Assembly of AUF1 Oligomers on U-rich RNA Targets by Sequential Dimer Association*

Many labile mammalian mRNAs are targeted for rapid cytoplasmic turnover by the presence of A + U-rich elements (AREs) within their 3′-untranslated regions. These elements are selectively recognized by AUF1, a component of a multisubunit complex that may participate in the initiation of mRNA decay. In this study, we have investigated the recognition of AREs by AUF1 in vitro using oligoribonucleotide substrates. Gel mobility shift assays demonstrated that U-rich RNA targets were specifically bound by AUF1, generating two distinct RNA-protein complexes in a concentration-dependent manner. Chemical cross-linking revealed the interaction of AUF1 dimers to form tetrameric structures involving protein-protein interactions in the presence of high affinity RNA targets. From these data, a model of AUF1 association with AREs involving sequential dimer binding was developed. Using fluorescent RNA substrates, binding parameters of AUF1 dimer-ARE and tetramer-ARE equilibria were evaluated in solution by fluorescence anisotropy measurements. Using two AUF1 deletion mutants, sequences C-terminal to the RNA recognition motifs are shown to contribute to the formation of the AUF1 tetramer·ARE complex but are not obligate for RNA binding activity. Kinetic studies demonstrated rapid turnover of AUF1·ARE complexes in solution, suggesting that these interactions are very dynamic in character. Taken together, these data support a model where ARE-dependent oligomerization of AUF1 may function to nucleate the formation of a trans-acting, RNA-destabilizing complex in vivo.

The control of cytoplasmic mRNA turnover plays a major role in regulating both the level and timing of expression of many gene products in eukaryotes (reviewed in Refs. 1 and 2). In many cases, sequence elements within individual mRNAs function as cis-acting determinants of their stability, either constitutively or in response to external stimuli. Conceptually, modulation of mRNA turnover rates may be envisioned as altering the activity or accessibility of one or more ribonucleases toward a specific transcript. At present, however, few mechanistic details are available linking cis-acting RNA sequence elements with the decay machinery necessary for hydrolysis of the target mRNA.
AREs 1 are potent cis-acting determinants of rapid cytoplasmic mRNA turnover in mammalian cells. They generally consist of one or more overlapping AUUUA pentamers contained within or near a U-rich tract (3)(4)(5). These elements are present in the 3Ј-untranslated regions (3Ј-UTRs) of many labile mRNAs, including several encoding inflammatory mediators, cytokines, oncoproteins, and G protein-coupled receptors (3, 6 -10). mRNA turnover mediated by AREs is usually characterized by rapid 3Ј to 5Ј shortening of the poly(A) tract followed by decay of the mRNA body (11)(12)(13)(14)(15). In addition, ARE-directed mRNA decay is dependent upon active translation of the mRNA in many cellular systems (16 -20), suggesting that some link also exists between protein synthesis and mRNA decay mechanisms.
AUF1 is an RNA-binding protein that exhibits many characteristics of a trans-acting factor participating in ARE-directed mRNA turnover (reviewed in Ref. 21). Current evidence indicates that AUF1 may function as a targeting system for AREs, either recruiting or promoting the assembly of multisubunit trans-acting complexes at these sites (22)(23)(24). Several cytoplasmic proteins co-immunoprecipitate with AUF1, indicating that AUF1 associates with additional factors in vivo (25). Some of these have been identified immunologically as the translation initiation factor eIF4G, poly(A)-binding protein, heat shock protein 70, and the 70-kDa heat shock cognate protein (26). The identification of these associated proteins is evidence of a physical link between AUF1 and factors involved in translation and mRNA turnover.
Elucidation of the mechanisms contributing to rapid mRNA turnover by AREs will require further understanding of both the molecular architecture of the trans-acting complex(es) as well as the molecular events involved in recognition of AREs by these factors. In particular, the direct interaction of AUF1 with target RNA sequences may serve to nucleate factor binding or transduce some signal to activate pre-assembled complexes. In this study, we have investigated the interaction of AUF1 in vitro with the following two U-rich oligoribonucleotides: the core ARE from tumor necrosis factor ␣ (TNF␣) mRNA, and a uridylate homopolymer. We present evidence that recognition of U-rich RNA sequences by AUF1 initiates the assembly of AUF1 multimers involving both RNA-protein and protein-protein interactions by sequential binding of AUF1 dimers. The application of fluorescence anisotropy to the study of RNA: AUF1 solution equilibria allowed estimation of the equilibrium constants for both initial and secondary binding events as well as an assessment of complex dynamics by off-rate analyses. Finally, we discuss potential functional consequences of RNA-dependent AUF1 oligomerization in the control of cytoplasmic mRNA turnover.

EXPERIMENTAL PROCEDURES
RNA Substrates-All RNA oligonucleotides (2Ј-hydroxyl) were synthesized by Dharmacon Research (Boulder, CO). The sequence of each RNA probe is listed in Fig. 1A. Following 2Ј-O-deprotection according to the manufacturer's instructions (27), RNA oligonucleotides were quantified by absorbance at 260 nm. Estimates of the extinction coefficients for each RNA probe at 260 nm were calculated as described (28). For fluorescein-tagged probes, absorbance at 260 nm was corrected by quantitation of the fluorescein moiety at 495 nm as described (29).
The substrate "TNF␣ ARE" corresponds to the core ARE from the 3Ј-UTR of human TNF␣ mRNA. The RNA substrate "U 32 " contains a uridylate homopolymeric sequence, and "R␤" encodes a fragment of the rabbit ␤-globin coding region. Duplicate oligoribonucleotides containing 5Ј-fluorescein labels were also synthesized and are designated Fl-TNF␣ ARE, Fl-U 32 , and Fl-R␤, respectively. For gel mobility shift assays, TNF␣ ARE, U 32 , and R␤ substrates were radiolabeled using T4 polynucleotide kinase (Promega, Madison, WI) and [␥-32 P]ATP (4500 Ci/mmol) (ICN Biomedicals, Costa Mesa, CA) to specific activities of 3-5 ϫ 10 3 cpm/fmol. Unincorporated radiolabel was removed by spin column chromatography using G-25 Quick Spin columns (Roche Molecular Biochemicals). Probe-specific activity was determined by liquid scintillation counting, and RNA integrity was verified by denaturing polyacrylamide gel electrophoresis and autoradiography. 32 P-Labeled RNA probes were detected as single bands (data not shown) indicating that they were predominantly (Ͼ99%) full length.
Preparation of Recombinant Proteins-The construction of plasmids pTrcHisB-AUF1-(1-257) and pTrcHisB-AUF1-(1-229) was described previously (24). pTrcHisB-AUF1-(1-257) encodes a stable, N-terminal His 6 -tagged mutant of human p37 AUF1 lacking 30 amino acid residues from the C terminus but showing comparable ARE binding activity to full-length p37 AUF1 (24). pTrcHisB-AUF1-(1-229) encodes a truncation mutant of p37 AUF1 lacking all sequences C-terminal of the RNA recognition motifs (RRMs). Recombinant His 6 -AUF1 mutant proteins were expressed and purified as described (30) and were judged to be Ͼ95% pure by SDS-PAGE. For protein cross-linking studies, His 6 proteins were dialyzed against 10 mM HEPES⅐KOH (pH 7.5) prior to concentration. Where indicated, a 3500-Da N-terminal fragment containing the His 6 tag was removed from His 6 -p37 AUF1 -(1-257) using the recombinant enterokinase kit (Novagen, Madison, WI) according to the manufacturer's instructions. A mock-digested reaction was also assembled to control for changes in protein activity resulting from prolonged incubation at room temperature (2 h). All recombinant proteins were quantified by the method of Bradford (31) using bovine serum albumin as standard. Protein concentrations were also evaluated by comparison of Coomassie Blue-stained SDS-PAGE gels containing recombinant proteins and a titration of bovine serum albumin. Determination of protein concentrations by both methods yielded estimates within 10%.
Protein-Protein Cross-linking-Dithio-bis(succinimidyl propionate) (DSP)-mediated protein cross-linking was performed in 10-l reactions containing 10 mM HEPES⅐KOH (pH 7.5), 100 mM potassium acetate, and 5 mM magnesium acetate. In this buffer system, HEPES-dialyzed His 6 -p37 AUF1 -(1-257) or His 6 -p37 AUF1 -(1-229) was diluted to 5 M in the presence or absence of 1.5 M RNA. DSP (Pierce) was then added to a final concentration of 2.5 mM, and reactions were allowed to proceed for 10 min at room temperature. Cross-linking was then quenched by addition of Tris⅐HCl (pH 7.5) to 1 M final concentration and incubation for a further 15 min. Reaction products were fractionated by SDS-PAGE in the absence of reducing agents. Complexes containing AUF1 were identified by probing immunoblots with anti-AUF1 antiserum (25). Secondary antibody detection was performed using the SuperSignal Chemiluminescent Detection Kit (Pierce) and exposure to x-ray film.
Fluorescence Anisotropy-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 assembled as described for gel mobility shift assays (above) except that no glycerol was added, and the final volume was 100 l. For equilibrium binding experiments, the polarimeter was operated in static mode, with each sample read as blank prior to addition of fluorescein-labeled RNA probes. Following probe addition, samples were incubated for 1 min before anisotropy was measured. Preliminary onrate analyses demonstrated that anisotropic equilibrium was reached within 10 s (data not shown). Data points represent the mean of 10 measurements for each binding reaction. Samples for off-rate analyses were similarly assembled, except that following an initial reading (t ϭ 0), a 5000-fold molar excess of unlabeled RNA competitor was added to the binding mixture and rapidly mixed. Anisotropy measurements were taken in kinetic mode in intervals of 15 s, with five measurements taken at each time point. All non-linear regression fitting of anisotropic data and statistical evaluations were performed using PRISM software version 2.0 (GraphPad, San Diego, CA).

AUF1
Binding to U-rich Oligoribonucleotide Substrates Generates Two Complexes in Vitro-Human TNF␣ mRNA contains an ARE in its 3Ј-UTR which contributes to its rapid turnover in vivo (32,33). It is also sufficient to destabilize a heterologous mRNA in transfected cell systems (12), and cytoplasmic proteins have been identified binding this element in vitro (34). The core sequence of the TNF␣ ARE is similar to others identified as high affinity AUF1-binding sites (23,25). Taken together, these features make the TNF␣ ARE a strong candidate for high affinity interaction with AUF1.
In gel mobility shift assays, binding of His 6 -p37 AUF1 -(1-257) to the TNF␣ ARE and U 32 sequences generated two distinct complexes ( Fig. 1, B and C). No detectable binding was observed to the rabbit ␤-globin substrate (Fig. 1D), consistent with the selectivity of AUF1 for A ϩ U-rich RNA sequences (23,25). For both the TNF␣ ARE and U 32 probes, the distribution of complexes with RNA was dependent on protein concentration, with the faster migrating complexes (complex I) appearing at lower concentrations of His 6 -p37 AUF1 -(1-257) (0.5-5 nM). Levels of complex I diminished as the abundance of complex II increased, consistent with the possibility of a precursor-product relationship. The diffuse smearing observed below binding complexes is likely due to RNA-protein dissociation in the gel. No additional binding events were observed in these assays, with protein concentrations tested up to 500 nM (data not shown).
AUF1 Forms Tetramers in the Presence of a High Affinity RNA-binding Site-Previous hydrodynamic studies demonstrated that AUF1 forms dimeric structures in solution involving an N-terminal alanine-rich region (24). Monomeric AUF1 was not detected in these experiments, indicating that dimers are generated with high affinity. To determine whether AUF1 oligomerization might contribute to the formation of complexes with the TNF␣ ARE, RNA-protein binding reactions were treated with the chemical cross-linker DSP, permitting covalent linkage through primary amino groups. In the absence of cross-linker, His 6 -p37 AUF1 -(1-257) migrates at an apparent M r Ϸ 43,000 by SDS-PAGE, larger than its predicted M r of 32,600 (data not shown). In cross-linking reactions lacking RNA, His 6 -p37 AUF1 -(1-257) was primarily detected as a dimer (Fig. 2, lane 1), consistent with the hydrodynamic studies of the full-length p37 AUF1 (24). Cross-links were generated specifically through the DSP linker, since they were cleaved following treatment with reducing agents (data not shown). The presence of a non-binding RNA substrate (R␤) did not alter the distribution of cross-linked protein products (Fig. 2, lane 2). However, larger protein complexes up to and including tetramers were observed in the presence of high affinity RNA targets (U 32 and TNF␣ ARE; Fig. 2, lanes 3 and 4). The generation of these larger AUF1 complexes in the presence of U 32 indicated that these species were unlikely to be the result of RNA bridging, since there is a paucity of primary amino groups contained within U 32 RNA. Furthermore, the lack of detectable AUF1 tetramers in the absence of a high affinity RNA target even at high concentrations of protein (5 M) suggests that dimerdimer association does not occur in the absence of U-rich RNA sequences.
Two binding complexes of AUF1 and RNA were detected by gel mobility shift assay (Fig. 1, B and C). The slowest mobility complex (complex II) likely represents RNA associated with the AUF1 tetramer, since it represents the largest change in mobility relative to the free probe, and its abundance increases with protein concentration. Complex I is observed at lower concentrations of protein and could thus represent an AUF1 monomer, dimer, or trimer associated with RNA. However, because unbound AUF1 is dimeric (24), interpretation of complex I as an AUF1 dimer rather than a monomer or trimer bound to RNA is the sole case in which monomeric AUF1 species are not required. Given both the absence of additional intermediate binding species observed by gel mobility shift analysis ( Fig. 1) and the absence of detectable monomeric AUF1 by gel filtration and sedimentation velocity experiments (24), the contribution of monomeric AUF1 to the binding equilibrium is likely to be negligible.
A Model for ARE-dependent Assembly of AUF1 Multimers Based on Sequential Association of Protein Dimers-Given the following findings, (i) His 6 -p37 AUF1 -(1-257) association with TNF␣ ARE and U 32 RNAs generates two RNA-protein complexes, (ii) formation of these complexes is dependent on protein concentration, (iii) AUF1 complexes as large as tetramers are associated with these RNA targets, and (iv) AUF1 dimers do not interact in the absence of U-rich RNA targets, we propose that AUF1 associates with these RNA probes by sequential dimer binding. In this model, complex I ( Fig. 1) represents the AUF1 dimer-bound RNA (P 2 R), and complex II represents an RNA-associated AUF1 tetramer (P 4 R). Whereas the re-iterative nature of AREs suggests that AUF1 oligomers may be the result of multiple binding sites on the RNA target, RNA-dependent tetramer cross-linking indicates that adjacent dimers are held in close proximity in the P 4 R complex, making interaction between these subunits likely. Furthermore, AUF1⅐ARE binding equilibria evaluated by remaining free probe concentration ( Fig. 1, see also Refs. 23,24,35) are not resolved by Scatchard analysis, suggesting that binding events involving multiple AUF1 dimers are not independent (data not shown). However, the model does not exclude the possibility that the initial dimer binding event may occur at one of several sites on a given RNA target. For this reason, the AUF1 dimer-RNA equilibrium may reflect an average of multiple simultaneous P 2 R variants.
An AUF1 Mutant Protein Lacking Sequences C-terminal of the RRMs Is Defective in RNA-dependent Tetramer Formation-The generation of tetrameric His 6 -p37 AUF1 -(1-257) structures on U-rich RNA targets indicates that protein-protein interactions may be generated between AUF1 dimers in an RNA-dependent manner. Whereas previous studies demonstrated that formation of AUF1 dimers in the absence of RNA required sequences near the N terminus, sequences C-terminal to the RRMs were dispensable for dimer assembly (24). In order to evaluate the contribution of C-terminal sequences to tetramer formation in the presence of an RNA target, gel mobility FIG. 1. Gel mobility shift assays of His 6 -p37 AUF1 -(1-257) with 32 P-labeled oligoribonucleotides. A, sequences of synthetic RNAs. Indicated 5Ј-end-labeled RNAs (0.15 nM) were used to program binding reactions containing varying concentrations of His 6 -p37 AUF1 -(1-257). Reaction products were then fractionated by non-denaturing polyacrylamide gel electrophoresis as described under "Experimental Procedures" (B-D). Protein concentrations listed represent monomeric His 6 -p37 AUF1 -(1-257). Distinct RNA-protein complexes generated with the TNF␣ ARE and U 32 RNA substrates are indicated by brackets, and the positions of free RNAs are identified. shift assays were also performed using the truncation mutant His 6 -p37 AUF1 -(1-229), which lacks all sequences C-terminal of the RRMs. Association of His 6 -p37 AUF1 -(1-229) with the TNF␣ ARE generated primarily a single complex consistent with P 2 R (Fig. 3A, complex I), whereas a complex consistent with P 4 R (complex II) was detected only weakly at higher concentrations of protein. Similar binding products were observed with the U 32 probe, and no binding activity was observed to R␤, indicating that RNA-binding specificity was not compromised by deletion of the C-terminal sequences of AUF1 (data not shown). Although association with TNF␣ ARE was observed at very low concentrations of His 6 -p37 AUF1 -(1-229) (0.5 nM), complete probe association was not detected (Fig. 3A), even at protein concentrations up to 2 M (data not shown). This suggests that maintenance of His 6 -p37 AUF1 -(1-229)-RNA complexes may be hindered by gel fractionation, possibly involving rapid dissociation in the sample wells or during electrophoresis.
Covalent cross-linking of His 6 -p37 AUF1 -(1-229) with DSP confirmed that this protein forms dimers in the absence of RNA (Fig. 3B, lane 1). A small amount of trimeric AUF1 was also detected but is likely the product of partial oxidation or aggregation in the protein preparation. Inclusion of the R␤ RNA oligonucleotide in the cross-linking reaction did not alter the distribution of covalently linked products (Fig. 3B, lane 2). Similarly, only minimal changes in the recovery of trimeric AUF1 were observed by addition of the TNF␣ ARE (Fig. 3B,  lane 4), and tetrameric species were not detected. However, binding reactions containing U 32 displayed both trimeric and tetrameric cross-linked species (Fig. 3B, lane 3). Taken together, these data indicate that the ability of AUF1 to form RNA-dependent tetramers is compromised but not completely abrogated by removal of sequences between amino acid residues 229 and 257. In particular, tetramer formation with His 6 -p37 AUF1 -(1-229) was observed only with the uridylate homopolymer, where the possibility of multiple identical binding sites exists on the RNA. This suggests that a bona fide ARE, like that in TNF␣ mRNA, presents a hierarchy of AUF1-binding sites consistent with a sequential binding model, in which AUF1 sequences between 229 and 257 contribute to secondary binding events.
Evaluation of AUF1⅐ARE Equilibrium Constants by Fluorescence Anisotropy-In order to evaluate the validity of the sequential dimer-binding model for AUF1 oligomerization on an ARE, it was necessary to first express the equilibrium relationships between each component mathematically. Subsequently, fluorescence-based solution binding experiments performed under equilibrium conditions were employed to test the accuracy of these equations in describing the sequential association of AUF1 dimers with a high affinity RNA target.
The steady state concentrations of P 2 R and P 4 R may be described in terms of the concentrations of RNA By using fluorescein-labeled RNAs, RNA-protein complexes are distinguishable in solution based on differences in the intrinsic fluorescence anisotropy exhibited by each species resulting from changes in molecular volume under conditions of constant temperature and viscosity (36,37). An initial experiment was performed to determine whether the quantum yield of the fluorescein-labeled RNA changed as a result of AUF1 binding. By using Fl-TNF␣ ARE and a titration of His 6 -p37 AUF1 -(1-229), no significant change in fluorescence intensity was observed with protein concentrations up to 250 nM His 6 -p37 AUF1 -(1-229) dimer (Fig. 4A), demonstrating that AUF1 binding did not alter the quantum yield of the fluorescein-labeled RNA. Accordingly, the measured anisotropy A t of the fluorescent RNA probe may be interpreted using Equation 3.
where A i represents the intrinsic anisotropy of each fluorescing species (in this case R, P 2 R, or P 4 R), and f i its fractional concentration (38 -40). Applying this to our binding model where the concentration of total fluorescent riboprobe [R] tot is limiting, the fractional concentration of each species is given by Equations 4 -6.
Changes in the fluorescence anisotropy of Fl-TNF␣ ARE resulting from His 6 -p37 AUF1 -(1-229) binding were used to test the sequential dimer association model (Fig. 4B, solid line) by non-linear least squares regression of A t versus a titration of [P 2 ] using Equation 8. The tetramer-defective AUF1 mutant was selected for this analysis because sequential binding events would be best resolved when K 1 and K 2 are significantly different. A R was determined by measurement of fluorescence anisotropy in the absence of added protein (A R ϭ 0.049) and was fixed in the regression function, whereas all other constants (A P2R , A P4R , K 1 , and K 2 ) were left unfixed. The utility of this binding algorithm is supported by a strong coefficient of determination (R 2 ϭ 0.9839) and by random positioning of residuals about the regression line (Fig. 4C). Since His 6 -p37 AUF1 -(1-229) appears defective in tetramer assembly by gel mobility shift assay, this data set was also fitted to a saturation function describing a single binding site shown in Equation 9.
Comparison of the sum-of-squares deviations for each regression function using the F test (GraphPad PRISM version 2.0) indicates that interpretation of these data is significantly improved using the sequential dimer binding model (p Ͻ 0.0001).
Calculation of AUF1 Equilibrium Binding Constants for Fluorescent RNA Substrates-In general, calculation of K 1 and K 2 values within reasonable confidence intervals from fluorescence anisotropy isotherms using Equation 8 requires knowledge of the intrinsic anisotropy constants A R and A P2R . A R can be measured directly as described above (0.049 for Fl-TNF␣ ARE; 0.039 for Fl-U 32 ). However, an approximation of A P2R is difficult to obtain unless K 1 and K 2 are significantly different. Accordingly, the most confident estimate of A P2R was given by association of His 6 -p37 AUF1 -(1-229) with the Fl-TNF␣ ARE RNA (Fig. 4B). In this case, regression of Equation 8 with A R held constant (0.049), and A P2R , A P4R , K 1 , and K 2 left unfixed yielded a solution for A P2R of 0.080 Ϯ 0.005. Triplicate assays yielded values of A P2R consistent with this estimate. Subsequently, solution of anisotropy plots using Equation 8 with A R and A P2R fixed in the regression allowed estimates of K 1 and K 2 to be calculated for the interactions of His 6 -p37 AUF1 -(1-257) and His 6 -p37 AUF1 -(1-229) with Fl-TNF␣ ARE and Fl-U 32 . A representative plot utilizing this technique for the solution of binding constants describing the His 6 -p37 AUF1 -(1-257): Fl-TNF␣ ARE equilibrium is shown in Fig. 5 (solid circles). Substitution of the HEPES-based buffer system used in crosslinking assays (Figs. 2 and 3B) did not significantly alter the anisotropy profile, but heparin was required to minimize nonspecific interaction of protein with RNA (data not shown). In an additional experiment, the N-terminal His 6 tag of His 6 -p37 AUF1 -(1-257) was removed using enterokinase. Changes in anisotropy generated by association of the digested protein with Fl-TNF␣ ARE were not significantly different than those observed using mock-digested protein (data not shown), indi- ) with A R-free held constant and A R-bound and K unfixed (dotted line). A residual plot was also prepared to detect any bias for data subsets resulting from the sequential dimer binding algorithm (C). cating that the His 6 motif does not contribute to RNA binding or complex assembly. Additionally, no significant change in anisotropy of the Fl-R␤ probe was observed with increasing dimer concentration of His 6 -p37 AUF1 -(1-257) (Fig. 5, open circles), consistent with the gel mobility shift results (Fig. 1D).
Calculated values of K 1 and K 2 for His 6 -p37 AUF1 -(1-257) and His 6 -p37 AUF1 -(1-229) binding to the Fl-TNF␣ ARE and Fl-U 32 RNAs are listed in Table I. Based on these data, we conclude that specific recognition of both the TNF␣ ARE and the uridylate homopolymer by AUF1 does not require sequences Cterminal of the RRMs. In addition, binding affinity of the second dimer (K 2 ) is enhanced relative to K 1 when multiple identical binding sites are present (Fl-U 32 : K 2 /K 1 Ͼ1). However, for the TNF␣ ARE, binding of the second His 6 -p37 AUF1 -(1-229) dimer occurred with much lower affinity than the first. This indicates that, unlike Fl-U 32 , the Fl-TNF␣ ARE probe does not present multiple high affinity sites for AUF1 binding. However, inclusion of the 29 amino acid residues C-terminal of the RRMs in His 6 -p37 AUF1 -(1-257) increases the affinity of AUF1 dimers for the AUF1 dimer-RNA complex, thus enhancing AUF1 tetramer formation on a physiologically relevant RNA target sequence.
Dynamic Nature of AUF1⅐ARE Equilibria-To evaluate the dynamics of AUF1-ARE interactions, the stability of His 6 -p37 AUF1 -(1-257) complexes with Fl-TNF␣ ARE was monitored in solution by off-rate analysis. First, binding reactions were generated with 10 nM His 6 -p37 AUF1 -(1-257) dimer and 0.2 nM Fl-TNF␣ ARE. Based on the estimates of K 1 and K 2 (Table I), this binding reaction produces a mixed population of binding products. This is supported by the measured anisotropy value (A t ϭ 0.091 Ϯ .002, n ϭ 3; Fig. 6, t ϭ 0), which falls between the estimated intrinsic anisotropy for P 2 R (0.080) and the calculated value of A P4R for this binding reaction (0.097 Ϯ 0.001), as well as the corresponding distribution of binding products identified by gel mobility shift assay (Fig. 1B, 20 nM His 6 -p37 AUF1 -(1-257) monomer). Addition of a 5000-fold molar excess of unlabeled TNF␣ ARE to the binding reaction resulted in a rapid decrease in measured anisotropy with an apparent halflife of less than 10 s (Fig. 6, solid circles). The decrease in anisotropy resolved to values for protein-free samples (open circles), indicating complete dissociation of fluorescent complexes. Similar dissociation kinetics were observed using the U 32 probe as a competitor (data not shown). However, His 6 -p37 AUF1 -(1-257)⅐Fl-TNF␣ ARE complexes were stable in the presence of excess unlabeled R␤ RNA, as minimal changes in fluorescence anisotropy were observed following its addition to the binding reaction (Fig. 6, triangles). These data demonstrate that the interaction of AUF1 with high affinity RNA targets is dynamic in solution and that U-rich RNA sequences are sufficient for protein recognition. DISCUSSION In this study, we have employed oligoribonucleotide target sequences to investigate mechanisms involved in the RNA-dependent oligomerization of AUF1. We propose that AUF1 associates with an ARE as a dimer in solution and that this initial binding event permits subsequent interaction with additional AUF1 dimers to form the oligomeric complex. In a previous work, AUF1 multimers as large as hexamers were identified complexed with the ARE from c-fos mRNA, a 75-nucleotide sequence contributing to the rapid decay of this transcript (24). In this study, shortened RNA targets (Ͻ40 nucleotides) served to limit the complexity of the AUF1 multimers, allowing mathematical models of their assembly to be derived and tested. Furthermore, the use of 5Ј-fluorescein-labeled RNA substrates allowed RNA-protein binding events to be evaluated in solution under true equilibrium conditions by fluorescence anisotropy.
Association of His 6 -p37 AUF1 -(1-257) and His 6 -p37 AUF1 -(1-  . A R was measured as the anisotropy of each RNA substrate in the absence of AUF1 (Fl-TNF␣ ARE, A R ϭ 0.049; Fl-U 32 , A R ϭ 0.039). A P2R was set to 0.080 based on the estimate of this constant for His 6 -p37 AUF1 -(1-229) binding to Fl-TNF␣ ARE (Fig. 4). Equilibrium constants are listed as mean Ϯ S.D. for the indicated number of experiments (n) with each experiment consisting of a minimum of 30 data points and were calculated based on 100% protein activity, hence reflecting the lower limit of AUF1 binding affinity. 229) with both the core ARE from TNF␣ mRNA as well as a uridylate homopolymer were well described by a sequential dimer binding model, allowing equilibrium constants for both stages of tetramer assembly at 25°C to be estimated. In each case, K d values resolved to the low nanomolar range indicating that C-terminal sequences are not requisite for efficient and specific recognition of RNA target sequences.
The sequential nature of the AUF1 binding mechanism is consistent with an induced fit model (41) for ARE recognition by this protein, in which structural rearrangements in one or more binding partners are requisite for high affinity interaction between them. In particular, binding of an AUF1 dimer to an "optimal" site on a target RNA may facilitate additional binding events at adjacent suboptimal site(s) due to free energy contributions from protein-protein contacts. This may partially explain the sequence heterogeneity observed among different AREs (3,9,21,42), since stringent sequence conservation may only be required at the initial AUF1 contact site. Although protein-protein cross-linking suggests interaction between adjacent AUF1 dimers on an ARE, it has not been proven whether binding of subsequent dimers necessarily includes the generation of additional RNA-protein contacts. However, the relative binding affinities (K 2 /K 1 ) for AUF1 binding to Fl-U 32 and Fl-TNF␣ ARE suggest that this is the case. With Fl-U 32 , the potential exists for multiple identical AUF1-binding sites on a single RNA target. The observation that K 2 Ͼ K 1 for both AUF1 mutants binding this probe, together with the potential for tetrameric cross-linked proteins on U 32 , indicates that the binding affinity of the second dimer is likely the result of both RNA-protein and protein-protein contacts. By contrast, the free energy contributing to K 1 is largely, if not solely, attributable to the RNA-protein interaction. By this model, the absence of cross-linked His 6 -p37 AUF1 -(1-229) tetramers on the TNF␣ ARE probe, combined with resolution of K 2 Ͻ Ͻ K 1 for this interaction, indicates that multiple identical AUF1-binding sites do not exist on this RNA. However, cross-linking of His 6 -p37 AUF1 -(1-257) tetramers on the TNF␣ ARE suggests that interaction between dimers of this AUF1 mutant may improve the energetics of P 4 R complex formation. The increased ratio of K 2 /K 1 for this binding event relative to His 6 -p37 AUF1 -(1-229) binding supports this model and furthermore implicates AUF1 sequences between residues 229 and 257 in RNA-dependent tetramer formation on the TNF␣ ARE.
In this study, interaction between AUF1 and the U 32 RNA was observed using several experimental techniques. These results demonstrate that AUUUA motifs are not requisite for AUF1 binding and further suggest that uridylate residues may be the primary determinants of RNA recognition by AUF1. However, in previous studies, gel mobility shift assays failed to detect a high affinity interaction between AUF1 and a U 32 homopolymer contained within a ␤-globin chimeric RNA fragment (23). This ␤-globin/U 32 cassette was also ineffective in initiating mRNA turnover in a transfected cell system (4). One explanation for this apparent discrepancy is that AUF1 may be sterically or conformationally hindered from binding U 32 in the context of the ␤-globin chimera as a result of flanking RNA sequences. Interaction between AUF1 and U-rich RNA sequences is also supported by the observation that cellular AUF1 can be purified by affinity chromatography over poly(U)agarose (25) and that adenylate residues are poorly represented in some U-rich sequences that act as mRNA destabilizers (42). Taken together, these data indicate that the role of interspersed adenylate residues in AREs may be secondary to AUF1 recognition. However, the conservation of these A residues in AREs across species (6, 7) suggests that they function at some other level, possibly involving the association of ancil-lary factors or by contributing to the positioning of AUF1 dimeric units in the oligomeric state. This may also account for the His 6 -p37 AUF1 -(1-229) tetramers generated by DSP-mediated cross-linking in the presence of U 32 but not TNF␣ ARE (Fig. 3B), since the lack of interspersed A residues in U 32 RNA may permit conformational variants of a P 4 R complex that are unfavorable on a bona fide ARE.
Another interesting feature of the AUF1⅐TNF␣ ARE binding equilibria were their rapid off-rates (t1 ⁄2 Ͻ10 s), which may contribute to the efficient recognition of AREs within complex RNA populations. Whereas some complex dissociation was observed by gel mobility shift assays, the clear resolution of AUF1⅐RNA binding events despite rapid dissociation kinetics (Fig. 1) suggests that gel "caging" effects (43) may be involved in maintaining these complexes during electrophoresis. Alternatively, the presence of multiple RNA-binding domains within the AUF1⅐ARE complex may allow for an "exchange" mechanism, where interaction of an ARE with accessible RNA recognition motifs in the AUF1 complex results in the release of an RNA molecule bound at another site. In this case, AUF1-ARE interactions may be stable under limiting RNA concentrations, whereas excess target RNA would induce rapid complex dissociation.
Current data indicate that the binding of AUF1 to AREcontaining mRNAs may function as a targeting system, perhaps directing the association of other factors necessary for the initiation of mRNA turnover (21). To this end, AUF1 oligomerization may represent the ARE-dependent generation of a binding surface for some other factor(s). This notion is also supported by the hydrodynamic properties of the c-fos ARE⅐AUF1 6 complex, where changes in calculated frictional ratios were consistent with maximization of protein surface area (24). Given the dynamic nature of the AUF1-ARE interaction, it is likely that binding of ancillary factors also serves to stabilize the complex in solution. Recently, several cytoplasmic proteins associated with an AUF1-containing complex have been identified, including factors participating in the regulation of translation (eIF4G) and mRNA turnover (poly(A)-binding protein) (26). Heat shock protein 70, which also co-immunoprecipitates with cytoplasmic AUF1 (26), can independently associate with A ϩ U-rich RNA sequences (44). Once assembled, this multisubunit complex may somehow target mRNAs for decay, possibly by directing specific ribonucleases to mRNAs in an AREdependent fashion.