Destabilization of Interleukin-6 mRNA Requires a Putative RNA Stem-Loop Structure, an AU-Rich Element, and the RNA-Binding Protein AUF1

Abstract

Interleukin-6 mRNA is unstable and degraded with a half-life of 30 min. Instability determinants can entirely be attributed to the 3′ untranslated region. By grafting segments of this region to stable green fluorescent protein mRNA and subsequent scanning mutagenesis, we have identified two conserved elements, which together account for most of the instability. The first corresponds to a short noncanonical AU-rich element. The other, 80 nucleotides further 5′, comprises a sequence predicted to form a stem-loop structure. Neither element alone was sufficient to confer full instability, suggesting that they might cooperate. Overexpression of myc-tagged AUF1 p37 and p42 isoforms as well as suppression of endogenous AUF1 by RNA interference stabilized interleukin-6 mRNA. Both effects required the AU-rich instability element. Similarly, the proteasome inhibitor MG132 stabilized interleukin-6 mRNA probably through an increase of AUF1 levels. The mRNA coimmunoprecipitated specifically with myc-tagged AUF1 p37 and p42 in cell extracts but only when the AU-rich instability element was present. These results indicate that AUF1 binds to the AU-rich element in vivo and promotes IL-6 mRNA degradation.

Interleukin-6 (IL-6) is a pleiotropic cytokine with multiple functions. Discovered as a B-cell growth and differentiation factor, it was subsequently implicated in the regulation of hematopoiesis and liver regeneration and the acute inflammatory response in liver (10, 58). IL-6 is secreted by many cell types and transcriptionally induced by external stimuli, notably the inflammatory cytokines IL-1 and tumor necrosis factor alpha (TNF-α) (27). Its expression must be carefully controlled, as high IL-6 levels are associated with various autoimmune diseases and tumor growth (27, 60). IL-6 overexpression in mice causes polyclonal plasmacytosis, which may evolve to malignant monoclonal plasmacytoma (25, 57). High IL-6 expression is also associated with other advanced tumors (60). Notably, in multiple myeloma, there exists evidence for autocrine growth due to IL-6 overexpression (28). Both transcriptional and posttranscriptional regulation appears to be involved, as IL-6 mRNA stability is modulated by external stimuli that activate the MAP kinase and other signaling pathways (12, 42, 44, 65). IL-6 mRNA shares with mRNAs of several cytokines (IL-2, IL-3, IL-8, and TNF-α), growth factors (granulocyte colony-stimulating factor [G-CSF], granulocyte-macrophage CSF [GM-CSF], and vascular endothelial growth factor), and growth-related transcription factors (c-myc, c-fos, and c-jun) the feature of being very unstable, thus keeping protein levels low. The 3′ untranslated region (3′UTR) of IL-6 mRNA is sufficient to confer a short half-life to reporter RNA (3, 44, 56). Human and mouse IL-6 3′UTRs comprise six conserved AUUUA sequences that resemble AU-rich elements (AREs) and are assumed to be important for mRNA instability. However, no detailed analysis has been carried out.

AREs are among the best-characterized mRNA-destabilizing determinants. Their deletion typically provokes increased mRNA levels and, sometimes, as in c-fos mRNA, oncogenic cell transformation (41). In certain cases, AREs are sufficient to confer instability when grafted into the 3′UTR of an otherwise stable mRNA (50, 59). Initially described as AUUUA repeats, they were later defined as tandem repeats of an UUAUUUA(U/A)(U/A) sequence (31, 71). However, microarray studies found that about 5% of all mRNAs in HepG2 and primary fibroblast cell lines are unstable, with half-lives of less than 2 h (69), and that many of them have 3′UTR sequences that do not conform to classical AREs (47, 69). Moreover, the prediction of AUUUA repeats in 3′UTRs correlated in only 15 to 30% of cases with a short mRNA half-life (34, 47, 69). This suggests that many individual mRNAs comprise either unpredicted ARE variants or other unknown destabilizing sequences. Even classical AREs are usually imbedded in neighboring sequence elements that are conserved in evolution but differ for each gene (11), and in several cases, such as G-CSF (46), TNF-α (53), and endothelin-1 (40), adjacent regions contribute to rapid mRNA degradation.

The precise mechanism by which AREs or other destabilizing elements induce rapid mRNA degradation is not fully understood. AREs induce a rapid shortening of the poly(A) tail, which precedes 3′-5′ degradation and is thought to be a prerequisite for decay (20, 63, 67). Recently, however, other mechanisms, including 5′-3′ degradation (14, 54) or endonucleolytic cleavage (29), have been postulated. Several proteins interact with AREs and promote or impede mRNA degradation. Best documented among degradation-promoting proteins are KSRP (15), the tristetraprolin (TTP) family members (6, 32, 52), and AUF1 (also known as hnRNP D) (70). AUF1 exists in four alternative splice variants of exons 2 and 7 (61). p45 has both the 19 amino acids of exon 2 and 49 amino acids of exon 7, whereas p37 has neither sequence. p40 and p42 comprise the exon 2 and exon 7 sequences, respectively. Although AUF1 is mostly nuclear, it shuttles to the cytoplasm, possibly attached to AREs (9). Evidence for destabilizing or stabilizing effects of AUF1 is contradictory and may vary depending on the cell line or isoforms analyzed. Ectopically expressed AUF1 p37 and p42 increased the instability of an ARE-containing mRNA in a hemin-treated differentiating human erythroleukemia cell line, K562 (38). In contrast, overexpression of each of the four myc-tagged AUF1 isoforms in NIH 3T3 cells stabilized β-globin constructs with either GM-CSF or c-fos AREs (68). Another study using similar conditions of transiently overexpressed flag-tagged AUF1 concluded, however, that p37 was limiting in cells when an ARE-mRNA reporter construct was at saturating levels for the decay machinery. Under such conditions, overexpression of p37 could overcome the saturation and accelerate ARE mRNA decay in the NIH 3T3, HeLa, 293T, and COS-1 cell lines, suggesting a destabilizing function (49). Recently, RNA interference (RNAi) was used to define the role of AUF1 in rapid mRNA decay. Targeting exon 2 with small interfering RNA (siRNA), which reduced the levels of p40 and p45, increased the half-life of a green fluorescent protein (GFP)-GM-CSF construct, but no such effect was seen after targeting all four isoforms or only p42 and p45 (48). The authors concluded that the relative ratio of isoforms might be important and that p40 had a destabilizing function in HT1080 cells. Others, however, found that siRNA targeting of all isoforms increased the half-lives of p21 and cyclin D1 mRNA (33). Whether AUF1 plays a role in IL-6 mRNA stability has not been investigated.

In view of the probable importance of posttranscriptional regulation in IL-6 mRNA expression, we have investigated which cis-acting elements in the human IL-6 3′UTR are essential for conferring instability on a GFP reporter mRNA. As expected, ARE sequences stand out as important, but they do not fully account for the overall instability. We have newly identified a 3′UTR sequence with a potential stem-loop structure that is equally required. We have further tested whether AUF1 plays a role in IL-6 mRNA stability by lowering or increasing its level. These assays were performed in stably transfected NIH 3T3 cells using newly designed retroviral vectors for RNAi and inducible expression of AUF1 isoforms. In parallel, we have carried out coimmunoprecipitation (IP) of myc-tagged AUF1 with the mRNA. We conclude that AUF1 p37 and p42 bind to the destabilizing ARE sequences in vivo and promote mRNA decay solely under conditions of adequate protein expression.

MATERIALS AND METHODS

Cell culture.

The simian virus 40 (SV40) large-T-antigen-transformed African green monkey cell line COS-7 (16) was grown in Dulbecco's modified Eagle's medium (Invitrogen, Carlsbad, CA), 10% fetal calf serum. A thymidine kinase-negative COS-7 cell clone was isolated after mutagenesis by three successive overnight treatments with 0.55 μl/ml methanesulfonic acid ethyl ester and subsequent selection with 30 μg/ml 5-bromo-deoxyuridine (Sigma, St. Louis, MO). This cell clone was used for stable plasmid cotransfections with the herpes simplex virus thymidine kinase gene by the calcium phosphate method (19) and selection in 100 μM hypoxanthine, 0.4 μM aminopterin, and 16 μM thymidine (HAT medium; Invitrogen). Mouse NIH 3T3 and Phoenix ecotropic virus packaging cells were cultured in Dulbecco's modified Eagle's medium, 10% fetal calf serum. Retrovirus was produced by stably transfecting plasmids into Phoenix cells by the calcium phosphate method (19). Two days later, culture supernatant was harvested and filtered through 0.45-μm filters (Millipore, Billerica, MA), supplemented with 8 μg/ml Polybrene, and used to infect exponentially growing NIH 3T3 or COS-7 cells. Infected cells were selected 2 days later with 10 μg/ml puromycin or 6 to 10 μg/ml blasticidin. Whole-cell populations were used for mRNA stability measurements. MG132 (C-2211; Sigma) was added at 40 μM from a 100 mM stock solution in dimethyl sulfoxide.

Plasmids and retroviral vectors for GFP fusion constructs.

For expression of GFP, an EcoRV-BamHI fragment of pcD-TR1-Δ322 (45) was replaced by the NheI-BamHI fragment of pEGFP-C1 (Clontech, Mountain View, CA). All three reading frames were blocked at the end of the GFP cDNA by insertion of the sequence CGTTAATTAATTAACG into the BspEI site after fill-in. This vector is referred to as pcDTR-GFPst or the GFP control vector. The wild-type human IL-6 3′UTR was amplified by PCR from expressed sequence tag (EST) AI085963 using as the forward primer CGCGGATCCGGTACCTAGCATGGGCACCACCTCAG and the reverse primer CCGGAATTCTTAAAATGCCATTTATTGGTAT. This fragment was cut with BamHI and EcoRI and cloned into the BglII-EcoRI sites behind the stop codon of pcDTR-GFPst. The plasmid was named pcDTR-GFPst-IL-6 and transcribes into GFP-IL-6 mRNA.

Deletion clones of the IL-6 3′UTR were PCR amplified with appropriate primers and subcloned by the same procedure. We adopted the nucleotide numbering for human IL-6 mRNA of GenBank entry M54894. The wild-type and mutant constructs extended over the following nucleotide positions: positions 687 to 1106 (wild type), 687 to 1008 (Δ1), 687 to 910 (Δ2), 687 to 810 (Δ3), 793 to 1106 (Δ4), 888 to 1106 (Δ5), 991 to 1106 (Δ6), 793 to 1008 (Δ7), 793 to 910 (Δ8), and 888 to 1008 (Δ9).

Linker scanning mutagenesis of pcDTR-GFPst-IL-6_Δ2 was carried out such that a linker with the sequence CGCAGATCTACAATTGGGA replaced the 3′UTR sequence every 15 bp. This was achieved by PCR amplification with appropriate primers. Fragments to the left or the right of the linker were first amplified separately. The reverse primer of fragment 1 and forward primer of fragment 2 were designed such that they could hybridize to create the linker. The separately amplified fragments were then mixed and reamplified with terminal primers, recreating a 224-base 3′UTR fragment that was cloned into the pcDTR-GFPst vector. Point mutations in the IL-6_Δ2 construct were introduced by PCR with appropriate primers, as for the scanning mutants.

For measurements by the Tet-Off system in NIH 3T3 and COS-7 cells, two retroviral vectors based on pBabe were devised. The details of these constructs will be reported elsewhere (B. Grisoni-Neupert et al., unpublished data). In short, the first vector, pZPCTHG, comprised the 7× Tet operator (17) next to a heterologous minimal promoter (26) and a cloning site for insertion of GFP-3′UTR constructs. The cloning site was followed by a polyadenylation signal of bovine growth hormone from pCDNA3 (Invitrogen). The same 7× Tet operator was used in the opposite orientation to drive the transcription of the puromycin resistance gene behind the cytomegalovirus (CMV) minimal promoter. The second vector, pBHTTA, comprised the tetracycline-sensitive trans-activator protein (17) under the control of the long terminal repeat promoter. The GFP-IL-6 fusion constructs were inserted into pZPCTHG as AgeI-BamHI fragments from pcDTR-GFPst-IL-6 or from mutants thereof. The human GM-CSF 3′UTR was amplified from EST AI655452 with forward (CGCGGATCCGGTACCGAGACCGGCCAGATGAG) and reverse CCGGAATTCAGAAGCATATTTTTAATAATAATT) primers. It was subcloned as an EcoRI-BamHI fragment into pcDTR-GFPst and from there into pZPCTHG. c-myc sequences were amplified from plasmid pACmyc11 (gift from Andreas Trumpp, ISREC) containing the mouse c-myc cDNA. The c-myc coding sequence with its 3′UTR was PCR amplified with the forward primer CCGGAATTCTGTACATGCCCCTCAACGTGAAC and reverse primer CCGGAATTCGCGGCCGCCTGTTATAAACGTTTTATTAAAG and cloned into the BsrGI and EcoRI sites of pEGFP-C1 (Clontech), resulting in the loss of the GFP stop codon. The GFP-myc fusion was subsequently cloned into the AgeI and NotI sites of pZPCTHG.

RNAi against mouse AUF1 in NIH 3T3 cells.

The retroviral expression vector pRetro Super carrying the human histone H1 polymerase III promoter was obtained from T. R. Brummelkamp (4). pSMLH1 was constructed by replacement of the puromycin resistance gene by the blasticidin resistance gene. Three different oligonucleotides overlapping exon 4 or 5 of the AUF1 coding region were inserted into pSMLH1 such that they formed RNA hairpin structures (in boldface in the sequences below) once transcribed from the H1 promoter. Oligonucleotide 299 had the sequence GATCCCCTAAGAGAGTACTTTGGTGGTTCAAGAGACCACCAAAGTACTACTCTCTTATTTTTGGAAA and gave rise to a processed RNAi sequence corresponding to coding region positions 593 to 613 in exon 4 of mouse AUF1. Oligonucleotide 300 had the sequence GATCCCCGACCAATAAGAGGCGTGGGTTCAAGAGACCCACGCCTCTTATTGGTCTTTTTGGAAA, corresponding to the mouse AUF1 exon 5 sequence from positions 654 to 674, and oligonucleotide 301 had the sequence GATCCCCTAAGAGGCGTGGGTTCTGTTTCAAGAGAACAGAACCCACGCCTCTTATTTTTGGAAA, overlapping the mouse AUF1 exon 5 sequence from positions 660 to 680. The oligonucleotides were annealed with the appropriate opposite strand and ligated into pSMLH1 at HindIII and BglII sites. The control vector was without an insert.

mRNA isolation and half-life measurements.

Total or cytoplasmic RNA from cell cultures at 70% confluence and RNA from immunoprecipitation assays were isolated with the RNeasy mini kit (QIAGEN, Hilden, Germany). For RNA half-life measurements, actinomycin D (Sigma) was added at 6 μg/ml at 0, 30, 60, 120, and 180 min prior to RNA extraction. For Northern blots, 10 μg cytoplasmic RNA was separated in 1.2% agarose-2% formaldehyde gels. RNA was transferred to Immobilon Ny⁺ membranes (Millipore) and cross-linked by UV light (1.2 × 10⁵ μJ) in a Stratalinker (Stratagene, La Jolla, CA). Hybridizations were carried out with cDNA fragments of the complete coding region of GFP or human GAPDH (glyceraldehyde-3-phosphate dehydrogenase). The radioactive probe was synthesized by random priming using 50 to 100 ng template DNA in the presence of 30 μCi [α-³²P]dCTP (3,000 Ci/mmol). The incorporation of label was at least 20%. Filters were prehybridized for 3 h at 42°C in hybridization buffer (50% formamide, 1% sodium dodecyl sulfate [SDS], 3.4× SSC buffer [1× SSC is 0.15 M NaCl plus 0.015 M sodium citrate], 10% dextran sulfate) with 200 μg/ml sonicated salmon sperm DNA. For hybridization, the probe was heat denatured, added to hybridization buffer, and incubated with the membranes for 17 h at 42°C. Membranes were washed twice for 30 min at 65°C in 0.2× SSC, 0.1% SDS. For autoradiography, Biomax films (Kodak) were exposed. Quantification was done with a model BAS-MP 2040S imaging plate (Fujifilm, Tokyo, Japan). For sequential hybridizations, membranes were stripped by boiling them twice for 5 min in 0.1× SSC, 0.5% SDS.

Real-time PCR measurements were carried out with the Applied Biosystems GeneAmp 5700 in conjunction with the Eurogentec quantitative PCR core kit or the Roche LightCycler in conjunction with either Roche FastStart DNA Master^PLUS hybridization probes or master SYBR green I according to the manufacturer's recommendations (Roche Diagnostics, Rotkreuz, Switzerland). The primer set for GFP was ACTACAACAGCCACAACGTCTATATCAT (forward) and ATGTTGTGGCGGATCTTGAAG (reverse), and the probe was 6-carboxyfluorescein (FAM)-CCTTGATGCCGTTCTTCTGCTTGTCG-6-carboxytetramethylrhodamine (TAMRA). The primer set for mouse acidic ribosomal protein P0 (mARP0) was CTTTGGGCATCACCACGAA (forward) and GCTGGCTCCCACCTTGTCT (reverse), and the probe was FAM-ATCAGCTGCACATCACTCAGAATTTCAATGGT-TAMRA. The primate β-actin SYBR green primer set was CGCGAGAAGATGACCCAGAT (forward) and GCGAGAAGATGACCCAGATCA (reverse).

Mifepristone-inducible protein expression system.

For inducible protein expression, we constructed two retroviral vectors based on the commercially available GeneSwitch system (Invitrogen). The first vector, pBSwitch, permits the constitutive expression of the trans-acting mifepristone-inducible Switch protein (5). To construct pBSwitch, we modified pBabeBleo (43) by removing the SV40 promoter and bleomycin resistance gene and inserting the sequence encoding the Switch protein of pSwitch (Invitrogen) behind the viral LTR promoter.

For the construction of mouse AUF1 isoforms with an amino-terminal myc tag, the vector pSBCMYC was derived from pZPCTHG. First, the CMV promoter from pCIneo (Promega, Madison, WI) was PCR amplified with the forward primer GCCCCGTCGACTCAATATTGGCCATTAGCCATA and a reverse primer containing the myc tag sequence (underlined), CCCCGCGGCCGCACGTGGATCCCGTACGGAATTCCAGATCCTCTTCAGAGATGAGTTTCTGCTCCATGGTGGTTCGAAGCTTCTCGAGCCCTATAGTGAGTCGTATTAAGT, and digested with NotI and SalI. Second, the SV40 promoter/bleomycin resistance gene cassette was isolated from pBabeBleo by digestion with SalI and ClaI. Third, the two above-named fragments were ligated into pZPCTHG after digestion with ClaI and NotI, which removes the puromycin resistance and GFP genes. The AUF1 p45 cDNA was amplified by PCR of an EST from the NIH using the forward primer GGGTACCATGTCGGAGGAGCAGTTC and the reverse primer GGGGATATCTTAGTATGGTTTGTAGCTATTT. The product was digested with Asp718I and EcoRV and cloned into pSBCMYC digested with BbuPI and BswiWI. This places p45 in frame with the myc tag.

This myc-tagged AUF1 p45 cDNA was subcloned into a Switch-regulated expression vector, pSLHGCG. pSLHGCG was derived from pZPCTHG by replacing the puromycin resistance coding sequence with the blasticidin resistance gene from pcDNA6-E (Invitrogen). The bidirectional Tet operator was replaced by another bidirectional promoter composed of the heterologous minimal promoter Gal4/Adeno/CMV/intron from pGene/V5-His A (Invitrogen) and the minimal promoter comprising the TATA box of the adenovirus major late promoter and the initiator of the terminal deoxynucleotidyl transferase from 5× GTTI-Gal4-VV (26). pSLHGCG was digested with HindIII and SphI. A fragment containing the myc-tagged AUF1 p45 cDNA and bovine growth hormone poly(A) signal was isolated from pSBCMYC-AUF1 p45 with HindIII and SphI and ligated to the above-named vector fragment. This produced pSLHGC-MYC AUF1 p45.

To obtain pSLHGC-MYC AUF1 p40, a deletion of exon 7 was made in p45 by PCR-mediated mutagenesis. First, exons 1 to 6 of pSLHGC-MYC AUF1 p45 were amplified with the forward primer GGGAATTCCGTACCATGTCGGAGGAGCAGTTCG and reverse primer CCATAACCACTCTGCTGATCTCCACCTCTTCCGCGAGCT. Second, exons 8 and 9 were amplified with the forward primer CGCGGAAGAGGTGGAGATCAGCAGAGTGGTTATGGGAAAGTATCCAGGC and reverse primer GAATAGGGCGGCCGC. Third, the above-described products were ligated due to complementary regions (in bold) using PCR with the forward primer at exon 1 and reverse primer at exon 9. The final PCR product was digested with BglII and NotI and subcloned into the corresponding region of pSLHGC-MYC AUF1 p45. pSLHGC-MYC AUF1 p42 was constructed from pSLHGC-MYC AUF1 p45 by replacing the region encompassing exon 2 with the corresponding region of EST AA154631, in which exon 2 is absent. pSLHGC-MYC AUF1 p45 was digested with EcoRI and BglII to remove the exon 2 region and ligated to an EcoRI/NcoI fragment from the same vector and the corresponding NcoI/BglII fragment from EST AA154631. In order to construct pSLHGC-MYC AUF1 p37, a deletion of both exons 2 and 7 was obtained by ligation of the exon 7 deletion PCR fragment used to obtain p40 and the exon 2 deletion fragment used to obtain p42. All expression constructs of AUF1 isoforms were sequenced.

Protein extracts, SDS-polyacrylamide gel electrophoresis, and Western blotting.

For total protein extracts, cells were lysed in CelLytic-M cell lysis reagent (C2978; Sigma). Protein concentration was determined with a protein assay reagent of Bio-Rad (Hercules, CA). SDS-polyacrylamide gel electrophoresis was performed with 12.5% polyacrylamide gels. For Western blots, proteins were transferred onto nitrocellulose ECL membranes (Amersham, Buckinghamshire, United Kingdom) with the Bio-Rad system. Membranes were preincubated in 1× TEN (20 mM Tris-HCl [pH 8], 1 mM EDTA, 140 mM NaCl), 5% milk powder three times for 15 min each time. Membranes were incubated with the first antibody for 2 h at room temperature or overnight at 4°C. Dilutions used were 1:1,000 for mouse monoclonal anti-myc tag antibody 9E10 (Roche Diagnostics) and 1:5,000 for monoclonal anti-γ-tubulin immunoglobulin G (IgG) (T-6557; Sigma). Membranes were subsequently washed three times for 20 min each time in 1× TEN, 5% milk powder. Except for the anti-myc tag antibody, which was directly linked to peroxidase, membranes were incubated with a horseradish peroxidase-coupled goat anti-mouse IgG antibody (A-9044; Sigma) at a 1:20,000 dilution for 1 h at room temperature. Following two 10-min washes in 1× TEN, 5% milk powder, membranes were further washed twice for 10 min each time with 1× TEN, 0.1% Triton X-100, 1.4 M NaCl. Endogenous AUF1 was measured with polyclonal rabbit anti-mouse AUF1 antiserum (07-260; Upstate Biotechnology, Charlottesville, VA) and horseradish peroxidase-coupled goat anti-rabbit IgG (W4011; Promega). Detection was performed with the Amersham ECL kit, and membranes were exposed for different times to a Fuji X-ray film.

Coimmunoprecipitation of mRNA.

To perform IP of mRNA with AUF1, 0.5 × 10⁶ to 1 × 10⁶ NIH 3T3 cells stably expressing myc-tagged AUF1 isoforms and GFP mRNA constructs were lysed in 400 μl CelLytic-M cell lysis reagent (C-2978; Sigma). Extracts were divided into two portions of 200 μl each. Of the first 200 μl (before IP), one-eighth (25 μl) was used for Western blot analysis of myc-AUF1 and the rest for RNA preparation (RNA-total). The second 200 μl was incubated at 4°C for 4 h with anti-c-myc agarose beads (A-7470; Sigma) according to the manufacturer's protocol. After a 15-s centrifugation at 10,000 × g, the 200-μl supernatant was collected, and again one-eighth (25 μl) was used for Western blots and the rest for RNA preparation (RNA-sup). The pellet of beads (IP) was washed twice with 1× IP buffer (I-5779; Sigma) and resuspended in 100 μl of 0.1× IP buffer. One-eighth of the pellet was used for Western blot analysis, and the rest of the sample was used for RNA preparation (RNA-IP). The Western blot analysis served as a control to show that all myc-tagged AUF1 was precipitated. Real-time PCR quantification of GFP and mARP0 mRNAs was performed with the LightCycler (Roche). The recovery of mRNA in the procedure was determined with the formula (RNA-IP + RNA-sup)/RNA-total and was generally between 30 and 80%. Experiments with lower rates of recovery were not considered. The efficiency of the IP of any mRNA was calculated as 100% × RNA-IP/(RNA-IP + RNA-sup).

RESULTS

Identification of IL-6 mRNA-destabilizing regions.

In order to identify cis-acting elements that destabilize human IL-6 mRNA, we inserted its 420-nucleotide 3′UTR into a mammalian expression vector immediately behind the stop codon of GFP. Transcription was initiated by the human transferrin receptor promoter and terminated by a polyadenylation signal of SV40 origin. The GFP vectors, alone or with the IL-6 3′UTR, were stably transfected into COS-7 cells. The GFP mRNA half-life was assessed after addition of the transcription inhibitor actinomycin D. The control vector gave rise to stable GFP mRNA with a half-life of more than 8 h (Fig. 1). In the presence of the IL-6 3′UTR, the GFP mRNA half-life diminished to 33 ± 5 min (Fig. 1). This confirms that the IL-6 3′UTR harbors the cis elements necessary to destabilize the reporter GFP mRNA. We also conclude that actinomycin D does not interfere with rapid IL-6 mRNA degradation.

FIG. 1.

The IL-6 3′UTR confers mRNA instability on a stable GFP mRNA. GFP mRNA with or without the 3′UTR of human IL-6 mRNA was stably expressed in COS-7 cells. Transcription was blocked by actinomycin D (Act D). (A) GFP mRNA was measured at different time points on Northern blots by hybridization with a GFP-specific probe and normalized to endogenous GAPDH mRNA. (B) The mRNA decay was plotted on a semilogarithmic scale and the half-life calculated by linear regression. Error bars indicate standard deviations (SD) from at least three independent experiments.

The extensive sequence conservation of the IL-6 3′UTR among mammalian species makes it impossible to predict regions that confer mRNA instability. Therefore, we constructed GFP constructs with shorter IL-6 3′UTR fragments to test them in stably transfected COS-7 cells (Fig. 2). mRNA half-life measurements indicated that destabilizing elements were contained mainly in the first 224 nucleotides of the 3′UTR (mutant Δ2) (Fig. 2), between positions 687 and 910 (numbering is as in GenBank entry M54894). However, a further subdivision of this region revealed a marked loss in its destabilizing capacity (compare mutants Δ3 and Δ8 with mutant Δ2). It suggested either more than one determinant in the 224-nucleotide 3′UTR of mutant Δ2 or the disruption of an essential element at the boundary of mutants Δ3 and Δ8. The result was unexpected, as all four ARE-like sequences of the GFP-IL-6_Δ2 cluster between nucleotides 825 and 885 and were present in the Δ8 construct. Yet, the finding was confirmed with mutant Δ4, which comprised all six AREs but was only very partially destabilized. Consequently, nucleotides 687 to 825 must contain an element that contributes to rapid mRNA decay and does not belong to the AU-rich class.

FIG. 2.

Essential regions of the IL-6 3′UTR that induce rapid mRNA degradation. (A) The human wild-type IL-6 3′UTR behind the GFP cDNA was shortened by various deletions, and the effect on mRNA stability was assessed in stably transfected COS-7 cells. Cytoplasmic mRNA was isolated at different time points after actinomycin D (Act D) treatment and probed for GFP and GAPDH mRNA by Northern hybridization as described for Fig. 1. The graphic also indicates the positions of AUUUA sequences with ARE-like properties in human IL-6 mRNA. T_1/2, half-life. (B) mRNA half-lives were calculated from typical decay curves by linear regression between 0 and 120 min. Values ± SD are based on at least two independent experiments.

The IL-6 mRNA-destabilizing region comprises two distinct elements.

In order to identify the essential sequences, we avoided shortening of the 3′UTR and carried out linker-scanning mutagenesis of the proximal 224 nucleotides of mutant Δ2. This was achieved by PCR-mediated mutagenesis replacing every 15 nucleotides by a constant linker sequence. The resulting 15 mutants, A to O, were tested in stably transfected COS-7 cells by measuring their mRNA half-lives by Northern blot hybridization after the addition of actinomycin D (Fig. 3). Compared to the wild-type construct or to mutants with low half-lives (C, D, I, J, or N), several mutants showed increased half-lives. Notably, mRNA of mutants F, G, H, and L was significantly more stable than wild-type GFP-IL-6 mRNA. We conclude that the 3′UTR of human IL-6 mRNA has at least two separate destabilizing regions, only one of which, region L, comprises an ARE. Both regions need to be present to obtain full destabilization.

FIG. 3.

Linker scanning mutagenesis of the destabilizing 3′UTR region reveals two important elements. The 224-base 3′UTR of GFP-IL-6_Δ2 (Fig. 2) was mutagenized by a linker sequence replacing the natural sequence every 15 bases. Each mutant was stably expressed in COS-7 cells, and the GFP mRNA half-life was assessed by Northern blot hybridization after 30, 60, and 120 min of actinomycin D treatment. The results were normalized to β-actin mRNA expression and are the averages of results from two to four independent experiments. Half-lives were compared to that of GFP-IL-6_wt by the two-tailed Student t test, assuming unequal variances. *, P < 0.05.

To verify the existence of two distinct 3′UTR elements, we also carried out measurements of the half-lives of mRNA in mouse NIH 3T3 fibroblasts by taking advantage of a retroviral Tet-Off vector system. This system consists of a constitutive tetracycline-sensitive trans activator that is stably expressed and that activates the transcription of GFP constructs from a heterologous Tet operator in a second retroviral vector (17). Upon the addition of doxycycline, a tetracycline analogue, the trans activator is specifically inactivated. mRNA half-life measurements with the Tet-Off system have an important advantage in that cell viability is not affected as with actinomycin D. We tested some representative constructs of Fig. 2 as well as additional mutants (Fig. 4). The wild-type GFP-IL-6 construct gave a short half-life of approximately 33 min in NIH 3T3 cells with the Tet-Off system or after the addition of actinomycin D (not shown). Similarly, the IL-6_Δ2 mutant construct with only the first 224 nucleotides of the 3′UTR was also strongly destabilized, although its decay was slightly slower than in COS-7 cells. Like in COS-7 cells, mutants G and L were partially stabilized compared to mutant Δ2, whereas the double mutant GL was slightly more stable. This shows that elements of both regions are necessary for destabilization in NIH 3T3 cells (Fig. 4). By extending linker region L, which covers a conserved ARE (Fig. 5), to the adjacent region M, comprising another conserved ARE-like sequence, we did not observe a notable increase in RNA stability (mutant IL-6_LM in Fig. 4). However, the IL-6_Δ11 construct, with only the first 140 nucleotides of the IL-6 3′UTR and lacking all sequences resembling AREs, was more stable, similar to the Δ3 construct in COS-7 cells (Fig. 2). Further deletion of the FGH region (construct IL-6_Δ12) increased the half-life to a value equal to that of GFP mRNA without IL-6 3′UTR (Fig. 4). This result suggests a moderate contribution of the FGH region in mutant IL-6_Δ11 to the instability. A similar observation was made with mutant IL-6_Δ13 without additional determinants outside the FGH region. Overall, we conclude that results from COS-7 and NIH 3T3 cells are qualitatively similar.

FIG. 4.

Half-life analysis of GFP-IL-6 constructs by the Tet-Off method in mouse NIH 3T3 cells. (A) GFP-IL-6 constructs were cloned in a retroviral vector behind a 7× Tet operator and induced with a tetracycline-sensitive transactivator protein, tTA (17). The transcription activator was then specifically blocked by the tetracycline analogue doxycycline (Dox). The hatched squares indicate a 19-nucleotide linker sequence which replaces the natural sequence at this position. Gray squares indicate the position of the putative stem-loop region, and black squares are the positions of AUUUA sequences. T_1/2, half-life. (B) The GFP half-life was measured by real-time PCR and normalized to that of mARP0 mRNA (30).

FIG. 5.

Sequence comparison of the destabilizing 3′UTR elements in IL-6 mRNA. The first 224 nucleotides of the 3′UTR of various species were aligned by ClustalW and the conservation of each position (below the alignment) analyzed by the JalView program (http://www.ebi.ac.uk/clustalw/index.html). The FGH and L regions as well as AUUUA sequences are highlighted. A possible stem and loop region in the FGH box is indicated. IL-6 mRNA species are from Oryctolagus cuniculus (GenBank accession no. AF169176), Sylvilagus nuttallii (AF169178), Sylvilagus audubonii (AF169177), Rattus norwegicus (NM_012589), Mus musculus (NM_031168), Marmota monax (AF012908), Homo sapiens (M54894), Ovis aries (X62501), Sus scrofa (M80258), Canis familiaris (U12234), Bos taurus (X57317), and Gallus gallus (AJ309540).

Mutagenesis of a putative RNA stem-loop structure.

While mutant L corresponds to a highly conserved AU-rich region with the sequence UUAAUUUAU, the sequences of F, G, and H (nucleotides 762 to 806 in human IL-6 mRNA) do not show any AU-rich properties (Fig. 5). Instead, using M-fold (72), we predicted a stem-loop structure in the FGH region with 5 paired nucleotides and a 6-nucleotide loop. From work on iron-responsive elements, it is known that six-membered RNA loops make intraloop base interactions at nucleotides 1 and 5 (1, 23). Taking this into account, the prediction in the FGH region gave an excellent consensus stem-loop structure for most IL-6 species in the GenBank database, except in three rabbit sequences (Fig. 6A). Notably, a predicted CG interaction in the loop is conserved but inverted in chickens. Moreover, the chicken sequence shows a UA-to-CG base pair change in the stem next to the loop that may be needed to increase the stability of this short stem (Fig. 6A). In all other species, additional pairing in flanking regions may stabilize the structure.

FIG. 6.

Point mutations in the putative RNA stem-loop structure stabilize IL-6 mRNA. (A) The FGH region (Fig. 3) was analyzed for potential RNA folding using the M-fold algorithm (72). A short stem-loop structure with a 6-nucleotide loop and 5-nucleotide paired region was predicted for most species (except for the rabbit Oryctolagus and Sylvilagus species). In Marmota monax, the putative stem-loop is not conserved below the horizontal line. (B) The GFP-IL-6_Δ2 construct (Fig. 2) was mutated by double point mutations on either strand of the putative stem near the loop such as to abolish correct folding of the structure. Alternatively, both strands were mutated simultaneously to restore the putative base pairing. (C) Each mutant was tested for its mRNA half-life in stably infected COS-7 cells using the Tet-Off vector system. RNA levels were measured at 0, 60, and 120 min after doxycycline (Dox) addition by real-time PCR and normalized to that of endogenous β-actin mRNA. Average half-lives (T_1/2) ± SD are reported for four independent experiments.

To test whether the predicted structure was indeed important for IL-6 mRNA stability, we made two different mutants (m1 and m2) with double mutations in the upper stem next to the loop of the putative structure, such as to weaken the base pairing (Fig. 6B). In addition, we made a complementary double mutant with the potential to restore the stem (m3). They were tested in the context of the Δ2 construct (Fig. 2) by using the retroviral vectors and Tet-Off analysis system in COS-7 cells. All three mutants showed a degradation rate similar to that of mutant G (Fig. 3) and were significantly more stable than the Δ2 construct (Fig. 6C). Notably, the complementary mutations in IL-6_m3 did not restore rapid mRNA decay, suggesting that not just structure, but probably also sequence, is important in this putative stem-loop region.

Overexpression of AUF1 and its interaction with IL-6 3′UTR.

In order to test whether specific proteins bind to the cis elements of IL-6 mRNA, we carried out in vitro binding assays with radiolabeled RNA probes encompassing either the first 224 bases of the IL-6 3′UTR or a specific probe with the putative stem-loop region and cytoplasmic extracts from COS-7 or NIH 3T3 cells. Invariably, with or without UV cross-linking, weakly appearing complexes were observed on nondenaturing gels but had to be considered nonspecific since they were easily competed with 0.5 mg/ml heparin or 5 μg/ml tRNA (data not shown).

As a consequence, we tested the importance of AUF1, a candidate protein known to interact with AREs that might be important for IL-6 degradation. Several recent studies using RNAi or overexpression of AUF1 have come to conflicting conclusions as to whether AUF1 stabilizes or destabilizes ARE-containing mRNAs (see the introduction). We first analyzed the effect of the overexpression of different AUF1 isoforms on the stability of the GFP-IL-6 construct. For this, we constructed retroviral vectors based on the commercially available GeneSwitch system (Invitrogen). It uses a trans-activating fusion protein with the Gal4 DNA binding domain, the activation domain of p65 of the NF-κB complex, and a mutated progesterone receptor ligand-binding domain (5). In the presence of the progesterone antagonist mifepristone, the chimeric trans-activator binds to the Gal4 promoter and induces target gene transcription. We determined that AUF1 protein was maximally expressed 24 h after induction with 1 nM mifepristone. Under these conditions, all four isoforms were strongly expressed (Fig. 7A). We estimated by Western blotting that the level of induced protein was between 10- and 30-fold that of endogenous AUF1 (not shown). The effect of AUF1 isoforms on the mRNA half-life was tested in stably transfected cell populations with the GFP-IL-6 construct under the control of the Tet-regulated promoter. AUF1 was induced (or not) for 24 h and then doxycycline added to block transcription of the GFP construct. Both AUF1 p37 and p42 increased the half-life of GFP-IL-6 mRNA by factors of 2.6- and 2.3-fold, respectively. In contrast, p40 and p45 failed to show any stabilization (Fig. 7B).

FIG. 7.

Overexpression of AUF1 p37 and p42 isoforms increases the GFP-IL-6 mRNA half-life but only in the presence of the critical ARE at site L. (A) The AUF1 isoforms p37, p40, p42, and p45 tagged with a myc epitope were expressed in mouse NIH 3T3 cells from a mifepristone-inducible vector. AUF1 expression was analyzed before (−) or after (+) a 24-h induction on a Western blot with a specific anti-myc tag antibody. (B) The cell lines were in addition stably transfected with GFP constructs behind a Tet-regulated promoter. The IL-6 3′UTR or the mouse c-myc coding region and 3′UTR were cloned in frame behind the GFP coding region. GFP mRNA levels were measured after the addition of tetracycline without AUF1 overexpression (open circles) or after a 24-h induction of AUF1 isoforms by mifepristone (filled circles). The half-life was calculated by regression on semilogarithmic plots. (C) The half-lives of various mutant constructs of GFP-IL-6 mRNA (Fig. 4) were tested in NIH 3T3 cells in the absence (open circles) and presence (filled circles) of mifepristone-induced AUF1 p37.

In order to broaden the observation of mRNA stabilization by AUF1 overexpression, we tested two additional unstable GFP constructs. The first one was with the human GM-CSF 3′UTR, considered a prototype ARE-containing region which confers strong instability to reporters; the second one was an in-frame fusion between the GFP and c-myc coding sequences and the c-myc 3′UTR. AUF1 p37 provoked a 2.1-fold increase of the GFP-GM-CSF mRNA (data not shown). In contrast, none of the overexpressed AUF1 isoforms increased the half-life of the GFP-myc fusion mRNA (Fig. 7B).

Taking advantage of the mutants described in Fig. 4, we next investigated which part of the IL-6 3′UTR is required for stabilization by AUF1. The GFP-IL-6_Δ2 construct was also 2.9-fold induced by p37 overexpression, indicating the presence of critical sequences in first half of the 3′UTR (Fig. 7C). Similarly, the scanning mutant GFP-IL-6_G lacking the putative stem-loop but with its AREs intact showed a 2.0-fold-prolonged half-life with p37. However, with mutant GFP-IL-6_L, which lacks one of the ARE-like sequences, no enhancement of the mRNA stability by p37 was observed (Fig. 7C). This result indicates the importance of this ARE for the stability change and suggests that AUF1 may directly interact with this ARE.

In order to test whether the different AUF1 isoforms bind to the 3′UTR of IL-6 mRNA, we immunoprecipitated the proteins through their myc tag and measured the amount of coprecipitated GFP-IL-6 mRNA. In three independent experiments, between 25 and 65% of the mRNA coprecipitated with myc-tagged AUF1 p37 and p42 but not with p40 or p45 (Fig. 8A). Coprecipitations of GFP-IL-6 mRNA without AUF1 induction and of mARP0 mRNA served as controls to assess the nonspecific background. This was usually less than 10%. We next investigated which part of the mRNA was essential for coprecipitation. We found that in the presence of the ARE at site L (mutants GFP-IL-6_Δ2 and GFP-IL-6_G), the mRNA was well coprecipitated (Fig. 8B). In contrast, the GFP vector alone or mutant GFP-IL-6_L, which lacks this ARE, was not coprecipitated.

FIG. 8.

AUF1 p37 and p42 bind to the IL-6 3′UTR in NIH 3T3 cells only in the presence of the critical ARE at site L. (A) Each of the four myc-tagged AUF1 isoforms was expressed in mouse NIH 3T3 cells in the presence of the GFP-IL-6_wt construct. The proteins of whole-cell lysates were immunoprecipitated with anti-myc tag antibody. The amounts of GFP-IL-6 mRNA bound to antibody-coupled beads and GFP-IL-6 mRNA left in the supernatant were quantified by real-time PCR. Their sum was considered the total mRNA recovered (100%). The ratio of coprecipitated to total mRNA was calculated for each isoform and is shown on the graph. mARP0 mRNA served as a control for nonspecific coprecipitation. The values reported are averages from at least three independent experiments ± SD. (B) The binding of myc-tagged AUF1 p37 to various GFP-IL-6 constructs was assessed by the same method. The values reported are averages from three experiments ± SD.

Modulation of AUF1 and IL-6 mRNA stability by the proteasome.

Previous studies have shown that the inhibition of ubiquitin-conjugating activity can stabilize mRNA containing the GM-CSF ARE (35, 36). In order to test whether IL-6 mRNA also responds to this type of treatment, we added the specific proteasome degradation inhibitor MG132 to NIH 3T3 cells expressing the GFP-wild-type IL-6 (IL-6_wt) construct and measured the mRNA half-life by the Tet-Off system. The mRNA stability was indeed 3.7-fold increased (Fig. 9A). As shown in Fig. 7, high AUF1 levels might be responsible for a change in mRNA stability. We therefore analyzed the effect of MG132 on the expression of myc-tagged AUF1 p37 in NIH 3T3 cells. We found that MG132 increased the level of p37 (Fig. 9B). This suggests that the inhibition of AUF1 p37 degradation by the proteasome may increase endogenous AUF1 levels to the extent that it stabilizes GFP-IL-6_wt mRNA in a way similar to that of exogenously overexpressed p37.

FIG. 9.

Effect of the proteasome inhibitor MG132 on IL-6 mRNA and AUF1 stability. (A) The decay of GFP-IL-6_wt mRNA in mouse NIH 3T3 cells was analyzed using the Tet-Off system and real-time PCR after a treatment for 4 h with 40 μM MG132. Average results ± SD of four experiments are shown, and half-lives were calculated by linear regression of semilogarithmic plots. (B) Exogenous myc-tagged AUF1 p37 was induced with mifepristone in NIH 3T3 cells and its level analyzed on a Western blot before and after treatment for 4 h with 40 μM MG132. Expression of γ-tubulin was unchanged by MG132 and served as a loading control.

RNAi against AUF1.

Taken together, the results show that overexpressed AUF1 p37 and p42 bind to the ARE at site L in cells. It is tempting to conclude that p37 and p42 are stabilizing rather than destabilizing proteins. However, in order to complete the study, we also wanted to know what happens in the absence of endogenous p37 and p42. For this we used RNAi against all AUF1 isoforms. Three oligonucleotides corresponding to sequences in exons 4 and 5 that are shared by all AUF1 isoforms were stably expressed from a retroviral vector based on pRetroSUPER with a histone H1 polymerase III promoter (4). These RNAs were transcribed as hairpin structures and are processed in cells to produce the corresponding siRNA molecules. Western blot analysis showed that oligonucleotide 301 lowered markedly the expression of endogenous AUF1, with only about 10% of the protein remaining (Fig. 10A). We then measured the half-life of GFP-IL-6 in the presence of RNAi and found that it was prolonged about 2.5-fold (Fig. 10B). This indicates a destabilizing role of AUF1 in IL-6 mRNA regulation under normal physiological conditions.

FIG. 10.

RNAi against endogenous AUF1 stabilizes GFP-IL-6 mRNA. (A) Western blot analyzed with polyclonal anti-AUF1 and anti-γ-tubulin antibodies after expression of three different interfering RNAs in mouse NIH 3T3 cells. Oligo, oligonucleotide. (B) Half-lives of GFP-IL-6 mRNA measured by the Tet-Off system before and after infection with a retroviral RNAi vector carrying the oligonucleotide 301 sequence, which targets AUF1 exon 5.

DISCUSSION

The present study identifies two distinct regions in the 3′UTR of IL-6 mRNA that are necessary to trigger the short half-life of about 30 min. It confirms initial observations that the 3′UTR of IL-6 mRNA is important for its destabilization (3, 44, 56). The IL-6 3′UTR reveals, however, greater complexity than expected and highlights the limitations of predicting ARE-like destabilizing elements. The four ARE-like elements between nucleotides 825 and 885 destabilized GFP reporter mRNA only partially, and it would have been impossible to predict which among them was more important. We show that not all ARE-like sequences are equal in inducing instability, suggesting subtle differences in their sequence or environment. The sequence covered by mutant L, which is most essential, conforms only loosely to previous ARE definitions (31, 71), as its consensus reads UUA-AUUUA-U(A/G). Yet, it fulfills the criterion of having a conserved nearby ARE-like sequence in tandem. Consistent with the finding that AREs are insufficient to account for the overall IL-6 mRNA instability, we have newly identified a second instability region defined by the mutants F, G, and H. We propose that this region folds into a stem-loop structure (Fig. 6). Point mutations in the predicted stem confirm the importance of this sequence and suggest that disruption of the structure may directly affect rapid mRNA degradation. However, since complementary mutations restoring a stem did not restore rapid mRNA decay, we cannot decide whether sequence alone or sequence within a structure is important. Experiments that address directly RNA folding will be needed to resolve this. Destabilization by the putative stem-loop was almost as strong as by the ARE region in COS-7 cells but slightly weaker in NIH 3T3 cells. It seems possible that testing a human RNA element in monkey cells gives more accurate results than in mouse cells. To account for the entire 10-fold-increased mRNA-destabilizing activity in NIH 3T3 cells, we attribute about a 2-fold increase to the FGH region, a 3-fold increase to ARE-like sequences, and another 1.5-fold increase to the second half of the 3′UTR that we did not analyze.

A 3′UTR with multiple destabilizing elements may have evolved to make IL-6 mRNA more robust against mutations that abolish the short half-life. Each element may also respond to separate RNA-protein interactions, and this might augment the versatility of potential posttranscriptional regulation by signaling cascades. While prototype AREs, as in GM-CSF mRNA, reside in a narrow region (50), the more complex situation of IL-6 mRNA is not unique. Studies of G-CSF mRNA identified a destabilizing element adjacent to AREs that forms a stem-loop and prevents mRNA stabilization by the calcium ionophore A23187 (3, 46). The sequence and folding of the putative IL-6 mRNA stem-loop are clearly distinct from those in G-CSF mRNA. A second instability element outside AREs was also identified in TNF-α mRNA (53). It destabilizes TNF-α mRNA in a mutant cell line lacking the ARE-binding protein Tis-11b. Again, we see no similarity between this element and our putative stem-loop. Recently, an auxiliary HuR binding element was identified in the vicinity of the main destabilizing AREs of IL-8 mRNA (64). In endothelin-1 mRNA, two destabilizing regions were found, of which at least one functions as an ARE (40). Moreover, in c-fos and c-myc mRNA there exists good evidence for destabilizing sequences both in the coding region and the 3′UTR (21, 22, 24, 66).

As we could not identify any stable RNA-protein complex with the IL-6 3′UTR in vitro, we investigated whether the known trans-acting protein AUF1 plays a role in IL-6 mRNA instability. This was also motivated by the prevalent uncertainty as to whether AUF1 is stabilizing or destabilizing (see the introduction). We show in a single cell line and for a defined target mRNA that both exogenous overexpression of p37 and p42 (Fig. 7) and strongly diminished expression of all four endogenous AUF1 isoforms by RNAi (Fig. 10) enhance mRNA stability two- to threefold. Thus, our results reconcile apparently opposite views of the role of AUF1. It appears that destabilizing or stabilizing effects of AUF1 are essentially linked to the level of its expression. Under normal conditions, the AUF1 concentration seems optimal to ensure rapid IL-6 mRNA degradation. When AUF1 is suppressed, however, its destabilizing function disappears and RNA stability increases. These results are in agreement with siRNA targeting of total AUF1 in HeLa cells, where it stabilized p21 and cyclin D mRNA (33), and in endothelial cells, where the endothelin 1 mRNA level increased (40). They differ slightly from a study of HT1080, in which targeting of all four isoforms had no effect, and the targeting of only AUF1 exon 2, and hence p40 and p45, increased the half-life of a GFP-ARE reporter mRNA (48). Nonetheless, all these studies conclude that at least certain AUF1 isoforms have a destabilizing function in cells, a conclusion that was anticipated by in vitro studies (2, 70).

Most intriguing is the fact that certain AUF1 isoforms also stabilize ARE-containing mRNA when they are overexpressed (Fig. 7). Several groups have reported on the overexpression of AUF1 with quite different results. Thus, in the special case of hemin-treated K562 cells, ARE-containing mRNAs were found to be more stable in the first place and overexpression of AUF1, notably the p37 and p42 isoforms, promoted RNA degradation (38). The same group, using NIH 3T3 cells, found overexpressed p37 and p42 to stabilize mRNA with AREs from GM-CSF and TNF-α, and even with a c-fos coding region determinant (68). Our results with IL-6 and GM-CSF 3′UTR constructs agree entirely with this study. However, unlike with c-fos mRNA (68), we see no stability change with a GFP-c-myc mRNA construct. Probably the instability determinant of the c-myc coding region cannot be inhibited by excess AUF1, and its dominance may have masked effects of AUF1 at AREs in the c-myc 3′UTR. Others reported more-complex changes in transgenic mice, when exogenous p37 was expressed from a β-actin promoter (18). It upregulated c-myc, c-fos, and c-jun mRNA in the muscles and livers of certain mice but had little effect in other tissues or even downregulated GM-CSF and TNF-α mRNA in the spleen. This suggests that relative levels of endogenous AUF1 and possibly other tissue-specific factors influenced the outcome. It was also shown that p37 can be limiting when an ARE target mRNA was overexpressed (49). Increased p37 expression could overcome the limitation and promote mRNA decay. The effect of the proteasome inhibitor MG132 on the IL-6 mRNA half-life (Fig. 9) is in line with results of previous studies of other ARE-containing mRNAs (35, 36). The modulation of AUF1 isoform levels through ubiquitination (37) supports the idea that AUF1 levels control mRNA degradation rates.

We can think of various models of how excess AUF1 might interfere with its function as a degradation-promoting protein. First, its interaction with AREs may become more extensive due to the higher cytoplasmic concentration and the mass law governing molecular interactions. Thus, fewer AREs would remain unoccupied, and a possible exchange for other ARE-binding proteins that also promote RNA degradation might be diminished. Several other proteins, notably TTP (6, 32, 55), the TTP family member Tis-11B (BRF1) (7, 52), and KSRP (15), were documented to bind to AREs and to promote mRNA decay. However, competition for ARE-binding sites, maybe already in the nucleus (9), would resemble stabilization by excess HuR (8, 13) and would not explain why AUF1 is a destabilizing protein in the first place. Second, ARE-bound AUF1 might serve as a platform for binding of a protein with enzymatic activity in RNA decay. Overexpressed unbound AUF1 might compete away such a protein and prevent the activity from reaching its target mRNA. Third, ARE-bound AUF1 interacts with other proteins on mRNA (22, 39). Its overexpression might alter these interactions through the formation of AUF1 dimers or multimers on the 3′UTR (62) such that the degradation-promoting function gets lost. This would imply that IL-6 mRNA degradation depends on the nature and stoichiometry of protein-protein complexes on the RNA.

Our experiments favor the third model, because mRNA stabilization by overexpressed AUF1 p37 as well as destabilization of GFP-IL-6 mRNA requires the presence of the same critical ARE identified by linker-scanning mutant L (Fig. 3 and 7C). Precipitation of up to 65% of the GFP-IL-6 mRNA with myc-tagged p37 or p42 indicates that the ARE at L is the binding site for AUF1 in vivo (Fig. 8B). In addition, only the two AUF1 isoforms with the highest affinity for AREs (62), p37 and p42, stabilize GFP-IL-6 mRNA, suggesting that stabilization requires AUF1 binding. We propose that endogenous AUF1 also binds preferentially to site L, as both the ARE at L and the expression of a minimum amount of AUF1 (Fig. 10) are required for rapid mRNA decay. A central problem for future experiments will be to find out how natural AUF1 at site L initiates deadenylation and mRNA instability and how AUF1 overexpression prevents this process. The major protein that needs to be removed from the poly(A) tail during its shortening is the poly(A)-binding protein 1 (PABP1). In this context it is interesting that two RNA recognition motives of PABP1 show a high affinity for AREs (51) and that PABP1 interacts with AUF1 on c-fos mRNA (22) and in vitro (39). Maybe the role of AUF1 bound to site L is to facilitate the binding of PABP1 to adjacent AU-rich sequences, and these interactions might be required for subsequent removal of PABP1 from poly(A) tails. As shown recently, AUF1 interacts also with eIF-4G in vitro, and translation might contribute to the removal of AUF1-PABP1 complexes from the 3′UTR (39). AUF1 overexpression, however, might promote AUF1 oligomerization on the RNA (62) at regions adjacent to site L and interfere with PABP1 binding.

A short mRNA half-life triggered by specific cis elements is a feature of several hundred different transcripts per cell (69). In spite of each 3′UTR being unique in sequence, rapid degradation converges probably in a limited number of pathways. Thus, if just a few specific proteins recognize instability elements and activate mRNA degradation, we may wonder whether these proteins show redundancy or specialize for certain mRNAs. Our RNAi experiments with NIH 3T3 cells suggest that no other protein can replace AUF1. It should be interesting to analyze whether candidate proteins, notably of the TTP family, are actually present and interact with IL-6 mRNA. Experiments will also be needed to explore how the stem-loop region contributes to the recruitment of destabilizing proteins in vivo. Finally, it remains to be tested whether signaling events that stabilize IL-6 mRNA require precise sequence elements and regulate the IL-6 mRNA-AUF1 interaction.