ (Fig.  6B) increased the fluorescence anisotropy of the RNA substrate in a dose-dependent manner. Regression solutions using Equation 6 with x ϭ 1 indicated that the affinity of Fl-TNF␣ ARE for these cations was somewhat lower than for Mg 2ϩ , however, giving equilibrium constants (K f ) of 3.1 Ϯ 0.4 ϫ 10 2 M Ϫ1 and 4.0 Ϯ 0.4 ϫ 10 2 M Ϫ1 for Ca 2ϩ and Mn 2ϩ , respectively. These data demonstrate that other divalent cations may also influence the higher order structure of the TNF␣ ARE.
In summary, the similarity in Mg 2ϩ -Fl-TNF␣ ARE binding equilibria described by FRET and anisotropy and the effects of Mg 2ϩ on the anisotropy of the Fl-TNF␣ ARE-2 and TNF␣ ARE-Fl RNA substrates reveal that association of the TNF␣ ARE with Mg 2ϩ induces or stabilizes one or more spatially condensed, AϩU-dependent RNA structures. Furthermore, the flexibility of the RNA appears to be restrained in this structure relative to the Mg 2ϩ -free RNA, particularly toward the 5Ј-end. Promotion of these RNA structures may not be specific for Mg 2ϩ , however, because Ca 2ϩ and Mn 2ϩ may also inhibit flexibility of the TNF␣ ARE near the 5Ј-end. Taken together with the correlation between cation-induced RNA structural changes and inhibition of AUF1 binding activity, these studies indicate that alteration of RNA conformation and/or flexibility may play a direct role in regulating the association of AUF1 with the TNF␣ ARE.
Mutually Exclusive Binding of AUF1 or Mg 2ϩ to the TNF␣ ARE Does Not Account for Inhibition of Protein Binding-The next series of experiments was designed to investigate the mechanisms whereby Mg 2ϩ -induced conformational changes in RNA substrates containing the TNF␣ ARE may inhibit the association of AUF1. The simplest model of this inhibitory mechanism is based on competition for free RNA between the reactions RNA ϩ Mg 2ϩ º RNA⅐Mg 2ϩ , described by K f , and the reaction series RNA ϩ P 2 º P 2 R and P 2 R ϩ P 2 º P 4 R, described by K 1 and K 2 , respectively (Fig. 7A). In essence, this represents an extension of the binding scheme described in Fig.  1E, in which the free RNA pool is in equilibrium between Mg 2ϩ -free (R) and Mg 2ϩ -bound (RЈ) states and where RNA conformational changes induced by association with Mg 2ϩ preclude binding of AUF1 dimers. Substitution of equations describing the equilibrium binding constants K f , K 1 , and K 2 in terms of the concentrations of free RNA, Mg 2ϩ , and AUF1 dimer into Equation 1 yields Equation 7.
To test this model for inhibition of AUF1 binding to the Fl-TNF␣ ARE substrate by Mg 2ϩ , nonlinear regression of an anisotropy data set generated by titration of His 6 -p37 AUF1 in the presence of 0.2 nM Fl-TNF␣ ARE and 5 mM Mg 2ϩ was applied to Equation 7 (Fig. 7B). Values of A R , A P2R , A P4R , K 1 , and K 2 were known from AUF1 titration experiments performed with [Mg 2ϩ ] ϭ 0 ( Fig. 1C and Table II). Likewise, the value of K f was established from Mg 2ϩ titration experiments performed in the absence of AUF1 ([P 2 ] ϭ 0; Fig. 5). Whereas a value for the intrinsic anisotropy of the Mg 2ϩ ⅐RNA complex (A RЈ ) was also provided by solution of Mg 2ϩ titrations in the absence of protein using Equation 6 ( Fig. 5), this constant was not fixed in subsequent regressions involving protein titrations because of the use of heparin as a competitor for nonspecific RNA-binding activity in these experiments. Although the inclusion of heparin in Mg 2ϩ titration experiments up to 5 mM Mg 2ϩ did not significantly affect the equilibrium constant describing the Mg 2ϩ -RNA interaction (K f ) (data not shown), small increases observed in absolute anisotropy values could potentially influence the solution of more global regression functions. As shown in Fig. 7B, however, the regression function described by Equation 7 (solid line) did not accurately reflect the observed anisotropy data. This function did generate a reasonable estimate for the intrinsic anisotropy of the Mg 2ϩ ⅐RNA complex (A RЈ ϭ 0.048 Ϯ 0.002), but the model overestimated anisotropy values for the global equilibrium of His 6 -p37 AUF1 association with Fl-TNF␣ ARE in the presence of 5 mM Mg 2ϩ , particularly for concentrations of protein dimer above 2 nM. Although no independent experimental evidence is available to dismiss this binding model, extensive mathematical simulations using these data sets reveal no situation in which the association of AUF1 with the Fl-TNF␣ ARE RNA substrate in the presence of 5 mM Mg 2ϩ can be satisfied without significant changes in the binding parameters defined by earlier experiments (Table II and Fig. 5). The negative pressure on measured anisotropy values as a function of protein concentration suggested the existence of another fluorescent binding complex (i.e. containing RNA) that exhibited low intrinsic anisotropy and/or was restrictive of RNA-dependent protein oligomerization. Accordingly, an alternative model for inhibition of AUF1 binding to the TNF␣ ARE by Mg 2ϩ was explored.
A Convergent Model of AUF1-TNF␣ ARE-Mg 2ϩ Binding Equilibrium-To accommodate the discrepancy between the "mutually exclusive binding model" described above and the measured anisotropy of the Fl-TNF␣ ARE substrate across a titration of AUF1 in the presence of Mg 2ϩ , it was postulated that an additional fluorescent species might exist, consisting of a ternary complex of Fl-TNF␣ ARE, Mg 2ϩ , and an AUF1 dimer. This complex could exert a negative influence on total measured anisotropy by exhibiting a low intrinsic anisotropy, forming with poor affinity for AUF1 dimers, forcing Mg 2ϩ to be ejected prior to formation of AUF1 tetramers, or any combination of these mechanisms. The assembly of this complex, denoted P 2 RЈ, would proceed either by association of an AUF1 dimer with a Mg 2ϩ -bound RNA, described by K 1 Ј, or by binding of Mg 2ϩ to the AUF1-dimer bound P 2 R complex, described by K f2 (Fig. 8A). Because this binding pathway converges at the Mg 2ϩ -free P 2 R species as a necessary prerequisite for AUF1 tetramer formation as [ This binding model was tested by fitting anisotropy values of Fl-TNF␣ ARE from a His 6 -p37 AUF1 titration experiment in the presence of 5 mM Mg 2ϩ by nonlinear regression. Similar to the test of the mutually exclusive binding model considered in Fig.  7, values of A R , A P2R , A P4R , K 1 , K 2 , and K f were fixed in Equation 8 based on prior experimentation (Table II and Table III. In addition, the rapid dynamics of all RNA-Mg 2ϩ and RNA-AUF1 interactions tested (Ref. 25 and data not shown) indicates that the global equilibrium is quickly re-established following changes in RNA structure or AUF1 concentrations. Although a binding species containing both an AUF1 tetramer and Mg 2ϩ may exist in this system, the precision of the regression solution in the absence of terms describing an additional complex indicate that its concentration is likely too low to significantly contribute to total measured anisotropy. The implications of the convergent binding model for regulating the formation of AUF1 multimers by changes in RNA structure and their potential effects on mRNA turnover rates are described under "Discussion." DISCUSSION Application of the convergent binding model to the global AUF1-TNF␣ ARE-Mg 2ϩ equilibrium satisfies the experimental data in many respects. The model includes the potential for independent association of Mg 2ϩ or protein with the RNA substrate. In the absence of Mg 2ϩ , AUF1 tetramer assembly proceeds by sequential binding of protein dimers described by K 1 and K 2 . In the absence of protein, Mg 2ϩ induces or stabilizes a conformational change in the RNA described by K f . However, in the presence of both Mg 2ϩ and AUF1, inhibition of protein association with the RNA may be achieved by two primary mechanisms. First, the binding affinity of a Mg 2ϩ -bound RNA (K 1 Ј) versus a Mg 2ϩ -free RNA (K 1 ) substrate for an AUF1 dimer is decreased approximately 5-fold (Table III). The resulting diminution of the change in free energy upon AUF1 binding when Mg 2ϩ is already bound to the RNA indicates that the folded RNA substrate presents reduced potential for intermolecular contacts with the protein. An appealing corollary of this premise is that the reiterative nature of AREs and the potential for multiple RNA-protein contact sites involving a single AUF1 dimer (four RNA recognition motifs/dimer) allow the possibility that RNA recognition motifs from both AUF1 subunits may contribute to optimal binding affinity. Second, the convergent binding model predicts that Mg 2ϩ must be ejected from the P 2 RЈ complex prior to tetramerization of AUF1 on the RNA. Because this event is accompanied by an increase in free energy, retention of RNA substrate in the P 2 RЈ state requires higher concentrations of protein to force the equilibrium in favor of AUF1 tetramers.
Although no physical evidence has been presented to unequivocally demonstrate the existence of the ternary P 2 RЈ complex, elucidated values of anisotropy and equilibrium binding constants associated with its formation hint at some of its properties. For example, although multiple regions of the TNF␣ ARE may be involved in its association with Mg 2ϩ , the resulting RNA conformation is likely more constrained toward its 5Ј-end in both the RЈ and P 2 RЈ states. In the formation of the RЈ state, this is evidenced by the decrease in ⌬A observed for the 3Ј-labeled TNF␣ ARE-Fl substrate compared with the 5Јlabeled Fl-TNF␣ ARE substrate upon Mg 2ϩ binding. Because the Mg 2ϩ -induced increase in anisotropy of these RNAs is largely, if not solely, due to inhibition of segmental motion accompanying changes in RNA conformation, restriction of bond rotations close to the fluorophore would exert a greater influence on its anisotropy than constraints on the mobility of more distal bonds. In the formation of the P 2 RЈ state from the RЈ precursor, the intrinsic anisotropy does not significantly change following protein binding (Table III), indicating that addition of an AUF1 dimer has little effect on probe anisotropy when Mg 2ϩ is already bound. This may reflect distal placement of the AUF1 dimer relative to Mg 2ϩ on the RNA or that the limitations on protein-RNA contacts imposed by the Mg 2ϩinduced RNA structure (described above) may prevent significant changes in RNA segmental motion following protein binding. The latter of these hypotheses is further supported by the observation that the intrinsic anisotropy of the Fl-TNF␣ ARE is dramatically increased in the high affinity P 2 R state (A P2R ) relative to the lower affinity P 2 RЈ complex (A P2RЈ ). However, this model also implies that changes in molecular volume following protein binding play a relatively small role in regulating the mobility of the fluorescein moiety in this system.
To our knowledge, this work presents the first evidence that AREs may adopt higher order RNA structures in the presence of Mg 2ϩ . The generation or stabilization of RNA tertiary structures by Mg 2ϩ has been well documented for many highly ordered RNA systems, including ribozymes (39,43,47), RNA pseudoknots (48,49), RNA helical junctions (50), internal ribosome entry sites (51), and transfer RNAs (52,53). AREs, however, generally appear to lack any significant potential for Watson-Crick base pairing, so contributions of classical duplexed structures are likely to be minimal. However, the observation that the Mg 2ϩ -induced increase in anisotropy of Fl-TNF␣ ARE is significantly greater than that observed with a a Values are listed as x Ϯ nϪ1 of triplicate experiments. The constants A R , A P2R , A P4R , K 1 , and K 2 were determined by AUF1 titration experiments performed in the absence of Mg 2ϩ as defined in Fig. 1 and Table II. The constant K f was determined by Mg 2ϩ titration experiments performed in the absence of protein as shown in Fig. 5.
b Values are listed as x Ϯ 95% confidence intervals for regression of the data set shown in Fig. 8B using Equations 8 and 9. A duplicate experiment yielded values that did not significantly vary from those listed above. similarly sized polyuridylate substrate (Fl-U 32 ) suggests that the interspersed adenosine residues that distinguish the TNF␣ ARE from poly(U) are likely involved. Future studies will focus on the elucidation of binding determinants regulating the association of Mg 2ϩ with an ARE.
The implications of Mg 2ϩ -dependent RNA folding on ARE function may be related to the differences in sequence composition of AREs from different mRNAs (7). For example, AREs from mRNAs encoding cytokines/lymphokines and inflammatory mediators typically contain several overlapping repeats of AUUUA, similar to the sequence of Fl-TNF␣ ARE. By contrast, these pentameric repeats are generally dispersed or even absent altogether in AREs from proto-oncogene mRNAs, which are largely divergent except for a high proportion of uridylate residues. Because Mg 2ϩ -induced changes in ARE conformation and subsequent association with AUF1 are dependent on RNA sequence (this work), the activity of a given ARE to promote rapid constitutive mRNA decay may be regulated by its potential for adopting higher order RNA structures. For example, an mRNA containing an ARE presenting strong structural constraints is likely to be less accessible for AUF1 recognition and hence will be more stable, because AUF1 binding activity closely correlates with the ability of an ARE to destabilize mRNA (16, 19 -22, 24). In the case of the Fl-TNF␣ ARE substrate, the RNA-Mg 2ϩ binding equilibrium constant (K f ) of 7.0 ϫ 10 Ϫ2 M Ϫ1 equates to a dissociation constant (K d ) of 1.4 mM. This is near the physiological range of intracellular Mg 2ϩ concentration (54), suggesting that structural variants of this ARE may play a significant role in regulating the accessibility of trans-acting factors in vivo. Similar mechanisms are also likely to regulate protein-binding events involving other RNA elements. A recent study demonstrated that Mg 2ϩ inhibits the association of insulin-like growth factor II mRNA-binding protein to the H19 RNA (55). Also, Mg 2ϩ -induced tertiary structures are involved in the association of ribosomal protein L11 with 23 S ribosomal RNA (56). In addition to potential roles in regulating constitutive mRNA decay rates, however, mechanisms involving ARE structures may also serve to modulate mRNA decay in inducible systems. For example, association of some "modulator" protein proximal to the ARE may act as an enhancer or repressor of AUF1 binding activity by remodeling the local topology of the RNA. In this manner, flanking RNA sequences may also serve as cis-modifiers of mRNA turnover, by imparting some thermodynamic or kinetic influence to an RNA structure involving the ARE or by acting as target sites for the association of other RNA-binding proteins.