Assembly of Functional Ribonucleoprotein Complexes by AU-rich Element RNA-binding Protein 1 (AUF1) Requires Base-dependent and -independent RNA Contacts*

Background: AUF1 post-transcriptionally regulates mRNA targets. Results: Short AU-rich sequences nucleate AUF1 ribonucleoproteins but require non-ionic contacts with flanking RNA to stabilize complexes and manipulate local RNA structure. Conclusion: AUF1 uses several molecular determinants to bind and remodel RNA targets. Significance: This model explains AUF1 interactome diversity and predicts allosteric effects on protein and microRNA trans-factor binding to proximal sites. AU-rich element RNA-binding protein 1 (AUF1) regulates the stability and/or translational efficiency of diverse mRNA targets, including many encoding products controlling the cell cycle, apoptosis, and inflammation by associating with AU-rich elements residing in their 3′-untranslated regions. Previous biochemical studies showed that optimal AUF1 binding requires 33–34 nucleotides with a strong preference for U-rich RNA despite observations that few AUF1-associated cellular mRNAs contain such extended U-rich domains. Using the smallest AUF1 isoform (p37AUF1) as a model, we employed fluorescence anisotropy-based approaches to define thermodynamic parameters describing AUF1 ribonucleoprotein (RNP) complex formation across a panel of RNA substrates. These data demonstrated that 15 nucleotides of AU-rich sequence were sufficient to nucleate high affinity p37AUF1 RNP complexes within a larger RNA context. In particular, p37AUF1 binding to short AU-rich RNA targets was significantly stabilized by interactions with a 3′-purine residue and largely base-independent but non-ionic contacts 5′ of the AU-rich site. RNP stabilization by the upstream RNA domain was associated with an enhanced negative change in heat capacity consistent with conformational changes in protein and/or RNA components, and fluorescence resonance energy transfer-based assays demonstrated that these contacts were required for p37AUF1 to remodel local RNA structure. Finally, reporter mRNAs containing minimal high affinity p37AUF1 target sequences associated with AUF1 and were destabilized in a p37AUF1-dependent manner in cells. These findings provide a mechanistic explanation for the diverse population of AUF1 target mRNAs but also suggest how AUF1 binding could regulate protein and/or microRNA binding events at adjacent sites.

Cellular homeostasis and function are intimately coupled to the precise control of gene expression. Accordingly, cells can regulate expression of specific genes at many levels, giving them strict control over the timing, abundance, and location of encoded gene products. For any protein coding gene, a key parameter contributing to the rate of protein production is the cytoplasmic concentration of its mRNA, which is established by the relative rates of mRNA synthetic processes including transcription, pre-mRNA processing, and nucleocytoplasmic transport, countered by the rate of cytoplasmic mRNA degradation (for review, see Refs. 1 and 2). Although cytoplasmic decay mechanisms regulate levels of all mRNAs, turnover kinetics vary widely across the cellular mRNA population. Gene-specific control of mRNA decay is mediated by cis-acting sequences located within each transcript, frequently localized to the 3Ј-untranslated region (3Ј-UTR). A prototypical family of mRNA stability determinants are the AU-rich elements (AREs) 3 frequently found in the 3Ј-UTRs of mRNAs encoding oncoproteins, cytokines, and inflammatory mediators (for review, see Ref. 3). AREs normally consist of a U-rich domain and frequent include one or more AUUUA pentamers, which genomic analysis indicates may occur in 5-8% of all human mRNAs (4).
The ability of AREs to control mRNA decay kinetics is governed by their association with cytoplasmic trans-acting factors, which may include both protein and nucleic acid components (2,5). For example, binding by members of the Hu family of proteins including the ubiquitously expressed HuR stabilizes ARE-containing mRNAs, whereas recruitment of tristetraprolin, KSRP, or the microRNA miR-16 is associated with accelerated mRNA decay (2,5,6). However, for the AUF1 family of RNA-binding proteins, the biochemical and functional conse-quences of mRNA substrate binding appear to be more complex. AUF1 is encoded by a single copy gene on chromosome 4q21 that generates four protein isoforms via alternative pre-mRNA splicing (7), named according to their apparent molecular weights as p37 AUF1 , p40 AUF1 , p42 AUF1 , and p45 AUF1 . All isoforms contain two adjacent, centrally positioned RNA recognition motifs (RRMs) followed by a glutamine-rich domain. The p37 AUF1 and p40 AUF1 isoforms lack a 49-amino acid sequence near the C terminus encoded by exon 7, whereas p37 AUF1 and p42 AUF1 lack a 19-amino acid domain immediately N-terminal of RRM1 that is encoded by exon 2. However, although sharing common canonical RNA binding domains, the various AUF1 isoforms exhibit distinct subcellular localization profiles (8 -11), diverse RNA binding affinities and effects on the structure of RNA ligands (12), and isoform-specific influences on the decay of mRNA targets (13)(14)(15)(16)(17).
All AUF1 isoforms form dimers in solution that can bind sequentially to RNA substrates to generate oligomeric ribonucleoprotein (RNP) structures (12,18). To date, most biochemical characterization of RNA recognition by AUF1 has focused on RNA substrates derived from extended, contiguous ARE sequences such as those from the 3Ј-UTRs of mRNAs encoding tumor necrosis factor ␣ (TNF␣) or c-fos or similarly sized polyuridylate ligands (18 -21). Although these studies have revealed many features regarding the mechanism and energetics of AUF1 RNP assembly, several details from these and other reports highlight deficiencies in our understanding of the RNA sequence requirements for high affinity AUF1 binding. First, using macromolecular binding density analysis, we recently demonstrated that the initial dimer binding event for either p37 AUF1 or p42 AUF1 occupies 33-34 nucleotides of RNA (12). Binding assays using truncated ARE substrates confirmed this as the lower size limit for high affinity AUF1 binding. This is a surprisingly large RNA footprint given that HuR, which contains three RRMs that contribute variously to its RNA binding activity, binds ARE substrates as small as 15 nt with low nanomolar affinity (22). A more analogous comparison might be the RNA binding domain of heterogeneous nuclear ribonucleoprotein (hnRNP) A1, which like AUF1 also contains two RRMs and forms dimers but interacts with only 11 nt of nucleic acid in each heteromolecular binding cleft (23). Second, transcripts containing Ͼ30 nt of contiguous AU-rich sequence are rare (24). Consistent with this, screening efforts have identified hundreds of AUF1-targeted mRNAs among the cellular transcript population but with few containing 3Ј-UTR sequences of 30 nt that consist solely of A and U residues (25,26) and to date have not delineated precise RNA requirements for high affinity AUF1 binding. Together these data suggest that only a subset of AUF1-associated RNA nucleotides need be AU-rich to nucleate assembly of functional RNP complexes.
In this study we have tested the hypothesis that, although AUF1 binding must be nucleated by base-specific contacts involving a U-rich RNA domain, stabilization of AUF1 RNP complexes also requires base-independent contacts with flanking RNA sequences. The p37 AUF1 isoform was used as a model for these experiments because it binds ARE substrates with the highest affinity (12) and is most closely associated with destabilization of ARE-containing mRNA targets (15,27). Fluores-cence-based assay systems permitted quantitative assessment of the thermodynamic contributions of RNA ligand subdomains to AUF1 RNP assembly and local RNA structural remodeling, whereas cultured cell models were used to determine whether minimal high affinity AUF1 binding sites were functional. Findings from these experiments provide an explanation for the 3Ј-UTR sequence heterogeneity observed among AUF1-targeted mRNAs but also suggest potential mechanisms for cooperative or competitive relationships between AUF1 and other trans-acting factors for mRNA targets.

EXPERIMENTAL PROCEDURES
RNA Substrates-RNA oligonucleotides were synthesized, 2Ј-hydroxyl-deprotected, and purified by Integrated DNA Technologies, Dharmacon, or Sigma. Lyophilized pellets were resuspended in 10 mM Tris (pH 8.0). RNA concentrations and fluorophore labeling efficiencies were quantified by absorbance, incorporating fractional contributions from fluorescein (Fl) and/or cyanine-3 (Cy3) labels to A 260 as described (20,28). RNA probe sequences are listed in Table 1. Sequences denoted "ARE xx " correspond to fragments of the core ARE from the 3Ј-UTR of TNF␣ mRNA. "R␤ xx " sequences are derived from a non-AU-rich fragment from the coding sequence of rabbit ␤-globin mRNA. RNA substrates named with "Fl" and/or "Cy3" affixes denote the presence and location of the indicated fluorophore. Where indicated, 5Ј-hydroxyl RNA probes were radiolabeled with [␥-32 P]ATP by T4 polynucleotide kinase to a specific activity of 3-5 ϫ 10 3 cpm/fmol as previously described (18).
Preparation of Recombinant AUF1 Proteins-Plasmid pBAD/HisB-p37 AUF1 was described previously (29) and expressed in Rosetta 2 Escherichia coli cells (Novagen). Recombinant His 6 -p37 AUF1 protein was purified from cell lysates using Ni 2ϩ affinity chromatography as previously described (12) except that buffer exchange into 10 mM Tris (pH 8.0) and protein concentration were both performed in Ambion ultracentrifugal filters. Recombinant protein purity and yield were assessed by Coomassie Blue-stained SDS-PAGE against a titration of bovine serum albumin.
Plasmids and Cell Lines-Plasmids expressing AUF1 and control short hairpin RNAs (shRNAs) were a gift from Dr. Gary Brewer. The expressed AUF1 shRNA targets a sequence encoded by exon 3, which is common to all isoforms (30). The plasmids were transfected into both HeLa and HeLa/Tet Off cell lines (Clontech) with Attractene transfection reagent, and stably expressing cell clones were selected with 400 g/ml hygromycin B (Calbiochem). After selection, amplification, and screening, shAUF1 or shControl clonal cell lines were maintained by adding 100 g/ml hygromycin B to regular growth media. Plasmid pcDNA/shR-p37 AUF1 -FLAG encodes a C-terminal FLAG-tagged shRNA-resistant p37 AUF1 , constructed by transferring the modified p37 AUF1 sequence from plasmid shR-p37 AUF1 (a gift from Dr. Gary Brewer; Ref. 31) into plasmid pcDNA3.1 ϩ -FLAG (generously donated by Dr. Bret Hassel) using standard subcloning techniques. Mammalian ␤-globin (␤G) reporter plasmids pTRER␤-wt and pTRER␤-ARE 38 were described previously (21). Plasmid pTRER␤-RAR was constructed by inserting a DNA duplex encoding the R␤ 17 -ARE 16 -R␤ 17 RNA substrate (Table 1) into the BglII site of pTRER␤-wt, thus positioning the R␤ 17 -ARE 16 -R␤ 17 sequence within the reporter mRNA 3Ј-UTR. All plasmid constructs were verified by restriction digest and automated sequencing.
Measurements of RNA-Protein Binding-The number of RNP complexes formed between AUF1 and selected RNA substrates was monitored using electrophoretic mobility shift assays (EMSA) essentially as described (32). Briefly, 32 P-labeled RNA (0.2 nM) was incubated for 15 min on ice with increasing concentrations of His 6 -p37 AUF1 in protein binding buffer (10 mM Tris-HCl (pH 8.0) containing 50 mM KCl, 2 mM DTT, 0.5 mM EDTA, 0.1 g/l acetylated bovine serum albumin, and 1 g/l heparin) plus 10% glycerol. Reaction components were resolved by electrophoresis through 6% native acrylamide gels run at 4°C. Gels were dried, and products were visualized using Typhoon FLA-9500 phosphorimaging (GE Healthcare).
Quantitative assessments of AUF1-RNA binding equilibria were performed using fluorescence anisotropy essentially as described (12,18,29). Binding reactions (100 l) were assembled as described for EMSAs but in buffer lacking glycerol and using Fl-labeled rather than radiolabeled RNA substrates. Reactions were incubated at 25°C for 1 min based on kinetic runs, which verified that equilibrium is attained within this time frame (Refs. 18 and 33 and data not shown). Subsequently, total reaction anisotropy (A t ) and fluorescence intensity were measured using a Beacon 2000 Fluorescence Polarization System (Panvera) equipped with a 490-nm excitation filter and a 535-nm emission filter.
For most protein-RNA binding events, fluorescence intensity did not significantly vary as a function of protein concentration indicating similar quantum yields for all fluorescent reaction components. In these cases, measured anisotropy is proportional to the fractional concentrations of each fluorescent species at constant temperature and viscosity (34,35). For RNA substrates where two p37 AUF1 dimers can bind to form a tetrameric p37 AUF1 -RNA complex, the observed dissociation constants describing the first (K d1-obs ) and second (K d2-obs ) p37 AUF1 dimer binding steps are resolved by nonlinear regres-sion of A t as a function of protein dimer concentration ([P 2 ]) using Equation 1, where K dx-obs ϭ 1/K x .
Here, A R is the intrinsic anisotropy of the free RNA ligand, whereas A P2R and A P4R represent the intrinsic anisotropy values for AUF1 dimer-bound and AUF1 tetramer-bound RNA substrates, respectively. In cases where only a single AUF1 dimer binding event is indicated, this function is simplified by insertion of K 2 ϭ 0 to yield Equation 2, However, for a subset of binding experiments involving short RNA ligands, fluorescence intensity was not constant across the tested range of His 6 -p37 AUF1 concentrations, indicating that the quantum yields of the free RNA ligand (q R ) and the His 6 -p37 AUF1 dimer-bound substrate (q P2R ) were not equivalent. In these cases the variations in quantum yield result in unequal contributions from bound and free RNA substrates to A t . Lundblad et al. (36) described a function to calculate the bound substrate fraction (F b ) under conditions where the quantum yield of free and bound substrate varied. Adapted to our nomenclature, this function is given as Equation 3, For simplification, the term f q has been substituted as the ratio of the fluorescence quantum yields of bound and free RNA substrates (ϭ q P2R /q R ). If the fluorescence intensity of Fl-tagged  24 AUUUAUUUAUUAUUUAUUUAUUUA ARE 16 AUUAUUUAUUUAUUUA ARE 15 UUAUUUAUUUAUUUA R␤ 17 -ARE 16 -R␤ 17 CCUGGCUCACAAAUACCAUUAUUUAUUUAUUUACCCUGGCUCACAAAUAC b R␤ 17 -ARE 14 -R␤ 17 CCUGGCUCACAAAUACCUUAUUUAUUUAUUUCCCUGGCUCACAAAUAC R␤ 17 Equation 4 was used to analyze all RNA binding isotherms where total fluorescence intensity varied by Ͼ20%. In these cases, f q was calculated by dividing the fluorescence intensity for the saturated protein-bound RNA complex (q P2R ) by the fluorescence intensity of the RNA substrate in the absence of protein (q R ). Thermodynamic Parameters Contributing to AUF1 RNP Complex Formation-The contributions of changes in enthalpy (⌬H°) and entropy (T⌬S°) to the free energy of His 6 -p37 AUF1 binding (⌬G°) to selected RNA targets were resolved using van't Hoff analyses. Briefly, observed equilibrium association constants (K obs ) for His 6 -p37 AUF1 binding to selected RNA substrates at various temperatures (T) between 5 and 37°C were measured by fluorescence anisotropy as described above and then plotted as lnK obs versus 1/T. If ⌬H°is constant across the temperature range tested, this plot will be linear and can be resolved using Equation 5, with 1.987 ϫ 10 Ϫ3 kcal⅐mol Ϫ1 ⅐K Ϫ1 as the universal gas constant (R) (37,38).
However, when these plots are nonlinear, the contributions of enthalpy and entropy to binding free energy are no longer constant with temperature. In these cases, lnK obs versus 1/T data sets were resolved using Equation 6, returning the change in molar heat capacity (⌬C°P ,obs ) and critical temperatures T H and T S , at which enthalpy and entropy, respectively, contribute no net energy to formation of AUF1 RNP complexes (38).
Solution of these parameters then allowed calculation of enthalpic and entropic contributions at any given temperature using Equations 7 and 8 (38).
T⌬S°ϭ T⌬C°P ,obs ⅐ln ͩ T T S ͪ (Eq. 8) Analysis of Ionic Contributions to AUF1 RNP Assembly-Due to the highly negative charge on nucleic acid backbones, ions can be a significant factor driving the formation of proteinnucleic acid complexes, dominated by the release of cations from the nucleic acid. For a single protein (P) binding event on RNA (R) this equilibrium can be expressed using Equation 9.
Here, ZM ϩ represents the ions that are released as a result of ion bridges formed between the protein and nucleic acid that neutralize backbone phosphates (39). Specifically, denotes the fraction of monovalent countercation (M ϩ ) associated with the RNA per phosphate that approximates the axial charge density of the nucleic acid, and Z is the number of ion pairs formed between the RNA and protein. In this manner, the protein can be considered as a Z-valent ligand that neutralizes Z phosphates on RNA (40,41). Following Le Chatelier's principle, the observed association constant between protein and RNA (K obs ) can thus be modulated by changes in the concentration of solution monovalent cation ([M ϩ ]) using Equation 10, Incorporating K T as the association constant at 1 M salt and taking logarithms yields Equation 11, When linear, the slope of a log K obs versus log[M ϩ ] plot can thus resolve Z if is known. is related to a parameter termed , which is sequence-, pH-, and counterion-dependent and proportional to the structural charge density (42,43). For polyU RNA, the value for Ϫ1 has been measured as 0.615 (44), thus resolving ϭ 1 Ϫ (2) Ϫ1 ϭ 0.693 (43). However, a nonlinear relationship between logK obs and log[M ϩ ] indicates that the displacement of cations from nucleic acid is not the only ionic event occurring upon protein binding. To account for these broader effects of ionic strength on RNP formation, we employed a counterion binding model derived by the Record (45) and Lohman (43) laboratories as adapted by Stickle and Fried (46,47). This model includes changes in the numbers of protein-associated cations (⌬m) and anions (⌬n) resulting from protein binding to the nucleic acid target in addition to the change in cations associated with the nucleic acid (⌬q). Incorporating these terms into Equation 11 and using [MX] in place of [M ϩ ] to indicate consideration of both cations and anions from the salt yields Equation 12 (45,47).
Langmuir isotherms in Equation 13 model the change in cation association with the protein upon binding RNA based on differences in the average concentration of cation near the RNA Mϩ is the population averaged dissociation constant for cation to protein (47).
The parameter m tot is the number of cation binding sites transferred from the bulk solution to the RNA environment upon assembly of the RNP complex. Because[M ϩ ] R is largely independent of bulk salt concentration between 10 mM to 1 M salt Here, ⌬t represents an aggregate ion release stoichiometry where ⌬t ϭ ⌬n ϩ ⌬q and K a M ϩ ϭ 1/K d M ϩ (46). To define ionic contributions to the stability of His 6 -p37 AUF1 RNP complexes, observed equilibrium binding constants (K obs ) were calculated from fluorescence anisotropy-based binding assays across a range of KCl concentrations (25-500 mM) and plotted as log K obs versus log [KCl]. Values of logK T , m tot , K a M ϩ , and ⌬t were resolved by nonlinear regression to Equation 14.
Analysis of RNA Folding Using Fluorescence Resonance Energy Transfer (FRET)-Protein-induced changes in the global structure of RNA substrates were monitored by alterations in the distance between 3Ј-Fl (donor) and 5Ј-Cy3 (acceptor) dyes appended to each RNA ligand using FRET essentially as described previously (12,21). FRET efficiency (E FRET ) is related to the interfluorophore distance by Equation 15 where r denotes the scalar distance between the donor and acceptor, and R 0 is the Förster distance defined as that yielding E FRET ϭ 0.5 (48). For Fl and Cy3 linked to single-stranded nucleic acids, R 0 has been calculated as 55.7 Å (49).
RNA binding reactions with varying concentrations of His 6 -p37 AUF1 were assembled as described for fluorescence anisotropy measurements but in paired samples containing RNA substrates (2 nM) labeled either with a 3Ј-Fl (donor alone) or both 3Ј-Fl and 5Ј-Cy3 dyes (donor-acceptor pair). Donor fluorescence from each sample was measured by excitation at 485 nm and emission at 520 nm or scanning from 500 to 620 nm (10-nm slit widths) using a Cary Eclipse fluorometer (Varian). Background fluorescence was measured from samples lacking RNA probe. Inner filter effects and photobleaching were insignificant under these conditions (data not shown). E FRET was calculated from the fluorescence of the donor in the presence (F DA ) or absence (F D ) of the FRET acceptor using Equation 16, where f represents the labeling efficiency of the Cy3 acceptor dye for each double-labeled RNA ligand (21).
Ribonucleoprotein Immunoprecipitation (RNP-IP)-HeLa/ shAUF1 cells were co-transfected with pcDNA/shR-p37 AUF1 -FLAG, specific ␤G reporter plasmids encoding transcripts containing or lacking putative p37 AUF1 -binding sites, and plasmid pEGFP-C1 (Clontech) to control for transfection efficiency. The amounts of transfected ␤G plasmids were optimized to yield similar steady-state reporter mRNA concentrations. RNP complexes containing FLAG-p37 AUF1 were then isolated from crude cell lysates by RNP-IP and precipitated RNA purified essentially as described (50,51). Levels of ␤G reporter and EGFP mRNAs recovered in unfractionated cytoplasm and immunoprecipitated RNP complexes were measured by multiplex TaqMan quantitative reverse transcription PCR (qRT-PCR) using the iScript One-Step qRT-PCR Kit for Probes (Bio-Rad) programmed with ␤G amplification primers GTGAACT-GCACTGTGACAAGC and ATGAGTAGACAGCACAATA-ACCAG, ␤G TaqMan probe Fl-CGTTGCCCAGGAGCCTG-AAGTTCTCA-Black Hole Quencher 1, EGFP primers GCGA-CACCCTGGTGAACC and GATGTTGTGGCGGATCTTG-AAG, and EGFP TaqMan probe Texas Red-CACCTTGATGC-CGTTCTTCTGCTTGTCG-Black Hole Quencher 2 (all 5Ј to 3Ј and from Integrated DNA Technologies). ␤G reporter mRNA levels measured in each RNP-IP reaction were then normalized to EGFP mRNA levels and expressed as a fraction of the total input level of each reporter transcript as described (26).
mRNA Decay Assays-The decay rates of ␤G reporter mRNAs were measured using doxycycline (Dox) time courses essentially as described previously (51,52). Briefly, HeLa/Tet-Off cells were cotransfected with plasmids encoding specific reporter transcripts and pEGFP-C1. After 48 h, transcription of ␤G mRNA was inhibited with Dox (2 g/ml; Sigma). Cells were harvested at time points thereafter, and DNA-free RNA was purified using the NucleoSpin RNA II kit (Macherey-Nagel). ␤G reporter mRNA levels were quantified relative to EGFP mRNA at each time point by qRT-PCR using the iScript One- Step qRT-PCR Kit for Probes (Bio-Rad) as described above. The percentage of reporter mRNA remaining in each sample was plotted as a function of time after Dox treatment, and first order decay constants (k) were resolved by nonlinear regression. From this, the mRNA half-life (t1 ⁄ 2 ) was calculated using t1 ⁄ 2 ϭ ln2/k.
Statistical Analyses-All regression and statistical analyses were performed using PRISM v3.03 software (GraphPad). Comparisons between resolved thermodynamic parameters, mRNA recoveries by RNP-IP, and mRNA decay kinetics were done using the unpaired t test. For all regression analyses, the appropriateness of analytical binding models was evaluated by the coefficient of determination (R 2 ) and analyses of residual randomness. Pairwise comparisons between analytical binding or van't Hoff models were performed using the F test where indicated. In all cases differences yielding p Ͻ 0.05 were considered significant.

p37 AUF1 Binding to Short (Յ 16-nt) ARE Substrates Is Dramatically Stabilized by the Addition of Flanking non-ARE
Sequences-Previously we showed that p37 AUF1 dimers bind sequentially with high affinity to ARE-based RNA substrates of 34 nt or more in length (12). Binding is substantially weaker on ARE substrates of 30 nt or less consistent with the p37 AUF1 dimer binding site size of 33 nt on RNA calculated using macromolecular binding density analysis (12). On ARE ligands Ն34 nt, His 6 -p37 AUF1 dimer binding generates two distinct RNP complexes (Fig. 1A, left panel) consistent with the formation of AUF1 tetramers on these substrates (18). As noted previously, p37 AUF1 RNPs are retained relatively weakly by EMSA, particularly at the first (P 2 R) binding event, likely because of the highly dynamic nature of these complexes (12,18). However, a two-step model of His 6 -p37 AUF1 RNP assembly on a 38-nt ARE ligand was further supported by resolution of fluorescence anisotropy isotherms using Equation 1 (Fig. 1B, left panel), which returns the observed dissociation constants describing both AUF1 dimer binding events. Conversely, shorter ARE ligands revealed only a single dimer binding event at protein concentrations up to 500 nM, both by EMSA (Fig. 1A, center panel) and resolution of anisotropy isotherms using the single site binding model described by Equation 2 (Fig. 1B, right panel), while calculated affinity constants affirm significant energetic penalties for His 6 -p37 AUF1 binding to shorter ARE substrates (Fig. 1E). These data indicate that extended RNA substrates are required to stabilize formation of His 6 -p37 AUF1 RNP complexes, far larger than the 9 -15-nt RNA ligands needed to form stable complexes with HuR or the RNA binding domains of tristetraprolin or hnRNP A1 (22,23,53). However, they do not establish whether AUF1 binding is restricted to extended ARE substrates or if only a subset of the RNA ligand needs to be AU-rich to nucleate RNP assembly.
To discriminate between these possibilities, we assessed His 6 -p37 AUF1 binding to a series of chimeric RNA substrates based on a 16-nt ARE flanked by sequences from a domain of the ␤-globin mRNA coding region (R␤ 31 ) that does not detectably bind His 6 -p37 AUF1 (Fig. 1, A, right panel, and B, left panel). By fluorescence anisotropy, His 6 -p37 AUF1 binding to the ARE 16 -Fl substrate alone was weak but detectable (Fig. 1B, right  panel). However, an RNA ligand incorporating 17 nt from the R␤ 31 sequence at both the 5Ј-and 3Ј-ends of the ARE 16 substrate (R␤ 17 -ARE 16 -R␤ 17 ) formed two distinct RNP complexes with His 6 -p37 AUF1 by EMSA (Fig. 1C), similar to the ARE 38 substrate. Anisotropy-based binding assays also supported a two-step binding mechanism for the R␤ 17 -ARE 16 -R␤ 17 -Fl ligand, as the single site model of Equation 2 yielded significant residual nonrandomness (p ϭ 0.006; Fig. 1D, dotted line) and significantly increased sum-of-squares deviations relative to regression solutions based on the two-step binding model using the F test (p Ͻ 0.0001; Fig. 1D, solid line). However, although His 6 -p37 AUF1 binding to the R␤ 17 -ARE 16 -R␤ 17 substrate demonstrated that extended ARE domains were not obligatory to form AUF1 tetrameric RNP complexes, more dramatic was the improvement in affinity of the initial His 6 -p37 AUF1 dimer binding event (K d1-obs ), which was strengthened by nearly 20-fold for R␤ 17 -ARE 16 -R␤ 17 versus the ARE 16 sequence alone (Fig.  1E). These data demonstrate that a smaller ARE sequence is sufficient to nucleate AUF1 RNP assembly if placed in a larger RNA context and that flanking nucleotides likely make baseindependent contributions to the stability of AUF1 RNP complexes. However, shortening the ARE domain further weakened His 6 -p37 AUF1 RNP formation (Fig. 1E).
A 3Ј Purine and Nonspecific 5Ј Nucleotides Significantly Stabilize p37 AUF1 Binding to a Minimal ARE Substrate-To define the features of non-ARE flanking sequences required to stabilize p37 AUF1 RNP assembly on a minimal ARE target, we measured the affinity of His 6 -p37 AUF1 binding across a panel of RNA substrates by fluorescence anisotropy. Each RNA substrate was based on an ARE 15 target for two reasons. First, its small size limits RNP complexes to a single AUF1 dimer, thus permitting the initial His 6 -p37 AUF1 binding event to be effectively separated from subsequent formation of protein tetramers, which conceivably may be influenced by different RNA structural determinants. Second, His 6 -p37 AUF1 binding to the ARE 15 -Fl ligand is very weak in the absence of flanking sequence (Fig. 2), allowing us to accurately quantitate the contributions of adjacent RNA elements to RNP stability over an extended range.
Appending 14 nt of R␤ sequence to the 3Ј-end of the ARE 15 substrate (ARE 15 -G-R␤ 13 -Fl) improved His 6 -p37 AUF1 binding affinity nearly 12-fold (⌬⌬G°ϭ Ϫ1.5 kcal/mol; Fig. 2). Interestingly, limiting the 3Ј-flanking sequence to the initial base, a guanine, was sufficient to confer most of this RNP stabilizing effect (⌬⌬G°ϭ Ϫ1.1 kcal/mol). The observation that the affinity of His 6 -p37 AUF1 for the ARE 15 -G-Fl substrate was significantly greater than that for ARE 16 -Fl (Fig. 1E) suggested that contributions of this 3Ј-G residue to RNP stability may be basespecific. To test this possibility, His 6 -p37 AUF1 binding was also measured to RNA substrates ARE 15 -A-Fl, ARE 15 -C-Fl, and ARE 15 -U-Fl. Although neither the 3Ј-C nor U residues significantly enhanced protein binding affinity relative to ARE 15 -Fl or ARE 16 -Fl, His 6 -p37 AUF1 bound the ARE 15 -A-Fl substrate with an affinity identical to ARE 15 -G-Fl, indicating that p37 AUF1 RNPs are stabilized specifically by a purine residue 3Ј of the core ARE sequence. Conceivably, the purine base could directly interact with AUF1 or stabilize the complex through an allosteric mechanism.
To assess the contributions of non-ARE residues upstream of the ARE 15 core, various lengths of R␤ sequence were appended to the 5Ј-end of the ARE 15 -G-Fl substrate. The affinity of His 6 -p37 AUF1 binding increased as a function of 5Ј-R␤ sequence length, with 19 nt (R␤ 19 -ARE 15 -G-Fl) improving protein affinity Ͼ5-fold (⌬⌬G°ϭ Ϫ1.0 kcal/mol). Experiments with two additional RNA substrates showed that both base-specific and non-base-specific interactions contribute to RNP stabilization by 5Ј-flanking sequences. First, adding 4 nt of R␤ sequence (R␤ 4 -ARE 15 -G-Fl) or 4 nt of ARE sequence (ARE 19 -G-Fl) 5Ј of ARE 15 -G yielded essentially identical improvements in the affinity of His 6 -p37 AUF1 binding (Fig. 2), indicating a base-independent effect on RNP stability. However, although the addition of 19 C residues upstream of ARE 15 -G-Fl also significantly stabilized p37 AUF1 RNP formation (C 19 -ARE 15 -G-Fl; ⌬⌬G°ϭ Ϫ0.5 kcal/mol), stabilization by the C 19 domain was only half that observed with the 5Ј-R␤ 19 sequence, suggesting that some primary or potentially secondary structural features of the R␤-based 5Ј-domain make additional contributions to His 6 -p37 AUF1 binding. 5Ј-and 3Ј-Non-ARE Nucleotides Make Distinct Thermodynamic Contributions to p37 AUF1 RNP Assembly-To identify thermodynamic features contributing to p37 AUF1 RNP formation by sequences flanking the ARE 15 RNA ligand domain, enthalpic and entropic contributions to RNP stability were assessed using van't Hoff analyses. The role of the 5Ј non-ARE domain was determined by comparing His 6 -p37 AUF1 binding to RNA substrates R␤ 19 -ARE 15 -G-Fl versus ARE 15 -G-Fl. For ARE 15 -G-Fl, a plot of lnK obs versus 1/T appeared to show modest downward curvature (Fig. 3A, top panel). Fitting these data to the linear van't Hoff function described by Equation 5 yielded reasonably confident values for the enthalpic (⌬H°) and entropic (T⌬S°) contributions to binding energy, and a residual runs test did not indicate significant deviation from linearity (p ϭ 0.117). However, an F test comparison of regression solutions to Equations 5 versus 6 strongly favored the latter model (p Ͻ 0.0001), and as such the thermodynamics of His 6 -p37 AUF1 binding to ARE 15 -G-Fl RNA were interpreted using the nonlinear function. By this model, p37 AUF1 binding to this RNA target is associated with a small but statistically significant negative change in heat capacity (⌬C°P ,obs ) ( Table 2) and a correspond-ing temperature dependence of both ⌬H°and T⌬S°. Solving these parameters as a function of temperature (Fig. 3A, bottom  panel) revealed that, at all temperatures tested, His 6 -p37 AUF1 binding to the ARE 15 -G-Fl ligand is driven by enthalpic effects (⌬H°Ͻ 0) and that above 10°C, this must overcome unfavorable changes in system entropy (T⌬S°Ͻ 0). A negative ⌬C°P ,obs is commonly seen in "induced fit" protein-nucleic acid binding mechanisms that involve conformational rearrangements within one or both binding partners (37,53,54).
By comparison, the negative curvature observed in the van't Hoff plot of His 6 -p37 AUF1 binding to RNA substrate R␤ 19 -ARE 15 -G-Fl was much more pronounced (Fig. 3A, top panel). These data resolved a significantly larger negative value for ⌬C°P ,obs than that observed for p37 AUF1 binding to the RNA ligand lacking the 5Ј-R␤ 19 domain ( Table 2), suggesting that upstream nucleotide contacts may induce new or more pronounced conformational changes within RNP complex constituents. Consistent with this model, there is a larger entropic penalty for p37 AUF1 binding to R␤ 19 -ARE 15 -G-Fl compared with the ligand lacking the 5Ј-R␤ sequence (ARE 15 -G-Fl) at 25°C, although enhanced enthalpic contributions by the 5Ј-R␤ 19 domain more than compensate for this ( Table 2). The improvement in ⌬H°associated with the 5Ј-R␤ 19 domain is consistent with additional protein contacts involving this sequence, whereas the increased entropic penalty suggests that these contacts may limit conformational freedom near or solvent exclusion from the binding interface.
The effects of a 3Ј-purine residue on the energetic contributions to p37 AUF1 RNP assembly were assessed using similar van't Hoff analyses but comparing His 6 -p37 AUF1 binding to RNA ligands R␤ 8 -ARE 15 -Fl and R␤ 8 -ARE 15 -G-Fl (Fig. 3B). Comparisons were performed in the presence of the 8-nt 5Ј-R␤ sequence because very weak protein binding by the ARE 15 -Fl ligand (Fig. 2) precluded accurate measurements of its affinity across the temperature range necessary to confidently resolve thermodynamic components of binding energy. The presence of the 3Ј-G residue did not significantly alter the negative value of ⌬C°P ,obs associated with His 6 -p37 AUF1 binding but did enhance both the enthalpic benefit and entropic penalties of RNP assembly at 25°C (Table 2), suggesting the formation of additional contacts between the protein and the 3Ј-terminal G.
A common base-independent mechanism that stabilizes protein binding to nucleic acids is through ionic interactions between positively charged functional groups on the protein and the negatively charged nucleic acid backbone (55)(56)(57). To determine whether new ionic contacts involving 5Ј-flanking sequences or a 3Ј-purine (using G as a model) stabilize p37 AUF1 binding to ARE 15 -containing RNA ligands, we measured the affinity of His 6 -p37 AUF1 for selected RNA substrates across a range of KCl concentrations, then plotted log K obs versus log- [KCl]. Frequently, these analyses are well described by linear functions, as high solution cation concentrations compete with the protein for the negatively charged phosphodiester nucleic acid backbone (22,53,55,58,59). However, in all cases tested logK obs displayed a multiphasic response to changes in log[KCl] (Fig. 4). In high salt, the RNA binding activity of His 6 -p37 AUF1 was weakened, consistent with competition for ion pairs contributing to RNP stability. However, under hypotonic condi- Anisotropy-based binding assays containing His 6 -p37 AUF1 and the indicated Fl-tagged RNA substrates were performed across a range of reaction temperatures (5-37°C). Observed association constants (K obs ) were resolved using Equation 2. A, van't Hoff plots of lnK obs versus 1/T. Each point is derived from the value of K obs calculated from an independent binding isotherm Ϯ the S.E. of regression (approximately half of the 95% confidence interval) calculated by PRISM v3.03 software. Data sets generated using each RNA ligand were then fit using Equation 6. Resolved thermodynamic constants are listed in Table 2. B, enthalpic (⌬H°, solid lines) and entropic (T⌬S°, dashed lines) contributions to the stability of His 6 -p37 AUF1 complexes on each RNA ligand were calculated as a function of temperature using Equations 7 and 8, respectively. Changes in enthalpy and entropy contributed no net energy to ⌬G°at their critical temperatures of T H and T S , respectively, indicated by the intersection of these functions with ⌬G°ϭ 0 (dotted line). The total free energy of RNP formation (⌬G°) was determined at each temperature using ⌬G°ϭ ϪRTlnK obs (open circles).
tions His 6 -p37 AUF1 binding was also impaired, suggesting that specific macromolecular interactions with ions in solution could also be required to stabilize p37 AUF1 RNP complexes or to maintain the protein in an active conformation as has been postulated for binding of the E. coli CAP and AraC proteins to cognate DNA targets (46,60). To incorporate broader roles for solvated ions in p37 AUF1 RNP assembly, logK obs versus log[KCl] plots were analyzed using a more inclusive counterion binding model described by Equation 14 (Fig. 4, solid lines). Among the resolved parameters, this model estimates the stoichiometry of ion release upon macromolecular binding (⌬t), which is the sum of anions released from His 6 -p37 AUF1 and cations released from RNA (Table 3). An increase in ion pairs formed between His 6 -p37 AUF1 and an RNA substrate would thus be expected to increase the value of ⌬t, as localized counterions would be ejected from macromolecular binding partners during RNP assembly. Interestingly, neither the 5Ј-non-ARE sequences ( Table 3, cf. ARE 15 -G-Fl versus R␤ 8 -ARE 15 -G-Fl and R␤ 19 -ARE 15 -G-Fl) nor the 3Ј-G residue (cf. R␤ 8 -ARE 15 -Fl versus R␤ 8 -ARE 15 -G-Fl) significantly altered ⌬t, suggesting that the enhancements in RNP stability conferred by each flanking RNA domain do not require the formation of new ionic contacts with the protein. This was not unexpected for the enhanced affinity conferred by the 3Ј-purine because of its specificity for an A or G nucleobase. However, given that RNP stabilization by the 5Ј-R␤-derived domains includes substantial base-independent components, these data suggest that non-ionic interactions, possibly including contacts with functional groups on ribose sugars or allosteric mechanisms, are primarily responsible for the contributions of 5Ј-flanking sequences to p37 AUF1 binding.
Sequences 5Ј of the Nucleating ARE Domain Are Required for p37 AUF1 -induced Condensation of Local RNA Structure-The observation that adding R␤-derived nucleotides 5Ј of the ARE 15 -G domain significantly increased the negative change in heat capacity upon His 6 -p37 AUF1 binding suggested that sequences upstream of ARE 15 may induce or enhance structural remodeling upon RNP formation. Conceivably, this could involve conformational changes in the protein and/or RNA moieties. However, given that we have previously reported a global condensation of local RNA structure associated with initial AUF1 dimer binding events for all isoforms (12,20,33), we investigated whether protein contacts 5Ј of the core ARE 15   Table 3.

TABLE 3
Ionic contributions to p37 AUF1 RNP formation on selected RNA substrates For His 6 -p37 AUF1 binding to each RNA substrate the number of cation binding sites transferred from the bulk solution to the RNA environment upon assembly of the RNP complex (m tot ), the population averaged association constant of K ϩ for His 6 -p37 AUF1 (K a M ϩ ), the affinity constant of His 6 -p37 AUF1 for each RNA ligand (K T ), and the aggregate ion-release stoichiometry of anions from AUF1 and cations from the RNA (⌬t) were resolved by nonlinear regression of logK obs versus log͓KCl͔ plots by Equation 14 as shown in Fig. 4. All parameter values are listed Ϯ the S.E. of regression (approximately half of the 95% confidence interval) calculated by PRISM v3.03 software.

RNA substrate
sequence were required for the RNA conformational remodeling activity of p37 AUF1 . Protein-dependent changes in the global conformation of RNA ligands were monitored by variations in FRET efficiency (E FRET ) between 3Ј-Fl and 5Ј-Cy3 fluorophores. For RNA substrate Cy3-R␤ 19 -ARE 15 -G-Fl, binding to His 6 -p37 AUF1 significantly decreased emission from the FRET donor (Fl) but only if the FRET acceptor (Cy3) was present (Fig. 5A, cf. emission at 520 nm in left versus right panels). Calculating E FRET from measurements of Cy3-R␤ 19 -ARE 15 -G-Fl donor emission across a titration of His 6 -p37 AUF1 shows that the 5Ј-and 3Ј-termini of the RNA ligand are forced closer together upon binding the protein (Fig. 5B, solid circles). Furthermore, the protein concentration dependence of RNA structural remodeling parallels the fractional concentration of RNA within RNP complexes (Fig. 5B, blue lines). Conversely, an RNA ligand lacking the 5Ј-R␤ domain (Cy3-ARE 15 -G-Fl) exhibited no change in the average distance between its 5Ј-and 3Ј-termini upon binding His 6 -p37 AUF1 (Fig. 5C). Together, these data indicate that local RNA remodeling by p37 AUF1 involves contacts with sequences upstream of the nucleating ARE 15 domain and that these conformational changes may be a major contributor to the change in heat capacity observed during AUF1 RNP formation on these RNA targets. These findings are consistent with previous observations of AUF1-induced conformational changes in extended ARE and polyuridylate ligands (12,20) but also indi-cate that condensation of local RNA structure is largely independent of the upstream sequence identity.
A Minimal p37 AUF1 -binding RNA Substrate Is a Functional AUF1-regulated mRNA-destabilizing Element in Cells-Our previous report demonstrated that high affinity p37 AUF1 binding required Ͼ34 nt of RNA (12). However, data from this study indicate that only a fraction of the AUF1-binding site on RNA requires AU-rich sequence and that a 15-16 nt ARE fragment can nucleate formation of stable (K d Ͻ 20 nM) complexes with His 6 -p37 AUF1 in vitro when positioned within a larger RNA context. Accordingly, we tested whether a reporter mRNA containing a minimal high affinity p37 AUF1 -binding site could be recognized and regulated by AUF1 in cells.
Stable HeLa and HeLa/Tet-Off cell lines were developed that expressed a non-targeting control shRNA (shControl) or one that targeted a domain common to all AUF1 mRNA variants (shAUF1). Western blots demonstrated robust suppression of all AUF1 protein isoforms in the shAUF1 lines (Fig. 6A). Transient transfection of a shRNA-resistant FLAG-tagged p37 AUF1 expression plasmid into shAUF1 clonal lines permitted selective expression of the p37 AUF1 isoform at near-or subphysiological levels (Fig. 6A, right lane), thus minimizing potential complications of protein overexpression. Into each of these cell backgrounds we also transfected reporter constructs under the control of Dox-responsive promoters, expressing either wild type ␤G mRNA (␤G-wt) or ␤G containing the R␤ 17 -ARE 16 -R␤ 17 sequence downstream of the translation termination codon (Fig. 6B, ␤G-RAR). An additional reporter mRNA included the 38-nt core ARE from TNF␣ mRNA (␤G-ARE 38 ) as a positive control, as a previous study indicated that this transcript bound AUF1 and was stabilized in shAUF1-expressing cells (26).
In RNP-IP experiments, inclusion of either the ARE 38 or RAR sequences significantly enhanced reporter mRNA recovery in immunoprecipitates containing p37 AUF1 -FLAG (Fig. 6C, p Ͻ 0.002 versus ␤G), indicating that both p37 AUF1 binding determinants can associate with the cellular protein. We next tested whether expression of AUF1 modulated reporter mRNA decay kinetics in HeLa/Tet-Off cell models using Dox time course assays, as the canonical role of AUF1, particularly the p37 AUF1 isoform, is to promote degradation of mRNA substrates (15,61). ␤G-wt mRNA, which lacks any ARE sequences, decayed very slowly in HeLa/Tet-Off cells expressing either shControl or shAUF1 (Fig. 6D, left panel), resolving mRNA half-lives of greater than 10 h in each cell background ( Table 4). As expected, insertion of the ARE 38 sequence into the 3Ј-UTR of ␤G mRNA shortened the mRNA half-life to 1 h in shControlexpressing cells (Fig. 6D, center panel), whereas suppression of endogenous AUF1 by shRNA stabilized ␤G-ARE 38 mRNA 2-fold. Inclusion of the RAR sequence in the 3Ј-UTR also significantly accelerated reporter mRNA decay but to a lesser extent than the ARE 38 sequence (Fig. 6D, right panel). However, decay of the ␤G-RAR mRNA was also regulated by AUF1, as this transcript was stabilized by ϳ60% in the shAUF1-expressing cell background (Table 4). Notably, expression of the shRNA-resistant p37 AUF1 -FLAG completely reversed shAUF1induced stabilization of both ␤G-ARE 38 and ␤G-RAR mRNAs. The results from these ectopic p37 AUF1 rescue experiments confirm that inhibition of reporter mRNA decay in shAUF1expressing cells was not an off-target effect of RNA silencing and that near-physiological levels of the p37 AUF1 isoform were sufficient to accelerate decay of each ARE-containing reporter mRNA. Together these data indicate that a minimal high affinity p37 AUF1 target sequence forms RNP complexes with this AUF1 isoform that accelerate mRNA turnover in a cellular context.

DISCUSSION
The nucleotide determinants required for AUF1 binding to RNA targets control the affinity and positioning of AUF1 recruitment and by extension the biochemical and functional consequences of these interactions. Our previous work using macromolecular binding density analysis and truncated ARE substrates demonstrated that for p37 AUF1 and p42 AUF1 , association of the first protein dimer occupies 33-34 nucleotides of RNA (12). Shorter RNA ligands showed significantly weaker binding, whereas above this site size a second AUF1 dimer binding event became measureable (12). Data from the current study are consistent with this unusually large RNA site size required to form high affinity p37 AUF1 RNP complexes but show that only a subset of the RNA ligand must be AU-rich. Although a short AU-rich domain is required to nucleate p37 AUF1 RNP assembly (Fig. 1, cf. binding to R␤ 31 versus R␤ 17 -ARE xx -R␤ 17 ligands), significant contributions to complex stability are made by contacts with 5Ј-flanking RNA, and even more if a purine residue is present immediately downstream of the ARE domain (Figs. 2 and 7A). Energetic contributions to RNP stability by nucleotides 5Ј of the core ARE domain include significant sequence-independent components, as additions of R␤ sequence (Fig. 2) yielded improvements in p37 AUF1 binding affinity comparable to those observed with similarly extended ARE sequences (12). Although base-independent stabilization of RNP complexes often involves ionic interactions between the protein and the RNA phosphodiester backbone, this is incon-  Table 4. sistent with the inability of 5Ј-R␤ sequences to alter the sensitivity of p37 AUF1 -RNA binding equilibria to changes in ionic strength ( Fig. 4 and Table 3). These data suggest that alternative interactions are formed between p37 AUF1 and these upstream nucleotides, which could include contacts with ribose moieties similar to those described for HuR (62). We were surprised to discover a strong enhancement to p37 AUF1 binding by the presence of a 3Ј-purine. Using guanosine as a model, the enthalpic benefit conferred by this residue coupled with the increased entropic penalty associated with p37 AUF1 binding suggests the formation of chemical contacts between this nucleotide and AUF1 (Table 3). Furthermore, the clear preference for a purine versus a pyrimidine base at this position indicates that such contacts likely involve direct base recognition. Further support for this model is given by the NMR structure of the C-terminal RRM of AUF1 bound to telomeric repeat d(TTAGGG) DNA, in which the RRM makes direct contacts with the central TAG sequence (63). In this structure the G nucleobase at position 4 is hydrogen-bonded to the carbonyl oxygen of Met-239 via N 1 and the ⑀-amino group of Lys-164 via O 6 while also forming a stacking interaction with Phe-208 (amino acid numbering follows p37 AUF1 sequence) (Fig. 7B). An adenosine at this position would be similarly expected to stack well with Phe-208 and form the hydrogen bond between N 1 and the Met-239 carbonyl oxygen, although it would not form the second predicted hydrogen bond with Lys-164 because it lacks the O 6 acceptor. By contrast, pyrimidines would be unlikely to reach as deeply into the binding cleft to make the hydrogen bonds or stack with Phe-208. Also, a single purine 3Ј of the U trinucleotide repeat of the core ARE was not sufficient to direct this enhancement of RNP stability (Fig. 2, cf. Fl-ARE 16 versus Fl-ARE 15 -A), consistent with interactions observed between AUF1 RRM2 and each of the consecutive T, A, and G bases in the NMR structure (63). However, given that p37 AUF1 can also bind with low nM affinity to extended (Ն34 nt) ARE domains that lack these additional 3Ј-purine residues (12), we suggest that the purine base is not necessary for association with all ARE-containing transcripts but could enhance binding on otherwise suboptimal targets. Furthermore, the presence of an additional purine 3Ј of the core AU-rich domain may pref-erentially direct AUF1 to a specific binding register on an ARE, whose reiterative nature might otherwise present a quasi-homogeneous lattice of potential AUF1 binding sites.
His 6 -p37 AUF1 binding modified the conformation of the 35-nt Cy3-R␤ 19 -ARE 15 -G-Fl substrate by bringing the RNA 5Јand 3Ј-termini into closer proximity, an effect that was not observed with the Cy3-ARE 15 -G-Fl RNA ligand (Fig. 5). Inclusion of the R␤ 19 domain also correlated with the significantly increased negative value of ⌬C°P ,obs associated with p37 AUF1 RNP formation ( Table 2), suggesting that protein-induced alterations in RNA confirmation may contribute to the R␤ 19dependent change in heat capacity. If so, it is possible that fewer nucleotides of sequence upstream of the ARE 15 core domain may be sufficient to induce these RNA conformational changes, as increased negative values of ⌬C°P ,obs were also observed for RNA substrates R␤ 8 -ARE 15 -G and R␤ 8 -ARE 15 versus ARE 15 -G upon binding to His 6 -p37 AUF1 ( Table 2). Based on Equation 15, the average distance between the 5Ј-and 3Ј-termini of the Cy3-R␤ 19 -ARE 15 -G-Fl substrate is ϳ48 Å in the p37 AUF1 RNP complex, in contrast to 60 Å between the ends of the unbound RNA ligand. These distances are similar to those observed for ARE 38based RNA substrates (12,33), suggesting that both extended ARE and chimeric R␤-ARE ligands assume comparable conformations in AUF1 RNPs. Formation of such condensed RNA structures would likely involve protein contacts at or near both ends of the RNA ligand that bring its termini into close proximity and restrict RNA mobility. The additional protein-RNA contacts required by this model would also be consistent with increased enthalpic contributions to complex stability (⌬H°Ͻ 0) while decreasing system entropy (⌬S°Ͻ 0) ( Table 2). Similar conformational rearrangements of bound single-stranded nucleic acid ligands have also been reported for the tandem RRM domain of hnRNP A1 and the RRM3ϩRRM4 domain of polypyrimidine tract-binding protein (PTB) (64,65). For PTB in particular, association of individual RRMs with separate 6-nt polypyrimidine sequences permits the intervening nucleotides to be looped, potentially allowing distal sequence determinants to be co-localized in the RNP complex (65). However, telomeric DNA sequences binding to the tandem RRMs of hnRNP A1 make contacts with only 11 contiguous nucleotides (23), whereas a 27-nt RNA ligand (two 6-nt polypyrimidine sequences separated by a 15-nt spacer) is sufficient to bind the PTB RRM3ϩRRM4 domain with low nM affinity (65). By contrast, optimal binding of p37 AUF1 requires at least 34 nt of RNA (Ref. 12 and this study). It is possible that the relative positioning of the two RRM domains in p37 AUF1 require access to disparate RNA subdomains analogous to the PTB model, although involving greater spatial separation. However, the stepwise improvement in His 6 -p37 AUF1 affinity observed as a function of total RNA size (Fig. 2) suggests that RNA interactions involving AUF1 domains outside the RRMs or with RRM domains from both proteins in the dimeric AUF1 complex are more likely explanations on extended RNA substrates.
At present, Ͼ20 different ARE-binding proteins have been identified that confer diverse effects on the stability and translational efficiency of targeted mRNAs (2,3). Given the variety of these factors expressed in cells, it is likely that many ARE-containing mRNA substrates are regulated by different proteins FIGURE 7. RNA contacts required for high affinity p37 AUF1 binding. A, schematic showing RNA domains contributing to stability of p37 AUF1 RNP complexes on minimal ARE substrates. Red arrows point to RNA sequences making base-specific contacts, whereas yellow arrows denote nonspecific contacts 5Ј-of the nucleating ARE sequence that are also required for AUF1induced remodeling of local RNA structure. B, the structure of the AUF1 RRM2 domain bound to a telomeric DNA repeat fragment (PDB code 1X0F from Ref. 63) shows potential protein interactions with the first G nucleotide (G4). Implications for AUF1 binding to minimal ARE ligands are discussed under "Discussion." through competitive and/or combinatorial mechanisms, a model supported by common mRNA substrates identified in ribonome-wide surveys of mRNAs targeted by distinct AREbinding proteins (50,66,67). However, although each AREbinding protein is recruited to AREs or similar sequences, biochemical analyses have indicated protein-dependent differences in ARE binding affinity, RNA sequence preferences, and sensitivity to local RNA structure (12,21,22,53,68). The current study provides a biochemical rationale for AUF1 binding to a broad range of ARE-like RNA targets beyond extended U-rich domains. Most notably, a 15-16-nt AU-rich sequence contained within a larger RNA context is sufficient to nucleate high affinity p37 AUF1 RNP complexes (Figs. 1 and 2) that can target substrate mRNAs for degradation in cells (Fig. 6). We predict that smaller ARE targets may focus trans-factor selectivity and minimize the potential for combinatorial binding mechanisms by limiting access to ARE sequence determinants beyond a single RNA-binding event. This model may explain the slower decay kinetics of the ␤G-RAR versus ␤G-ARE 38 mRNAs in both shControl-and shAUF1-expressing HeLa cells (Table 4). Although p37 AUF1 bound both mRNAs in cells and significantly accelerated their decay kinetics, it is likely that some other cellular ARE binding factors were less efficiently recruited to the RAR reporter transcript and hence did not contribute to its turnover.
Finally, this study demonstrates that interactions between p37 AUF1 and RNA sequences 5Ј of a nucleating ARE site mediate AUF1-induced changes in local RNA structure. Although the functional significance of this activity is currently unknown, we predict that these conformational changes may enhance or obstruct access for other RNA binding factors or microRNAs to nearby binding sites on mRNA targets. These factors could include members of a multisubunit trans-acting complex that may mediate AUF1-dependent effects on mRNA decay such as the heat shock proteins Hsp27 and Hsp/Hsc70, the translation initiation factor eIF4G, and poly(A)-binding protein (30,69,70). Most of these factors also possess RNA binding activity (30,51,71,72), so local RNA remodeling by AUF1 could influence their recruitment by exposing adjacent sequence determinants. Alternatively, AUF1 binding may alter the accessibility of proximal RNA binding sites to other ARE-binding proteins or microRNAs. Future studies will determine which of these factors are impacted by AUF1-induced RNA remodeling and the functional significance of these RNA allosteric relationships.