The Internal Ribosome Entry Site (IRES) Contained within the RNA-binding Motif Protein 3 (Rbm3) mRNA Is Composed of Functionally Distinct Elements*

Although the internal ribosome entry sites (IRESes) of viral mRNAs are highly structured and comprise several hundred nucleotides, there is a variety of evidence indicating that very short nucleotide sequences, both naturally occurring and synthetic, can similarly mediate internal initiation of translation. In this study, we performed deletion and mutational analyses of an IRES contained within the 720-nucleotide (nt) 5′ leader of the Rbm3 mRNA and demonstrated that this IRES is highly modular, with at least 9 discrete cis-acting sequences. These cis-acting sequences include a 22-nt IRES module, a 10-nt enhancer, and 2 inhibitory sequences. The 22-nt sequence was shown to function as an IRES when tested in isolation, and we demonstrated that it did not enhance translation by functioning as a transcriptional promoter, enhancer, or splice site. The activities of all 4 cis-acting sequences were further confirmed by their mutation in the context of the full IRES. Interestingly, one of the inhibitory cis-acting sequences is contained within an upstream open reading frame (uORF), and its activity seems to be masked by translation of this uORF. Binding studies revealed that all 4 cis-acting sequences could bind specifically to distinct cytoplasmic proteins. In addition, the 22-nt IRES module was shown to bind specifically to 40 S ribosomal subunits. The results demonstrate that different types of cis-acting sequences mediate or modulate translation of the Rbm3 mRNA and suggest that one of the IRES modules contained within the 5′ leader facilitates translation initiation by binding directly to 40 S ribosomal subunits.

Translation of eukaryotic mRNAs begins with recruitment of the translation machinery at either the 5Ј-m 7 G cap structure or an internal ribosome entry site (IRES) 1 (reviewed in Refs. 1 and 2). The cap-dependent mechanism for initiating translation is generally thought to be more common; however, the number of mRNAs reported to initiate translation internally is growing (3). Internal initiation seems to facilitate the transla-tion of particular viral and cellular mRNAs under conditions or subcellular locations that render the cap-dependent mechanism less efficient, for example, during the G 2 /M phase of the cell cycle, under conditions of mild hypothermia, or in dendrites (4 -7). Although IRESes are defined by using operational criteria, such as the ability to enhance second cistron translation when tested in the intercistronic region of a dicistronic mRNA (8), various lines of evidence indicate that they comprise a diverse group of sequences that may recruit the translation machinery by a variety of mechanisms (1,2).
Accumulated data also suggest that there are intrinsic differences between viral and cellular IRESes. Picornavirus IRESes, for example, seem to be highly structured, well defined functional entities of up to several hundred nucleotides. These IRESes have been categorized on the basis of sequence and structural similarities, and some of these features have been shown to be functionally important (9). In contrast, cellular IRESes do not seem to contain conserved secondary structures, and their sequences are not obviously similar to each other or to viral IRESes.
A feature noted in some cellular but not in viral IRESes is that they seem to be composed of shorter elements that can function independently when tested in isolation (10 -12). The modular nature of some cellular IRESes is suggested by numerous observations. For example, it has been difficult to define discrete 5Ј and 3Ј boundaries for particular cellular IRESes. Moreover, the activity of some cellular IRESes seems to be contained within two or more nonoverlapping fragments (e.g. Refs. [13][14][15][16][17]. In our analysis of the Gtx 5Ј leader, we identified four nonoverlapping fragments that functioned as IRESes (10), and the activity of one of these fragments was contained within a 9-nt segment. Other short nucleotide sequences that function as IRESes include two cis-acting sequences selected from libraries of random nucleotides (18) and short GA-rich repeats, which were shown to function as IRESes in both plants and animals (19).
The 9-nt Gtx IRES module is complementary to a segment of 18 S rRNA, and we previously showed that this sequence could bind directly to 40 S ribosomal subunits by base pairing to its rRNA complement (20). In other studies, we noted that many cellular mRNAs contain segments with complementarity to rRNA (21) and showed that numerous short complementary matches also occur in both cellular and synthetic IRESes (6,11,18). Other studies showed that complementary mRNA sequences could bind to ribosomes and affect the translation of the host mRNA (e.g. Refs. 22 and 23). These observations prompted the ribosome filter hypothesis, which postulates that the ribosomal subunits themselves are regulatory elements that modulate patterns of protein expression by reducing the translation of some mRNAs while enhancing that of others (12). In addition to recruiting ribosomes directly, we expect that some IRES elements might recruit ribosomes indirectly by binding to initiation factors or other trans-factors.
In a previous study, we noted that a 720-nt 5Ј leader of the cold stress-induced Rbm3 mRNA, which encodes a putative RNA-binding protein, contains 13 uORFs (6). Although uORFs should block translation of the main ORF (24,25), translation of a reporter mRNA containing the Rbm3 5Ј leader was relatively efficient. The identification of an IRES within this 5Ј leader suggested a mechanism by which the uORFs were bypassed. In our initial characterization of this IRES, its activity was confirmed by using a number of criteria, both in cells and in cell-free lysates. The notion that short nucleotide motifs may be key elements controlling the activity of some cellular IRESes is further developed in the present study.

MATERIALS AND METHODS
DNA Constructs-Reporter constructs were based on vectors that use the simian virus 40 (SV40) promoter to transcribe a dicistronic mRNA that encodes Renilla and Photinus luciferases as the first and second cistrons, respectively. These vectors were kindly provided by Dr. Anne E. Willis (14), who subsequently renamed them pRF and phpRF (hairpin). For consistency with our previous publications, we use the alternative nomenclature of RP and RPh, respectively. Fragments of the Rbm3 5Ј leader were generated by PCR amplification using Pfu DNA polymerase with either the full-length Rbm3 5Ј leader or oligonucleotides as templates and cloned into the intercistronic region of the RP or RPh vectors. Amplification oligonucleotides for cloning into RP contained EcoRI and NcoI restriction sites, whereas oligonucleotide templates for cloning into RPh contained SpeI and EcoRI. The 5Ј-nucleotides flanking the Rbm3 sequences in the 5Ј deletion series are ACUAGU. In some deletions, the 5Ј-nucleotide of the Rbm3 sequence is U and overlaps with the 3Ј-nucleotide from the 5Ј-flanking sequence. Oligonucleotides containing various combinations of the AclI, HindIII, and NcoI restriction sites were used to introduce mutations into the full-length Rbm3 5Ј leader by replacement of wild-type sequences. To generate promoterless dicistronic constructs, the SV40 promoter was excised using BglII and BlnI restriction sites. Particular fragments and mutations of the Rbm3 5Ј leader were also tested in a monocistronic context 5Ј of the Photinus luciferase cistron. mRNA levels in these constructs were estimated by using a second cistron which encodes a synthetic Renilla luciferase protein (RЈ, Promega) expressed via the IRES of the encephalomyocarditis virus (EMCV) (P(EMCV)RЈ). This construct was modified from the P(EMCV)-CAT construct (6) by replacement of the second cistron (CAT) with the synthetic Renilla luciferase.
Transfection and Cell-free Analyses-Reporter constructs (0.5 g) were transfected into cells (1 ϫ 10 5 ) using FuGENE 6 (Roche Applied Science). Cell lines were mouse neuroblastoma N2a, rat glial tumor C6, mouse fibroblast NIH 3T3 (3T3), and human neuroblastoma SK-N-SH (SK). Transfection efficiencies were normalized by cotransfection with 0.2 g of a LacZ reporter gene construct (pCMV␤, CLONTECH). Cells were harvested 24 h after transfection, and reporter gene activities were determined (10). Note that, in this paper, the Photinus:Renilla luciferase expression ratio is referred to as IRES activity, even for putative cis-acting sequences. For cell-free translation studies, capped dicistronic mRNAs were transcribed by using the mMESSAGE mMACHINE T7 transcription kit (Ambion, Austin, TX). Translation reactions were performed by using 0.5 g of mRNA in the presence or absence of 0.2 mM cap analogue (m 7 G(5Ј)ppp(5Ј)G, Roche Diagnostics) in C6 cell-free lysates (6,20). mRNA integrity and size after translation were determined by Northern blot analyses (26) using a Photinus luciferase riboprobe.
Electrophoretic Gel Mobility Shift and Nitrocellulose Filter Binding Analyses-RNA oligonucleotides (Dharmacon Research Inc.) were 5Јend-labeled using [␥-32 P]ATP with T4 polynucleotide kinase and tested in electrophoretic mobility gel shift assays based on the method previously described (27) using cytoplasmic extracts prepared from N2a, C6, or 3T3 cells (20). Labeled RNA oligonucleotides corresponding to cisacting sequences with complementarity to 18 S rRNA were tested for their ability to bind to puromycin-dissociated N2a ribosomes and purified C6 40 S ribosomal subunits (28) in nitrocellulose filter binding assays (based on Ref. 29) using 1.25 mM MgCl 2 in the binding buffer. Nonspecific competitor RNA (SI/SIII) was based on mouse ␤-globin 5Ј leader sequences and poly(A) (10). The relative amounts of probe retained on nitrocellulose membranes was quantified with a Phosphor-Imager (Amersham Biosciences).

RESULTS
The Rbm3 IRES Contains Numerous Segments That Function Independently and Contribute to Overall Activity-To determine whether the Rbm3 IRES contains a discrete 5Ј boundary, the full-length Rbm3 5Ј leader and sequential deletions of ϳ100 nucleotides from its 5Ј end were tested in dicistronic constructs in four cell lines (Fig. 1). In the cell lines tested, the IRES activity of the Rbm3 5Ј leader was up to ϳ50-fold over the background level of activity obtained with the parent construct RP. Sequential deletions of this 5Ј leader progressively reduced IRES activity and indicated that the Rbm3 IRES does not contain a discrete 5Ј boundary; this suggests that IRES activity is derived from the summation of a series of shorter functional elements that are distributed throughout the Rbm3 5Ј leader rather than from a single, well defined functional entity. This suggested modular composition was further investigated by arbitrarily fragmenting the 720-nt 5Ј leader into seven segments ( Fig. 1, I-VII) and testing these for IRES activity. Six of the fragments had IRES activity at least 2-fold over back- ground in at least one cell line. In addition, the activities of two of the fragments (V and VII) varied in the different cell lines. Both were active in SK cells and inactive in C6 and 3T3 cells; however, in N2a cells, fragment V was active and VII was inactive. Fragment III was the most active segment in all four cell lines, and in two cell lines (N2a and SK) it was more active than the full IRES.
In a previous publication, we showed that the 5Ј leader of the Rbm3 mRNA functioned as an IRES in a cell-free lysate and that the activities obtained were not caused by the production of monocistronic mRNAs corresponding to the second cistron (6). In this study, we noted that the activity of one of the 100-nt fragments (III) was greater than that of the full-length 5Ј leader in N2a and SK cells. We sought to fine-map the cisacting elements contained within this fragment; however, before proceeding, we performed an additional control. To ensure that the enhanced activity obtained with the construct containing fragment III was not caused by promoter activity leading to the production of monocistronic Photinus luciferase mRNAs, we deleted the SV40 promoter from the Rbm3(III)/RP construct ( Fig. 2A). After transfection into N2a cells, the activities of both cistrons were reduced to a background level compared with the activities obtained from the Rbm3(III)/RP construct containing the SV40 promoter, demonstrating that the activities obtained from fragment III were not caused by promoter activity.
To begin to identify cis-acting sequences within fragment III, it was deleted and fragmented at intervals of ϳ20 nucleotides (Fig. 2B). As in our original characterization of the Rbm3 IRES, we included a hairpin structure in the 5Ј leader of these constructs to minimize the contribution of the first cistron through reinitiation or leaky scanning. In these experiments, one of the 20-nt segments (III-g) had a level of IRES activity ϳ5-fold over the background level of activity obtained with the parent construct RPh, but its activity was enhanced dramatically by nucleotide sequences located immediately 3Ј of it (Fig. 2B, compare fragments III-g to III-c), even though the 3Ј sequences themselves (III-d) had no detectable IRES activity. These results may indicate that fragment III-g contains a truncated IRES module; alternatively, the nucleotides 3Ј of III-g contain an independent cis-acting sequence that enhances the activity of the IRES module.
A 100-nt Fragment Contains Four cis-Acting Sequences That Affect IRES Activity-Sequential deletion of fragment III in steps of 1-5 nucleotides in a 5Ј to 3Ј direction revealed two activities (Fig. 3A, constructs d1-d14). The first, observed with deletion d4, resulted in an almost complete loss of IRES activity, although activity was recovered by deletion of an additional 4 nucleotides. Deletion d4 destroyed the putative initiation codon of a 15-nt uORF; the effects of this uORF on translation are addressed later. The second activity, observed with deletion d8, resulted in diminished IRES activity and localized the 5Ј boundary of a putative IRES module to nucleotide U Ϫ286 . Deletion of the next five nucleotides (d10) decreased IRES activity further, and activity was completely lost with deletion d11 (nucleotide U Ϫ277 ).
The 3Ј boundary of the cis-acting sequence defined by the 5Ј deletions d8 -d11 was determined by mutating 3Ј sequences to adenosine. The use of mutations rather than deletions avoids any changes in activity related to the spacing of cis-acting sequences relative to the downstream cistron. Sequential 3Ј to 5Ј mutations of fragment d6 (Fig. 3A, m1-m12) localized the 3Ј boundary of a putative IRES module to nucleotide A Ϫ265 (m2), with a level of activity ϳ8-fold over background. We refer to the segment defined by nucleotides U Ϫ286 to A Ϫ265 as the putative 22-nt IRES module.
Mutation of fragment III also revealed two other activities. An inhibitory sequence is contained between nucleotides C Ϫ228 and U Ϫ242 , and mutation of these nucleotides increased IRES activity from ϳ40-fold to ϳ90-fold over background (Fig. 3A, compare d6 to m10). The 3Ј boundary of a second cis-acting sequence was localized 16 nucleotides 3Ј of the putative 22-nt IRES module to G Ϫ249 (m8). These 16 nucleotides enhanced IRES activity more than 9-fold (Fig. 3A, compare m2 to m8), even though they did not have detectable IRES activity by themselves (d14). To address whether these 16 nucleotides represent the 3Ј end of a putative IRES module or contain a discrete enhancer element, nucleotides located immediately 3Ј of the putative IRES module were sequentially mutated to adenosine (m13-m19). Mutation of the first 7 nucleotides (m13-m16) diminished IRES activity somewhat from ϳ40-fold to ϳ30-fold over background; however, activity was still enhanced relative to that of the putative IRES module (Fig. 3A, compare m16 to m2). These results suggest the existence of a discrete enhancer element with a 5Ј boundary at nucleotide G Ϫ257 (m16). To investigate this possibility further, the spacing between the putative IRES module and enhancer sequences was varied by using poly(A). Mutation of the 7 nucleotides between the putative IRES module and enhancer sequences (m20) resulted in a level of IRES activity comparable with that of m16, even though these constructs differ in their 3Ј se- quences: m20 contains a stretch of poly(A), and m16 contains Rbm3 sequences. When the putative IRES module was spaced 9, 11, or 19 nucleotides from the sequences with enhancer activity (m21-m23, respectively), IRES activity was still enhanced relative to the activity of the putative IRES module alone (m2 or m19), indicating that the IRES and enhancer activities are separable. However, these activities were not completely independent, as the enhancer did not function when tested 5Ј of the putative IRES module (data not shown). Therefore, we refer to this segment as the 3Ј enhancer element.

The 22-nt Putative IRES Module Functions
Independently-To control for the possibility that deletions and mutations generated an IRES element (for example, by juxtaposition of intercistronic and Rbm3 sequences), we tested the putative 22-nt IRES module in isolation and observed that this sequence also seemed to function as an IRES. Its activity was 115-fold over background in N2a cells (Fig. 3A, 22-nt), 86-fold in C6 cells, 95-fold in 3T3 cells, and 46-fold in SK cells. The activity of the 22-nt IRES module is up to ϳ15-fold greater than that of the EMCV IRES. In addition, these activities were greater than However, in some constructs, i.e. those beginning with U, the 5Ј-nucleotide is contributed by this flanking sequence. IRES activities are normalized to 1.0 for that obtained with the RPh construct. Horizontal lines indicate S.E. A, constructs d1-d14 are 5Ј deletions of fragment III encompassing nucleotides Ϫ307 to Ϫ228. Constructs m1-m19 contain mutations of constructs d6 and d7. In constructs m20 -m23, the putative IRES module and enhancer sequences are spaced 7-19 nucleotides apart with poly(A). This spacing is indicated as a numerical value to emphasize the exact spacing between the putative 22-nt IRES module and the putative enhancer sequence. Construct 22 nt is the putative IRES module tested in isolation. The full-length EMCV IRES is represented as a hatched bar. B, schematic representation of the putative cis-acting sequences contained within fragment III. those observed for the full-length IRES or those expected from the results of the deletion and mutational analyses. The results suggest that the activity of the putative 22-nt IRES module may be affected by flanking sequences or by its spacing relative to the Photinus luciferase initiation codon, both of which differ in these particular constructs. In summary, deletions and mutations of fragment III identified four cis-acting sequences that affected its translation, which together are schematically represented in Fig. 3B.
The 22-nt Putative IRES Module Is Confirmed to Be an IRES by Various Criteria-To validate that the 22-nt cis-acting sequence functions as an IRES, we excluded other possibilities that might yield similar results, i.e. the possibility that this sequence functions as a transcriptional promoter, enhancer, or splice site that leads to the production of monocistronic second cistron mRNAs. The possibility of promoter or enhancer activities was considered unlikely because the 22-nt cis-acting sequence accounts for the activity of fragment III, and fragment III was shown not to function as a promoter in Fig. 2A. However, to further investigate this possibility, we also showed that the activity of the 22-nt cis-acting sequence depended on the production of the dicistronic mRNA (data not shown). In addition, we tested this sequence in a cell-free translation reaction (Fig. 4), which also addressed the possibility that the putative 22-nt IRES module might generate monocistronic Photinus luciferase mRNAs by functioning as a splice site that results in excision of the first cistron. For these experiments, dicistronic mRNAs were transcribed and capped in vitro and translated in cell-free lysates that lack nuclear splicing factors. The 22-nt cis-acting sequence increased translation of the second cistron relative to a control mRNA (RP) lacking these sequences by ϳ70-fold. This translation was cap-independent because it was not blocked by the presence of a cap analogue, which decreased the translation of the first cistron by ϳ80% (Fig. 4A). These results also indicate that the 22-nt sequence does not facilitate translation of the second cistron by mechanisms that depend on translation of the first cistron (e.g. reinitiation). In fact, when translation of the first cistron was blocked, second cistron expression actually increased by ϳ1.8-fold. Nuclease hypersensitivity or splicing activities in these lysates were monitored by Northern analysis of the dicistronic reporter mRNAs after in vitro translation (Fig. 4B). The results showed no obvious differences in either mRNA integrity or size after 60 min of incubation. Taken together, the observations from the transfection and cell-free studies confirm that the 22-nt cis-acting sequence functions as an IRES.
Translation of a uORF Appears to Mask an Inhibitory cis-Acting Sequence-The presence of a fourth cis-sequence, located upstream of the IRES module, was suggested from deletion d4 (Fig. 3A). This deletion destroyed a putative initiation codon for a 15-nt uORF and led to an almost complete loss of IRES activity. However, deletion of the next 4 nucleotides (d5) restored activity, suggesting that translation of this uORF was not required for activity but rather masked an inhibitory sequence that overlaps the first 5 nucleotides. To investigate this possibility, this uORF was disrupted by mutating its initiation codon from AUG to UUG, CUG, AAG, or AUA (Fig. 5A, m24 -m27, respectively) and tested in the intercistronic region of a dicistronic mRNA in N2a cells. In this context, all four mutations inhibited IRES activity by ϳ60 -80%, consistent with the notion that translation of the uORF masks an inhibitory activity.
However, using dicistronic mRNAs, it is not possible to distinguish between a loss of IRES activity and an inhibition of translation, because the second cistron is translated at a background level in the absence of an IRES. To distinguish between these possibilities, we tested the IRES sequences in the 5Ј leader of a monocistronic mRNA, which is translated at a high level even in the absence of the IRES (Fig. 5B, m28 -m31). In this context, fragment III increased translation by ϳ30% relative to the parent construct. Mutations that disrupted the initiation codon of the uORF inhibited Photinus luciferase activity by up to ϳ40%, which was slightly less than that of the parent construct, suggesting that disruption of the uORF inhibits internal initiation but does not affect cap-dependent translation. A second potential initiation codon in fragment III at nucleotides Ϫ290 to Ϫ288 did not affect IRES activity (Fig.  3A, compare d5 and d6). This second uORF overlaps and is in the same reading frame as the Photinus luciferase cistron, indicating that it is either not recognized or results in a luciferase fusion protein, the activity of which is indistinguishable from the unfused enzyme.
In Different Cell Lines, the Four cis-Acting Sequences Contribute Differentially to the Activity of the Full-length Rbm3 IRES-Deletions and fragments of the Rbm3 5Ј leader indicated that it contains at least six IRES elements that can function in isolation (Fig. 1). To determine the contribution to overall IRES activity of the four cis-acting sequences identified in fragment III (Fig. 3B) and to address the possibility that the deletions and mutations used to define these elements may have generated novel cisacting elements, we mutated these sequences within the context of the full-length 5Ј leader (Fig. 6). The results of these mutations in N2a cells, the cell line in which these cis-acting sequences were first defined, were consistent with the hypothetically expected results. Mutation of the IRES module (m33) decreased IRES activity by up to ϳ40%, and mutation of the 3Ј enhancer (m34) had a small inhibitory effect. However, the simultaneous mutation of both the IRES module and 3Ј enhancer (m35) did not reduce activity more than mutation of the IRES module alone. Disruption of the uORF by an AUG to UUG mutation (m32) had only a small inhibitory effect in N2a cells, but mutation of the 3Ј inhibitor sequence (m36) increased IRES activity in all four cell lines. With the exception of the 3Ј inhibitory sequence, the activities of the other three cis-acting sequences seemed to vary in the other cell lines. For example, mutation of the IRES module decreased IRES activity in two of the cell lines but had almost no effect in SK cells. These results may either reflect the differential contribution of cis-acting sequences that are contained within other regions of the Rbm3 5Ј leader (see Fig. 1) or the differential activity of the four cis-acting sequences, perhaps because of differences in the expression of trans-factors in the various cell lines that affect the activities of these cis-acting sequences.
The Four cis-Acting Sequences Each Bind Specifically to Distinct Cytoplasmic Factors-To investigate potential interactions between the four Rbm3 cis-acting sequences and transfactors, each of the cis-acting sequences was tested for its ability to bind specifically to cytoplasmic proteins in an electrophoretic gel mobility shift assay. After incubation with cell extracts, an RNA probe corresponding to the 22-nt IRES module seemed to form ribonucleoprotein (RNP) complexes not observed in a reaction that contained the probe alone (Fig. 7A). One complex of similar mobility formed in cell lysates prepared from N2a, C6, or NIH3T3 cells, and a second complex formed in the N2a lysate. Binding specificity was assessed by competition with specific or nonspecific unlabeled RNA competitors corresponding to the probe itself or to a fragment containing ␤-globin and poly(A) sequences (SI/SIII (10)), respectively (Fig. 7B). The presence of an increasing concentration of the specific competitor resulted in a loss of formation of both RNPs. Similar concentrations of the nonspecific competitor failed to block RNP complex formation, indicating sequence-specific interactions. An RNA probe corresponding to the uORF formed diffusely migrating RNP complexes (Fig. 7A). The formation of these RNP complexes was blocked by the specific competitor and by the highest concentration of the nonspecific competitor (Fig. 7B). The uORF and 22-nt IRES module overlap by 10 nucleotides, and it is possible that the same proteins are contained within the lower molecular weight RNPs. The RNA probe corresponding to the 3Ј enhancer formed two RNP com-plexes with the three lysates (Fig. 7A); however, formation of only the higher molecular weight RNP was sequence-specific ( Fig. 7B and data not shown). Finally, an RNA probe corresponding to the 3Ј inhibitor specifically formed two RNP complexes in all three cell lysates (Fig. 7, A and B).
The 22-nt IRES Module Binds Ribosomes Directly-Computer searches identified complementary sequence matches to 18 S rRNA within the 22-nt IRES module and the uORF, raising the possibility of base-pairing interactions. To determine whether these cis-acting sequences bind ribosomes, RNA probes corresponding to them were incubated with puromycindissociated ribosomal subunits and filtered through nitrocellulose membrane. The nitrocellulose retains the RNA probe only if it is ribosome-bound. When tested, an RNA probe corresponding to the 22-nt IRES module bound to ribosomes and to purified 40 S ribosomal subunits (Fig. 8, A and B, respectively). However, no binding was observed for RNA probes corresponding to the uORF or a control sequence (SI/SIII, Fig. 8A). Binding of the IRES module probe to 40 S ribosomal subunits was specifically competed for by the unlabeled IRES module probe (ϳ80%) but was not affected by a similar molar concentration of an unlabeled control RNA (Fig. 8B). Ongoing studies will determine whether the complementary sequence match identified between the 22-nt IRES module and 18 S rRNA contributes to the binding observed. DISCUSSION cis-Acting sequences in mRNAs affect numerous post-transcriptional processes, including mRNA splicing, stability, methylation, and translation (reviewed in Ref. 30). One type of cis-acting sequence that has received increasing attention is the IRES. Cellular IRESes compose a diverse group of sequences, and although there are indications that some cellular IRESes might be modular, i.e. composed of shorter elements, little effort has been focused on defining these subcomponents. The possibility of IRES modularity was supported by our analysis of the Gtx IRES, in which a 9-nt IRES module was identified on the basis of its complementarity to 18 S rRNA (10). In addition, numerous other studies have shown modularity in IRESes from naturally occurring mRNAs (11)(12)(13)(14)(15)(16)(17), and selection studies have identified several short sequences (Ͻ18 nucleotides) that function as IRESes (18). In the present study, a functional approach was used to demonstrate the modularity of the Rbm3 IRES, and the results revealed the presence of at least three types of cis-acting sequences that directly mediate or affect IRES activity.
Little is known about the importance of RNA secondary and tertiary structures for cellular IRESes. However, the short size of the cis-acting sequences identified in this study is not consistent with RNA secondary structures of the types proposed to be contained within some viral and cellular IRE-Ses (31,32). Moreover, an RNA-folding algorithm (Mfold, version 3.1, Ref. 33) failed to identify any potentially stable secondary structures within the 22-nt IRES module. These and other results suggest the importance of primary sequence motifs in recruitment of the translation machinery (e.g. Refs. 10, 18, and 34). The identification of short IRES modules does not necessarily rule out the possibility that RNA secondary or tertiary structures exist or are critical for the activity of particular IRESes. In fact, higher order RNA structures may influence the presentation of primary sequence motifs and thus affect their accessibility for interaction with cytoplasmic factors or ribosomes. For example, in the analyses of the Apaf-1 IRES, for which RNA structures were experimentally determined, it was shown that disruption of particular RNA structures by either protein transfactors or mutation enhanced IRES activity (35). We expect that individual IRES elements might recruit ribosomes indirectly by binding to initiation factors or other trans-factors or directly by binding to ribosomal subunits. Sev-eral trans-factors have been shown to interact with and, in some cases, be functionally important for the activities of some cellular IRESes. For example, the polypyrimidine tract binding protein and the upstream of N-ras protein have been shown to bind to and enhance the activity of the Apaf-1 IRES (35).
The results of the present studies also raise the possibility that the 22-nt Rbm3 IRES module may recruit ribosomes directly. Other IRESes shown to bind ribosomes directly include those contained within hepatitis C and cricket paralysis virus RNAs (e.g. Refs. 36 -38). Inasmuch as a segment of the 22-nt IRES module has a complementary match to 18 S rRNA, a possible binding mechanism may involve base pairing between these sequences. This possibility is consistent with the site of complementarity, which is contained within an accessible region of 18 S rRNA at nucleotides 808 -819 (39). Based on cryo-electron microscopy analyses of yeast 80 S ribosomal complexes (40), this region of the 18 S rRNA seems to be located within expansion segment 6, which extends out from the solvent side of the ribosome platform. This interaction might A, gel mobility shift analysis of Rbm3 cis-acting sequence probes after incubation with extracts from NIH-3T3, C6, or N2a cell lines. RNA probes corresponding to the Rbm3 cis-acting sequences were electrophoresed alone (Probe) or after incubation in cell extracts. Sequences of RNA probes (uORF, IRES module, 3Ј enhancer, and 3Ј inhibitor) are indicated below autoradiograms. B, evaluation of specificity of RNP formation in N2a cell lysates by competition with specific (unlabeled probe) or nonspecific (SI/SIII) RNA competitors at up to a 1000-fold molar excess, as indicated. The two panels for the 3Ј inhibitor competition represent the two mobility shift bands seen in A.

FIG. 8. Binding of the Rbm3 22-nt IRES module to ribosomes.
A, nitrocellulose filter binding analysis of the 22-nt Rbm3 IRES module, uORF, and nonspecific RNA probes (SI/SIII) with an increasing amount of puromycin-dissociated ribosomes. B, binding and competition of binding between 22-nt IRES module and 40 S ribosomal subunits with a 100-fold molar excess of specific and nonspecific unlabeled competitor RNAs, the 22-nt IRES module, and SI/SIII RNAs, respectively. enhance translation initiation by increasing the local concentration of ribosomal subunits, as postulated by the ribosome filter hypothesis (12).
The Rbm3 IRES module was shown to bind specifically to a cytoplasmic factor (Fig. 7) and to 40 S ribosomal subunits (Fig.  8). Given that numerous ribosomal proteins have an extraribosomal representation (41), one possible explanation is that this cis-acting sequence binds to a ribosomal protein that is present both in ribosomal subunits and in the cytoplasm. Alternatively, the IRES module may bind to a cytoplasmic protein that either enhances or blocks the ability of this sequence to bind ribosomes. Additional experiments are required to determine whether any of these binding activities affect translation initiation.
The cis-acting sequences identified in this study contribute to the overall activity of the full-length 5Ј leader, but their contributions were found to vary in different cell lines. These results are consistent with the results of the deletion and fragment analyses, which identified other IRES elements distributed throughout the Rbm3 5Ј leader. The activities of two of these IRES elements, i.e. those contained within fragments V and VII (Fig. 1A), appeared to be restricted by cell type and may reflect the differential expression of trans-factors in these cell lines. Three of the cis-acting sequences identified in fragment III affected the activity of the 22-nt IRES module but did not have IRES activity themselves. Each of these cis-acting sequences was shown to bind specifically to cytoplasmic factors; ongoing studies are focused on identifying these trans-factors and evaluating the biological significance of these interactions for IRES activity.
For most mRNAs, uORFs inhibit the translation of downstream cistrons (24). This is not the case with the Rbm3 5Ј leader, which contains 13 uORFs (6). Indeed, the full-length 5Ј leader, which contains the most uORFs, has more IRES activity than 5Ј-deleted fragments that contain fewer uORFs. These results are unexpected because the 5Ј-most fragments are not particularly active as IRESes when tested in isolation (Fig. 1A). The presence of 13 upstream AUGs seems to be inconsistent with translation initiation by reinitiation or leaky scanning mechanisms (24). One possible explanation is that ribosomes recruited by some IRES modules detach and are re-recruited by binding to nearby IRES modules contained within the mRNA. This might resemble shunting, or it may be a clustering mechanism that increases the local concentration of ribosomes and increases the probability that a subunit will be proximal to the initiation codon, whereby it can either scan to it or bind directly to it by means of the initiator Met-tRNA that is a component of the ternary complex (12).
Although a uORF usually inhibits translation of a downstream ORF (24), one of the Rbm3 uORFs (Fig. 5) appeared to have the opposite effect. Other examples of stimulatory uORFs include the cauliflower mosaic virus and the yeast GCN4 mRNA, which facilitate shunting or reinitiation, respectively (24,25). There is no evidence that similar mechanisms operate in the context of the Rbm3 uORF, because deletion of part of the uORF restored activity, suggesting that its translation masks an inhibitory activity. Within the Rbm3 5Ј leader, the uORF overlaps the 22-nt IRES module and binds to cytoplasmic proteins. Binding of these proteins to the uORF may block IRES activity, and this binding may be blocked by the translation of this uORF.
In this study, we have identified short mRNA cis-acting sequences that affect translation initiation; however, the overall contribution to the proteome of such cis-acting sequences remains poorly understood. The identification of cis-acting sequences, both naturally occurring and synthetic, may provide reagents that will allow us to determine the underlying mechanisms by which these sequences influence translation initiation. The notion that short mRNA cis-acting sequences affect translation initiation seems to have functional parallels with the short DNA sequences contained within promoters that enhance or block transcription. This observation suggests that, like DNA transcription, mRNA translation can be controlled or fine-tuned by combinations of cis-acting sequences that mediate their effects by binding to trans-factors.