Iron regulatory proteins 1 and 2 bind distinct sets of RNA target sequences.

Iron regulatory proteins (IRPs) 1 and 2 bind with equally high affinity to iron-responsive element (IRE) RNA stem-loops located in mRNA untranslated regions and, thereby, post-transcriptionally regulate several genes of iron metabolism. In this study we define the RNA-binding specificities of mouse IRP-1 and IRP-2. By screening loop mutations of the ferritin H-chain IRE, we show that both IRPs bind well to a large number of IRE-like sequences. More significantly, each IRP was found to recognize a unique subset of IRE-like targets. These IRP-specific groups of IREs are distinct from one another and are characterized by changes in certain paired (IRP-1) or unpaired (IRP-2) loop nucleotides. We further demonstrate the application of such sequences as unique probes to detect and distinguish IRP-1 from IRP-2 in human cells, and observe that the IRPs are regulated similarly by iron and reducing agents in human and rodent cells. Importantly, the ability of each IRP to recognize an exclusive subset of IREs was conserved between species. These findings suggest that IRP-1 and IRP-2 may each regulate unique mRNA targets in vivo, possibly extending their function beyond the regulation of intracellular iron homeostasis.

Iron regulatory proteins (IRPs) 1 and 2 bind with equally high affinity to iron-responsive element (IRE) RNA stem-loops located in mRNA untranslated regions and, thereby, post-transcriptionally regulate several genes of iron metabolism. In this study we define the RNA-binding specificities of mouse IRP-1 and IRP-2. By screening loop mutations of the ferritin H-chain IRE, we show that both IRPs bind well to a large number of IRE-like sequences. More significantly, each IRP was found to recognize a unique subset of IRE-like targets. These IRP-specific groups of IREs are distinct from one another and are characterized by changes in certain paired (IRP-1) or unpaired (IRP-2) loop nucleotides. We further demonstrate the application of such sequences as unique probes to detect and distinguish IRP-1 from IRP-2 in human cells, and observe that the IRPs are regulated similarly by iron and reducing agents in human and rodent cells. Importantly, the ability of each IRP to recognize an exclusive subset of IREs was conserved between species. These findings suggest that IRP-1 and IRP-2 may each regulate unique mRNA targets in vivo, possibly extending their function beyond the regulation of intracellular iron homeostasis.
A second IRE-binding protein has been characterized in rodents Guo et al., 1994) and in human cells (Samaniego et al., 1994), and is now commonly referred to as IRP-2. IRP-2 is a 105-kDa protein that binds specifically to all known mRNA IREs with an affinity equally as high as that of IRP-1 Guo et al., 1994). The two proteins are encoded by separate genes (Rouault et al., 1990), and their cDNA sequences (Rouault et al., 1992;Guo et al., 1995a) reveal conservation of cysteine residues known to ligate a [4Fe-4S] cluster in IRP-1 (Philpott et al., 1993;Hirling et al., 1994), although IRP-2 is apparently enzymatically inactive (Guo et al., 1994). The IRPs both respond to iron, but via different pathways. IRP-1 is post-translationally converted between active and inactive RNA-binding forms (reviewed by ). IRP-2, however, is induced following iron starvation through renewed synthesis of stable IRP-2 protein (Henderson and Kü hn, 1995;Pantopoulos et al., 1995), and its inactivation by iron reflects degradation of IRP-2 protein (Guo et al., 1994;Samaniego et al., 1994) by a translation-dependent mechanism (Henderson and Kü hn, 1995). The IRPs are able to inhibit the translation of IRE-containing mRNAs in vitro (Guo et al., 1994;Kim et al., 1995), and therefore both proteins are potential mRNA regulators in vivo.
The possibility that the IRPs may bind additional IRE-like hairpins, and thus regulate an extended repertoire of mRNAs, is supported by a recent study in which a wide range of suboptimal IRP-1 binding sequences were selected from a partially degenerate pool of ferritin IRE RNAs . Interestingly, several of these bound exclusively to IRP-1 and not to IRP-2. These findings suggest that the RNA-binding specificities of these two related proteins are overlapping, but different. We postulated that IRP-2 might also bind not only to wild-type IREs, but exclusively to a unique set of RNA targets. In this study we investigated this premise, using the ferritin IRE as a model for mutagenesis, as this sequence is the optimal IRP target and a range of well characterized mutants is already established for comparison . By focusing on sequential mutagenesis of different IRE loop bases, we directly define and compare the RNA-binding specificities of mouse IRP-1 and IRP-2. Our new findings demonstrate the favorable effect that a structured IRE loop has on RNA binding by both IRPs and identify a novel set of four RNAs that bind exclusively to IRP-2. Furthermore, an assay is described for the discrimination of human IRP-1 from IRP-2, using IRP-specific RNA probes. Finally, we show that several features that distinguish the regulation of rodent IRP-1 and IRP-2 are well conserved in humans.

MATERIALS AND METHODS
Cell Culture and Treatments-Mouse B16.F1 melanoma cells and human FEK-4 fibroblasts and HL60 promyelocytic leukemia cells were grown in ␣-minimal essential medium. Human Molt-4 T-cell leukemia, HeLa cervical carcinoma cells, HFF fetal fibroblasts, TK6 lymphoblasts, and HT-1080 fibrosarcoma cells were cultured in Dulbecco's modified Eagle's medium. The human colon carcinoma cell lines SW480 and Col l5 were grown in L15 medium. All cell cultures were supplemented with 10% fetal calf serum, except for cell lines SW480, Col l5, and FEK-4, which required 15% fetal calf serum. The chelation of intracellular iron was achieved by 20 h of treatment with 100 M desferrioxamine (Desferal; gift from Ciba-Geigy, Basel, Switzerland). For iron treatments, 60 g/ml ferric ammonium citrate was added to cells after washing once with phosphate-buffered saline.
Cell Fractionation Procedure-Mouse B16 cells were treated for 20 h with 100 M desferrioxamine in order to fully induce IRE-binding proteins, and the cells were then harvested at 4°C in lysis buffer containing 10 mM Hepes, pH 7.5, 3 mM MgCl 2 , 40 mM KCl, 5% glycerol, 0.3% Nonidet P-40, and 1 mM phenylmethylsulfonyl fluoride. The cytoplasmic protein extract (36 mg) was next diluted to 24 ml in Buffer A (20 mM Tris⅐HCl, pH 8, 8 mM 2-mercaptoethanol (2-ME), 5% glycerol), resulting in a final salt concentration of 5 mM KCl. The extract was prefiltered through a 0.25-m Nalgene filter prior to loading onto a Mono-Q anion exchange column (Pharmacia Biotech Inc.) equilibrated in Buffer A. The column was first washed in Buffer A, and flow-through and 10 1-ml washes were collected. Subsequently, bound proteins were eluted from the column in 40 1-ml fractions using a linear salt gradient of 5 mM to 600 mM KCl (in Buffer A). Phenylmethylsulfonyl fluoride was added to 1 mM to each fraction, and samples were stored frozen at Ϫ70°C. Every second fraction was tested by band-shift assay for IRE binding activity. The fractionation was performed three times at 4°C, giving similar elution profiles for IRP-1 and IRP-2 each time.
Screening Fractionated Extract with Degenerate IRE Probes-Fractions containing IRP-1 or IRP-2 were detected by band-shift assay, using a wild-type ferritin heavy chain IRE probe (transcribed from a plasmid (clone 42); Henderson et al. (1994)). Positive identification of the IRP-enriched fractions was made by UV cross-linking experiments and Western blot analysis using antisera that cross-react with human IRP-1 (Ab1) or both IRP-1 and IRP-2 (Ab3) (see Henderson et al. (1993)).
Fractions were then analyzed by band-shift assay using 32 P-labeled degenerate IRE stem-loops. Two sets of four partly degenerate IREs were designed, which essentially represent variations of a previously described hairpin sequence, identical to the wild-type ferritin IRE except for randomization of the 7 conserved bases in the loop and bulge (for details of sequence, see Henderson et al. (1994)). The first group of 4 IRE hairpins (Series 1) was made partly degenerate only in the loop. Each Series 1 pool maintained a fixed base at loop position 1, while the remaining 5 bases were completely randomized except for the base at position 6, which could not pair with the first loop base. Series 2 RNAs differed from those of Series 1 only in that the base at loop position 5 was constrained to pair with that of position 1. The loop sequence and degeneracy of these RNAs is shown in Fig. 2. The RNAs were labeled and used to probe B16 fractions. In each case, protein was in excess, and the detection of IRP complexes was limited by the number of actual IRP-binding sequences in each RNA mixture. The detection limit of our assay was about 100 cpm/complex, and mixed RNA probes were used at levels (5 ϫ 10 4 to 10 5 cpm) sufficiently high to detect a protein binding to a single sequence with high affinity. While certain degenerate IRE pools detected IRP-1 or IRP-2, all RNA mixtures detected some additional proteins, none of which were sequence-specific as judged by band-shift competition assays.
Selection of IRP-binding Sequences-Oligonucleotide templates corresponding to the Series 2 RNA pools C 1 G 5 and G 1 C 5 , were made double-stranded by annealing a T7 promoter primer and extending with Klenow polymerase (Boehringer Mannheim). The degenerate DNA duplexes were then cut with SacI and BamHI, and directionally cloned into pGEM-3Zf(Ϫ) (Promega) as outlined in Fig. 3. About 300 clones of each degenerate pool were screened for binding to mouse IRP-1 or IRP-2. Briefly, DNA was prepared from pools of five bacterial colonies and used as template for in vitro transcription of 32 P-labeled RNA probes. 5 ϫ 10 4 cpm of each gel-purified probe was mixed with 12 g of B16 total cytoplasmic extract and IRP⅐IRE complexes detected by bandshift assay. Individual IRP-binding sequences were subsequently identified by rescreening positive pools, and sequenced by the method of Sanger et al. (1977) using the M13 reverse primer and Sequenase version 2.0 (U. S. Biochemical Corp.).
RNA Preparation-Degenerate oligonucleotide templates were made double-stranded by polymerase chain reaction (see Henderson et al. (1994)), to avoid folding of the DNA during transcription. Plasmid DNA templates (in pGEM-3Zf(ϩ/Ϫ)) were first linearized with BamHI prior to RNA preparation. In vitro transcription was performed with T7 RNA polymerase (Promega) in a reaction containing 1.5 mM unlabeled ATP, GTP, and UTP, and 40 -60 Ci of [␣-32 P]CTP (800 Ci/mmol; Amersham Corp.) for probes, or 1.5 mM CTP for unlabeled competitor RNAs. 32 P-Labeled RNAs were purified after separation on 15% polyacrylamide gels, and unlabeled transcripts were purified from agarose gels as described previously .
RNA Band-shift Assay-RNA-protein complexes were resolved by band-shift assay. Gel-purified 32 P-labeled RNAs were incubated for 10 min at 25°C with cytoplasmic extract, enriched IRP fractions, or recombinant human IRP-1 expressed as fusion proteins with glutathione S-transferase (Drapier et al., 1993) or a (His) 6 tag , and resolved on 6% non-denaturing polyacrylamide gels at 4°C as described previously (Mü llner et al., 1989). Unless otherwise stated, the binding reactions (20 l final volume) were performed in the presence of 5 mg/ml heparin to displace nonspecific RNA-protein interactions, and with 2% 2-ME to activate IRPs in vitro.
More specific conditions were required to detect individual human IRPs using selected IRP-specific probes. Importantly, human IRP-2 appears more susceptible to oxidation than rodent IRP-2, and thus extracts were prepared fresh and processed quickly for experiments to be performed in the absence of reductant. Up to 4 g of human cytoplasmic extract was preincubated with 300 ng of yeast tRNA, and then mixed with 5 ϫ 10 4 cpm of IRE probe (mutant GG1 or CG125) for 10 min in the presence or absence of 1% 2-ME (higher concentrations of 2-ME can decrease human IRP-2 activity), and in the presence of 5 mg/ml heparin. Gels were scanned and bands quantitated using a Compaq PhosphorImager equipped with Molecular Dynamics Image software.
Measurement of RNA-binding Affinities-The dissociation constant (K d ) for binding of a wild-type ferritin IRE to IRP-1 is between 10 and 90 pM (Haile et al., 1989;Barton et al., 1990). Rather than determine dissociation constants for all the selected mutant IREs, which would prove difficult for sequences that bind poorly to the IRPs and are less likely to reach saturable binding, we adopted two approaches to measure relative binding affinities. The first approach was a competition assay to determine the D 1 2 values of each IRP-binding sequence selected, where D 1 2 is the concentration of competitor required to reduce IRP binding of a 32 P-labeled wild-type IRE probe by 50% (see Dingwall et al. (1991)). The IRE probe (clone 42 RNA; 10 4 cpm, 0.093 nM) was added simultaneously with a 1-1000-fold excess of unlabeled competitor RNA to IRP-enriched B16 fractions, incubated for 15 min at 25°C in the presence of 2% 2-ME, and IRE⅐IRP complexes detected by band-shift assay. IRP in the cell extract used was pre-determined by titration curves to bind [ 32 P]IRE at a 1:1 molar ratio. Competition curves were plotted for each RNA, and the ratio of the D 1 2 values of wild-type and mutant IRE sequences yielded K Rel , the relative binding efficiency of each RNA (further details given by Henderson et al. (1994)).
In addition to competition assays, which were performed in the absence of heparin, each selected RNA was 32 P-labeled and gel-purified, and then tested for binding to IRP-1 or IRP-2 (protein in excess) by band-shift assay. The amount of IRE⅐IRP complex formed by each probe (2 ϫ 10 4 cpm; 0.04 ng of RNA/reaction) was determined relative to the wild-type IRE, and a "binding ratio" then assigned to each sequence.
UV Cross-linking of RNA-Protein Complexes-For UV cross-linking, 2 ϫ 10 6 cpm of gel-purified [ 32 P]RNA was incubated with an excess of IRP-containing cytoplasmic extract, in the presence of 1.5% 2-ME, for 5 min at room temperature. The specificity of the reaction was enhanced by the addition of 100 ng of yeast tRNA and 200 ng of a random stem-loop RNA . Unbound probe was degraded by addition of 0.1 unit of RNase T1 (Calbiochem) for 5 min, after which heparin was added to 5 mg/ml for 10 min. The reaction mixtures were then transferred on ice and irradiated 2 cm below a Philips TUV 15-watt UV lamp for 25 min, followed by an additional 5-min incubation at 25°C with 1 unit of RNase T1. Samples were then denatured in SDS buffer and separated on a denaturing 8% SDS-polyacrylamide gel.

Sequence-specific Binding of Two Distinct Proteins to Ironresponsive Elements-
The objective of this study was to define and compare the RNA-binding specificities of IRP-1 and IRP-2. Since the cellular forms of each IRP can undergo modification (Eisenstein et al., 1993;Henderson and Kü hn, 1995), we began with a systematic approach to enrich for all IRE-binding proteins within a single cell line, and then screened for IRPbinding sequences. Cytoplasmic extract from B16.F1 mouse melanoma cells (treated with desferrioxamine for 20 h to ensure IRP activation) was fractionated over a Mono-Q anion exchange column, and proteins eluted by linear gradient from 5 to 600 mM KCl. IRP-1 and IRP-2 activities were identified by RNA band-shift assay (see Fig. 1), and their specificity confirmed by UV cross-linking and RNA competition experiments (data not shown). Immunoblot analysis revealed a close correlation between IRP RNA binding activities and protein levels (see "Materials and Methods"). While most IRP-1 was recovered in the flow-through as previously observed (Barton et al., 1990;Henderson et al., 1993), a minor peak was eluted with 150 -160 mM KCl (Fig. 1). This small peak of IRP-1 may represent modified protein; however, it did not appear to differ in RNA-binding specificity compared to the major form of IRP-1 (data not shown) and awaits more detailed characterization. Recovery of IRP-2 peaked at 175 mM, then trailed off, with some IRP-2 still detectable in 400 mM salt fractions. Interestingly, a third RNA-binding protein complexed with the wildtype ferritin IRE probe (see fractions 16 -19, Fig. 1); however, this protein bound equally as well to other RNA stem-loops of different structure (e.g. yeast tRNA; data not shown) and therefore was not studied further.
As expected, the above analysis identified IRP-1 and IRP-2 as the only sequence-specific IRE-binding proteins in B16 cells. To test whether IRP-related proteins may exist that bind RNA hairpins of similar structure but different sequence to the IRE, we probed B16 fractions with a 32 P-labeled ferritin IRE probe made degenerate at the 7 conserved loop and bulge nucleotide positions (see Henderson et al. (1994) and Fig. 2A), and detected several additional RNA-protein complexes (data not shown). In order to assess the specificity of the extra band-shift complexes, a set of four less degenerate RNA mixtures were prepared, in which bases at the bulge and the first loop position were fixed (see Series 1 RNA pools in Fig. 2A). Band-shift analysis of the mouse B16 fractions revealed binding of IRP-1 and IRP-2 to certain [ 32 P]RNA pools (e.g. C 1 N 5 and G 1 N 5 probes in Fig. 2B); however, following detailed examination and a series of competition assays, we identified no other proteins with specificity for IRE-like sequences. We therefore have no evidence to suggest the existence of other IRPs.
RNA Recognition by IRP-1 and IRP-2 Requires a Structured IRE Loop-Previously, we reported binding of IRP-1 to IREs containing C 1 G 5 or U 1 A 5 loops (base pairing between positions 1 and 5), but not to G 1 C 5 -type IREs . Since the G 1 C 5 IREs tested could alternatively fold as tetraloops, which showed reduced binding to rodent IRP-1 and in particular to IRP-2 , we have re-examined the IRP requirement for intra-loop base pairing using paired RNA mixtures incapable of tetraloop formation. A second series of IRE-like hairpin mixtures was prepared, less degenerate than the Series 1 RNAs (see Fig. 2A). The Series 2 RNA pools differed from the first series only in that the fifth base of the loop was constrained to pair with the first. This single change increased IRP-binding of certain IRE-like RNA pools by up to 4-fold (see band-shift in Fig. 2B). Quantitation of paired RNA probes ( Fig. 2A) confirmed the preference of IRP-1 for C 1 G 5 and U 1 A 5 IREs, and in addition revealed strong binding of IRP-1 to G 1 C 5 IREs (a 2-fold increase was expected, as IRP-1 also binds well to G 1 G 5 IREs) . Surprisingly, IRE binding by IRP-2 demonstrated a similar requirement for 1:5 loop pairing; however, IRP-2 did not recognize U 1 A 5 IREs, and neither IRP bound detectably to IREs with an A 1 U 5 loop (summarized in Fig. 2A).
Relative Binding Affinities of Selected IRP-1 and IRP-2 Binding Sequences-We predicted that 1:5 base pairing might position the IRE loop nucleotides 2, 3, and 4 for direct protein contact (see Henderson et al. (1994)). These bases may in part determine the binding specificity of IRP-1 and IRP-2. Working on this premise, we cloned degenerate DNA fragments corresponding to the CNNNGN and GNNNCN loop-containing IRE probes (described in Fig. 2) into pGEM-3Zf(Ϫ), and screened about 300 clones from each mixture for binding to IRP-1 or IRP-2 (procedure outlined in Fig. 3A). Individual positive clones were sequenced, and the binding affinities of in vitro transcripts then tested by competition assay and band-shift assay as described (see Henderson et al. (1994) and "Materials and Methods"). We identified 11 different positive C 1 G 5 -type FIG. 1. Detection of IRE-binding proteins in mouse extract. Mouse B16 melanoma cytoplasmic extract was fractionated by linear gradient over a Mono-Q column, and 1 l of selected fractions analyzed by band-shift assay. Samples were probed with 5 ϫ 10 4 cpm of 32 P-labeled IRE, and IRE-protein complexes corresponding to IRP-1 and IRP-2 are clearly indicated. A third IRE-binding protein was detected in fractions 16 -19; however, this RNA-binding protein (RBP) was not specific for IRE stem-loops. The gel was scanned with a Molecular Dynamics PhosphorImager, and relative signal strength is plotted in the lower panel. The graph does not accurately portray the total relative amounts of IRP, as while each fraction volume was 1 ml, most IRP-1 eluted in the flow-through (24 ml volume). Cross-contamination of IRP fractions was minimal; Ͻ 0.3% of total IRP-2 activity eluted in the first wash, and in the major IRP-2 peak fraction (175 mM KCl), IRP-1 comprised Ͻ0.2% of total IRE binding activity.

IRP-1 and IRP-2 Differ in RNA-binding Specificity
RNAs and 7 different G 1 C 5 -type hairpins. Selected 32 P-labeled transcripts were gel-purified and compared for binding to mouse IRP-1 and IRP-2 relative to a wild-type ferritin IRE (see Fig. 3B). The 5Ј-GAGUCG-3Ј loop IRE (mutant GC6), which is wild type at loop positions 2-4, bound both IRP-1 and IRP-2 almost as well as the C 1 G 5 wild-type sequence. The binding ratios (quantitation of band-shift complex formed relative to ferritin IRE) for individual sequences are summarized in Fig. 4 and are presented with relative affinity values (K Rel ) calculated from competition assays. The two approaches indicated a similar trend in the binding affinities of each RNA, although the competition assays were performed in the absence of heparin and generally (with a few exceptions) scored higher values than the binding ratios obtained in the presence of 5 mg/ml heparin (see Fig. 4 and Henderson et al. (1994)).

Identification of IRP-2-specific RNA Target Sequences-
Some of the C 1 G 5 sequences tested (CG mutants 305, 153, and 218) had been selected previously using a PCR selection/amplification approach , and similar binding affinity values were obtained. Several sequences from each group bound well to both IRP-1 and IRP-2; however, we selected no sequence with preference for mouse IRP-1. By contrast, we identified four different suboptimal sequences, which bound preferentially to IRP-2 (binding 12 to Ͼ140 times stronger than to IRP-1; see Table I). Each of the IRP-2-specific RNAs maintain a G at loop position 3, but differ from the wild-type sequence at positions 2 and 4. Two of these RNAs share the same central loop sequence CGC, but differ in their type of intraloop base pair (compare mutants CG147 and GC157). The loop nucleotide G 3 is clearly important in this context, as its replacement by U 3 abolishes binding to both IRPs (see negative G 1 C 5 mutant GC-1, Figs. 3B and 4).
The IRE mutant CG125 (loop: 5Ј-CCGAGC-3Ј) showed the greatest specificity for IRP-2, as judged by comparison of K Rel values and binding ratios (Fig. 4, Table I). We tested further the specificity of this IRE variant by band-shift analysis (Fig.  5). Binding of the mutant CG125 probe to mouse IRP-1 and IRP-2 was compared to a wild-type IRE, in the presence or absence of heparin. Remarkably, in the absence of heparin, the CG125 probe bound mouse IRP-2 50% as well as the wild-type IRE, but it showed no detectable binding to IRP-1 (see Fig. 5). Fig. 1) is shown at top, adjacent to a degenerate IRE made by randomizing the 7 conserved nucleotides of the IRE bulge (cytosine) and loop (top right; numbering of loop bases is indicated). Two series of RNA pools (degenerate only in the loop) were then prepared and tested for binding to the IRPs (see "Materials and Methods"). The degeneracy of each randomized IRE pool is given in brackets; Roman N is any base, and italic N is any base that does not pair with the first loop position. Equivalent limiting amounts (5 ϫ 10 4 cpm) of each 32 P-labeled, gelpurified RNA was used to probe the mouse B16 fractions shown in Fig.  1. The ability of Series 2 RNAs to bind IRPs is shown relative to that of Series 1 RNAs (lower panel, box). B, examples of band-shift assay gels are shown, comparing the C 1 G 5 and G 1 C 5 RNAs of Series 2 with their Series 1 counterparts. Fractions were preincubated with 200 ng of yeast tRNA, and additional nonspecific complexes are visible in each sample above the IRP complexes.

FIG. 3. Selection of IRP-binding sequences.
A, degenerate DNA fragments corresponding to the Series 2 C 1 G 5 and G 1 C 5 RNA pools were cloned, transcribed, and individual IRP-1 or IRP-2 binding sequences identified by band-shift assay using the selection procedure outlined. B, equal amounts (2 ϫ 10 4 cpm) of selected RNAs were labeled with [ 32 P]CTP, gel-purified, and compared for binding to IRP-1 or IRP-2 by band-shift assay. The IRPs were in excess in total or enriched B16 fractions. The wild-type human ferritin IRE (clone 42) is included as positive control; a randomly selected non-binding sequence (mutant GC-1) is a negative control.

IRP-1 and IRP-2 Differ in RNA-binding Specificity
Increasing the stringency of the reaction, by adding heparin, reduced binding of the CG125 probe to IRP-2 as expected (binding decreased to about 12% of wild type). This experiment confirms mutant CG125 as an IRP-2-specific RNA target.

Conserved Specificity of IRE Variant Probes for Human IRP-1 and IRP-2-
The range of RNA sequences targeted by mouse IRP-1 and IRP-2 overlap, but are clearly different. If these differences are of functional importance, then they should likewise be observed in human cells. Currently, there is no means to distinguish the RNA binding activities of human IRP-1 and IRP-2, as the human IRP⅐IRE complexes co-migrate on band-shift gels . We therefore investigated whether RNA probes specific for mouse IRP-1 or IRP-2 might show similar preference for the human IRPs. In addition to demonstrating conservation of the binding specificity, such probes would provide a valuable tool to distinguish IRP-1 from IRP-2 in different human cell types. Toward this goal, we tested two partially characterized variant IREs, one specific to mouse IRP-2 (mutant CG125; this study), and a previously selected IRP-1-specific RNA, mutant GG1 (loop sequence, 5Ј-GAGAGU-3Ј; Henderson et al. (1994)). The variant IREs bind to rodent IRP-1 or IRP-2 with affinities ϳ50% and ϳ14%, respectively, that of a wild-type IRE in the presence of 5 mg/ml heparin (see Fig. 4 and Henderson et al. (1994)).
We next tested the ability of each IRP-specific probe to bind different forms of bacterially expressed, purified recombinant human IRP-1 (Fig. 6A). The IRP-2-specific probe (mutant CG125) bound to saturating amounts of histidine-tagged IRP-1 fusion protein less than 1% as effectively as the IRP-1-specific (mutant GG1) probe. The slightly better binding (5% of wild type) of the IRP-2-specific probe to a glutathione S-transferase-IRP-1 fusion protein reflected nonspecific binding, as determined by comparison with other RNA stem-loops (data not shown). This result confirms the difference in affinity of these probes for human IRP-1. We then assessed the ability of mutant GG1 and mutant CG125 32 P-labeled RNAs to form UVcross-linked covalently bound complexes with mouse and human cytoplasmic proteins. Conditions of partial RNase T1 digestion were attained, which demonstrated clear UV crosslinks to proteins of ϳ100 kDa (Fig. 6B). The wild-type and mutant IRE probes cross-linked specifically to mouse IRP-1 and/or IRP-2 proteins from enriched mouse B16 cell fractions. It was not possible under these conditions to distinguish between the two human IRPs, which are of similar size (IRP-1, ϳ98 kDa; IRP-2, ϳ105 kDa; Rouault et al. (1992)). However, UV cross-linking of human Molt-4 extracts showed strong binding of all probes to a ϳ100-kDa protein. We expect, given the poor affinity of the CG125 probe for recombinant human IRP-1, that the ϳ100-kDa human protein cross-linked by this probe is most likely IRP-2.
RNA Probes That Detect and Distinguish IRP-1 from IRP-2 FIG. 4. Sequence and binding affinities of selected RNAs. The loop sequence of selected RNAs is shown relative to the wild-type sequence (clone 42: 5Ј-CAGUGC-3Ј). Bases capable of pairing (positions 1 and 5) are in large type, and non-wild-type nucleotides are underlined. The relative binding affinities of each RNA were determined by competition assay (K Rel values), or by 32 P-labeling each RNA and comparing IRP binding of each sequence with a wild-type IRE probe (see "Binding Ratios"; see Fig. 3B for an example). Both approaches utilized the band-shift assay (see "Materials and Methods" for details), and all binding affinities are relative to the wild-type IRE, set a value of 1. Binding assays were performed in the presence of 2% 2-ME. ND, not determined. Mutant GC-1 (loop: 5Ј-GCUCCG-3Ј) is a negative control.

TABLE I Selected IRE sequences specific for IRP-2
The values shown are the binding ratios (taken from Fig. 4) and represent relative binding of RNAs to IRP-1 or IRP-2 in the presence of 5 mg/ml heparin. Bold type indicates those bases with the potential to pair; underlined bases differ from wild type.

Loop sequence position
Relative binding ratio 5. Identification of an IRP-2-specific RNA sequence. 32 P-Labeled mutant CG125 RNA was compared to wild-type IRE for binding to mouse IRP-1 or IRP-2. Equal amounts (2 ϫ 10 4 cpm) of each probe was incubated with enriched IRP fractions in the presence (ϩ) or absence (Ϫ) of 5 mg/ml heparin, after addition of 2% 2-ME. The bandshift gel was scanned by phosphorimager, and the relative band intensities graphed (lower panel). The mutant CG125 probe bound exclusively to IRP-2, independent of heparin treatment.
in Different Human Cell Lines-Cytoplasmic extracts from different human cell types were treated in vitro with 2-ME to activate IRP RNA binding activity and analyzed by band-shift assay using gel-purified RNA probes (Fig. 7A). The migration of human [ 32 P]RNA-protein complexes was compared to that of mouse IRP-1 and IRP-2, included as a reference (lane 1). As shown in Fig. 7A, while the wild-type probe recognized both mouse IRP proteins, the IRP-1-and IRP-2-specific probes detected only the expected IRP. The band-shift signals were quantitated by phosphorimaging, and the amounts of IRP-1and IRP-2-specific complex were normalized relative to the wild-type probe, to compensate for differences in RNA-binding affinities (see Fig. 7B).
The IRP-1-specific probe generated a single human bandshift complex with a pattern very similar to that of the wildtype ferritin IRE probe; IRP-1 activity was lowest in human TK6 lymphoblasts and HL60 promyelocytic leukemia cells. By contrast, the IRP-2-specific probe formed a band-shift complex migrating at the same position as that of IRP-1 in human cells; however, the pattern was markedly different. IRP-2 activity was strongest in Molt-4 T-cell leukemia cells and markedly lower in fibroblasts (HFF, FEK) and different tumor-derived cell lines. The same expression pattern was observed independently using a different IRP-2-specific probe (mutant GC157; loop sequence: 5Ј-GCGCCG-3Ј), although the background was somewhat higher (data not shown).
These results confirm that the exclusive binding specificities of IRP-1 and IRP-2 are at least partly conserved between rodents and humans. Furthermore, the IRP-specific probes now provide a useful assay for the differential detection of the human proteins, and indicate that their expression is cell type-specific.

Regulation of Human IRP-1 and IRP-2 Activity by Iron and
Reducing Agents-We next tested whether the activity of IRP-1 and IRP-2 is iron regulated in human cells. As illustrated in Fig. 8, IRP-1 was strongly induced by 20 h of desferrioxamine treatment and repressed by the re-addition of iron salts for 4 h in HL60 and HeLa cells, but not in Molt-4 cells. IRP-1 activities were mostly normalized following in vitro reduction with 2-ME. FIG. 6. Specificity of different probes for human IRP-1 and IRP-2. A, 4 ϫ 10 4 cpm of wild-type, IRP-2-specific (mutant CG125) and IRP-1-specific (mutant GG1) IRE probes were compared for binding to different forms of recombinant human IRP-1 by band-shift assay. The probes were incubated with saturating amounts of a glutathione Stransferase-IRP-1 fusion protein (GST fusion) or histidine-tagged IRP-1. The data were quantified and plotted relative to wild-type IRE, which is set at 100%. B, the above probes (2 ϫ 10 6 cpm) were UV-crosslinked to proteins in mouse IRP-1 or IRP-2 fractions, and in human Molt-4 cytoplasmic extract (see "Materials and Methods"). Cross-linked RNA-protein complexes were separated on a denaturing 8% SDS-polyacrylamide gel. The probes covalently bound to ϳ97-kDa proteins corresponding to mouse IRP-1 and/or IRP-2 as expected. All probes, including mutant CG125, cross-linked to a similar sized ϳ97-kDa protein in human Molt-4 extracts, in addition to some smaller sized proteins that may include IRP degradation products. The autoradiographs shown represent 4-day exposures of x-ray film to the gels. FIG. 7. Detection of IRP-1 and IRP-2 RNA binding activities in human cells. A, cytoplasmic extract (4 g) from different human cell lines was probed with gel-purified wild-type and IRP-specific mutant RNAs (5 ϫ 10 4 cpm) in the presence of 1% 2-ME and 5 mg/ml heparin, and analyzed by band-shift assay. 2 g of total mouse B16 extract was included as a control. All probes detected human IRE-protein complexes migrating at the same position as mouse IRP-2 (mIRP-2); however, IRP-1 and IRP-2-specific probes generated different patterns, corresponding to variations in expression of the human IRPs (hIRPs). The autoradiographs shown represent different exposures (e.g. ratio of exposure times was 1:2:9 for wild-type, mutant GG1, and mutant CG125 probes, respectively) to compensate for differences in binding affinities of the probes. This scaling was made according to the relative affinities of these probes for the mouse IRPs, as evident by comparison of the mouse IRP signals. B, gels were quantitated and the relative signal intensities plotted as shown, after normalizing the values for IRP-1 (mutant GG1; ϫ 2) and IRP-2 (mutant CG125; ϫ 9) probes against the wild-type IRE. To aid comparison, the IRP-2 activity profile is shown superimposed over the wild-type expression profile.
Interestingly, IRP-2 responded in much the same way as IRP-1 to changes in iron levels; however, its activity could not be recovered by 2-ME in extracts from iron-treated cells (Fig. 8). This finding is in accord with the distinction between IRP-1 and IRP-2 in rodent cells . Our data suggest that iron decreases IRP-2 protein levels in HL60 and HeLa cells, which is supported by recent Western blot analysis of IRP-2 protein in HeLa cells (Guo et al., 1995b). IRP-2 was more active in untreated Molt-4 cells and HeLa cells than in HL60 cells. IRP-2 activity, like that of IRP-1, was also refractory to iron modulation in Molt-4 cells.
This experiment reveals that endogenous human IRP-1 and IRP-2 activities are iron-regulated. In addition, we observed that pretreatment of HeLa cells with cycloheximide blocked the inactivation of IRP-2 by iron, but not the inactivation of IRP-1 (see Fig. 8, lower panel, for quantitation). The same result was observed with the translation inhibitor anisomycin (data not shown) and correlates perfectly with recent data on rodent IRP-2 (Henderson and Kü hn, 1995). When considered together, our findings strongly suggest that our probes do indeed detect and discriminate between the two human IRPs. Furthermore, the features that distinguish binding activity of IRP-1 from IRP-2 are well conserved between human and rodent cells. DISCUSSION The IRP⅐IRE interaction represents a unique cellular mechanism underlying the post-transcriptional regulation of several genes. Recent evidence supports the view that IRP-1 and IRP-2 each function as mRNA trans-regulators involved in maintaining intracellular iron homeostasis, as both proteins (i) are regulated by iron levels , (ii) inhibit IRE-containing mRNA translation in vitro (Guo et al. 1994;Kim et al., 1995), and (iii) bind ferritin and transferrin receptor mRNA IREs with equally high affinity Guo et al., 1994). In this study, we further show that the two IRPs have conserved the ability to individually recognize exclusive and distinct sets of RNAs. This novel finding suggests the possibility that the IRPs may regulate additional mRNA targets. We also describe features of the IRE hairpin that permit binding by either or both IRPs and have used this information to establish an assay for the detection of individual IRP RNA binding activities in human cells.
IRPs Differ in RNA-binding Specificity-A two-step screening protocol was designed to define the RNA-binding specificities of mouse IRP-1 and IRP-2 (Figs. 2 and 3). The first stage revealed that IRE-binding by both IRPs required a (1:5) base pair interaction between positions 1 and 5 of the 6-base IRE loop. This was previously discovered as a critical feature of RNA-binding by IRP-1 . The fact that optimal binding by IRP-2 also favored a defined loop structure may implicate some similarity in the RNA-binding domains of the two proteins. The type of loop (1:5) base pair influenced IRP binding efficacy, and, perhaps in turn, the stability of different 1:5 loop base pairs was affected by IRP contact. In the final selection, we identified C 1 G 5 and G 1 C 5 base pairs as the only ferritin IRE loop interactions tolerated by both IRPs.
IRP-1 was generally more tolerant of IRE base changes than was IRP-2, particularly of specific base alterations in the bulge and loop (1:5) base pairs. Only IRP-1 recognized IRE loops containing a U 1 A 5 pair, and some less stable non-Watson-Crick base pairs . While none of the 600 randomly selected C 1 G 5 or G 1 C 5 -loop RNAs were specific for IRP-1, we identified four such sequences that preferentially bound to IRP-2. Each contained a double mutation of two loop bases (positions 2 and 4) predicted to be unpaired and accessible for protein contact. These findings are not restricted to the ferritin IRE, as fusion of an IRP-2-specific loop (mutant CG125: 5Ј-CCGAGC-3Ј) to the erythroid 5-aminolevulinic acid synthase IRE stem resulted in comparable selectivity for IRP-2, although the binding affinity was greatly reduced (data not shown). We have thus dissected the wild-type IRE loop and identified specific changes in paired or unpaired nucleotides that are tolerated by only IRP-1 or IRP-2. These findings reveal that the IRPs bind overlapping, but quite distinct, sets of RNA targets.
The list of regulatory RNA-binding proteins is growing rapidly, and several of these bind with specificity to defined RNA hairpin structures (reviewed in Varani and Pardi (1994), Burd and Dreyfuss (1994), and McCarthy and Kollmus (1995)). Aside from the IRPs and tRNA synthetases, there are relatively few sets of related proteins known to discriminate between similar RNA hairpin targets via differences in loop sequence. The eukaryotic splicing snRNP proteins U1A and U2BЉ are one example; they share strong sequence similarity and bind closely related U1/U2 small nuclear RNA stem-loops (reviewed by Nagai and Mattaj (1994)). Furthermore, the bacteriophage MS2 and GA coat proteins are 62% identical in sequence and FIG. 8. Iron regulation of human IRP-1 and IRP-2. Cytoplasmic extract was prepared from Molt-4, HL60, and HeLa cells, untreated in log-phase (L) or treated for 20 h with 100 M desferrioxamine (D), or 20 h of desferrioxamine followed by washing and 4 h of treatment with 60 g/ml ferric ammonium citrate (Fe). The effect of pretreating HeLa cells with 10 g/ml cycloheximide 15 min before iron treatment was also tested (Fe/C). 2 g of each extract was mixed with 6 ϫ 10 4 cpm of gel-purified mutant GG1 (IRP-1-specific) and mutant CG125 (IRP-2specific) RNA probes in the presence of 300 ng yeast tRNA, 5 mg/ml heparin, and the presence (ϩ) or absence (Ϫ) of 1% 2-ME, and complexes analyzed by band-shift. Autoradiographs were exposed for 24 h (IRP-1 gel) or 48 h (IRP-2 gel). Gel bands were quantitated and the relative intensities plotted in the lower panel, after normalizing values to account for differences in binding affinity. This experiment was performed twice with similar results. act as translation repressors that recognize similar RNA hairpins differing primarily in the 4-base loop sequence (see Lim et al. (1994)). Discernment of the different RNA loop sequences by MS2 and GA coat proteins was attributed to a single amino acid residue (Lim et al., 1994). Thus, while the relatedness of IRP-1 and IRP-2 (Rouault et al., 1992) may account for their shared capacity to bind a wild-type IRE, it should prove at least equally as interesting to learn the amino acid differences which decide the individual specificities of the two IRPs.
In our view, intraloop base pairing induces a conformational change within the IRE loop, such that the intervening three bases (positions 2, 3, and 4; see Fig. 2A for numbering) are made accessible for contact with IRP-1 and IRP-2 (model outlined by Henderson et al. (1994)). We propose that the 3 unpaired loop bases splay outward to form hydrogen bonds with specific IRP amino acids, as was shown previously for other RNA⅐protein interactions including that between the tRNA Gln and tRNA Asp anticodon loops and their respective glutaminyland aspartyl-tRNA synthetases (Rould et al., 1991;Cavarelli et al., 1993), and recently for the U1 small nuclear RNA stem-loop and U1A spliceosomal protein (Oubridge et al., 1994). The distinction in binding by the IRPs arises following base substitution at IRE loop positions 2 and 4, which is clearly far better tolerated by IRP-2. It is interesting that IRP-1 is more tolerant of different 1:5 base pair combinations in the IRE loop. Perhaps after initial binding, IRP-1 is better able to fold the IRE into a conformation favoring the 1:5 loop interaction, thereby enabling base pairs with lower energy to form. This explains the ability of IRP-1, but not IRP-2, to bind IRE-like hairpins with less stable non-Watson-Crick base pairs previously observed in other RNAs (Gutell, 1993;Henderson et al., 1994). Similar reasoning predicts RNA hairpins with C 1 G 5 and G 1 C 5 -type loops to be the most stable, possibly even forming spontaneously in solution, and thus facilitating easier recognition by IRP-2. Indeed, the wild-type IRE loop may form a C 1 :G 5 base pair in solution (Sierzputowska-Gracz et al., 1994), and it will be interesting to learn if alternate loop base pairs also form prior to RNA interaction with the IRPs. Ultimately, refined structural analyses of the IRP⅐RNA complexes are required to confirm these proposals.
Coordinate Iron Regulation of IRP-1 and IRP-2 in Human Cells-A novel and practical application was found for the selected RNA sequences, which were used as 32 P-labeled probes to distinguish human IRP-1 from IRP-2. Unlike the rodent IRPs, human IRP-1 and IRP-2 are of similar charge, and their RNA binding activities cannot be distinguished by bandshift assay using conventional IRE probes. This has precluded any detailed analysis of individual endogenous human IRP activities; as a consequence, earlier studies that examined human IRE binding activity actually detected a combination of the two IRPs. Recent investigations have managed either to detect recombinant human IRP-2 expressed following transfection into human cells (Samaniego et al., 1994) or to supershift human IRP-2⅐IRE complexes with specific antibodies (Guo et al., 1995a).
In this study, we demonstrated that IRE-like hairpins specific for mouse IRP-1 or IRP-2 showed similar selectivity for the human IRPs. Despite an increase in background caused by a lower binding affinity of the IRP-2-specific probes, the availability of different IRP-2-specific and non-binding sequences enables a controlled assay for human IRP-2 activity. A survey of human cell lines revealed different IRP expression patterns, and as in most rodent cell lines (Henderson and Kü hn, 1995), IRP-1 activity was always highest. In fact, the human IRPs displayed most of the previously defined characteristics that distinguish rodent IRP-1 and IRP-2. Both of the human pro-teins were regulated by iron, but only IRP-1 activity was recovered in extracts from iron-treated cells by in vitro reduction with 2-ME. We recently reported that iron-mediated inactivation of mouse IRP-2, but not IRP-1, was translation-dependent (Henderson and Kü hn, 1995). This difference in the iron regulatory pathways of the two IRPs also applies in human cells. These findings establish several aspects of IRP regulation that are well conserved and describe a general assay to detect individual human IRP RNA binding activities.
Growing Complexity in IRP Regulation and Function-IRP-1 and IRP-2 are related proteins that share several common features; they each bind strongly to "wild-type" IREs conserved phylogenetically in all known IRE-containing mRNAs (Theil, 1994), their RNA binding activities are coordinately modulated by iron levels  this study), and they inhibit translation of IRE-containing transcripts in vitro (Guo et al., 1994;Kim et al., 1995). Recent work, however, has uncovered some interesting differences between the two proteins. For instance, while iron inactivates both proteins, only IRP-2 is selectively degraded (Guo et al., 1994;Samaniego et al., 1994), and by a translation-dependent mechanism (Henderson and Kü hn, 1995), suggesting that IRP-2 levels are independently controlled by another protein. Moreover, IRP-2 responds preferentially to certain stimuli, and in rat liver cells IRP-2 activity increases during tissue regeneration (Cairo and Pietrangelo, 1994), but decreases following oxidative stress in vivo . These differences in IRP gene or protein regulation may reflect in the localization of these proteins. Expression patterns of IRP mRNA (Samaniego et al., 1994), protein (Guo et al., 1995a) and RNA binding activities  vary among different tissues. The cell type-specific expression reported in this study further implies that the IRPs might function at different cellular locations or stages of development. Our current findings add to this complexity, by showing that each IRP has a distinct RNA-binding specificity, which is conserved between species. This implies that each IRP may regulate its own distinct set of mRNA targets in vivo. We are currently searching the nucleic acid data bases to identify such mRNA target sequences, in the hope that this might further elucidate the function of these proteins, and potentially link iron metabolism with other cellular processes.