Mapping of multiple RNA binding sites of human T-cell lymphotropic virus type I rex protein within 5'- and 3'-Rex response elements.

Interaction between the human T-cell lymphotropic virus type I Rex protein and viral transcripts in the nucleus is essential to the cytoplasmic appearance of unspliced and singly spliced viral RNA. Rex has been shown to mediate its function through direct interaction with a highly ordered secondary structure in the 3'-untranslated region of all human T-cell lymphotropic virus type I mRNAs termed the Rex response element (3'-RxRE). Part of the 3'-RxRE sequence is also present in the 5'-end of viral transcripts (5'-RxRE), and we demonstrate that Rex binds to this RNA with essentially the same affinity and specificity as to the 3'-RxRE. We have analyzed the secondary structures and binding sites of Rex within the 5'- and 3'-RxREs by enzymatic probing and chemical modification interference and show that multiple Rex molecules bind within a stem-loop, which is similarly structured in the two RxREs. Our experiments confirm the presence of a previously characterized Rex binding site but also identify a common motif within an extended region that comprises an additional Rex binding site. This suggests that Rex oligomerizes on the RxREs similarly to what has been observed for binding of the human immunodeficiency virus type 1 Rev protein to the Rev response element.

Interaction between the human T-cell lymphotropic virus type I Rex protein and viral transcripts in the nucleus is essential to the cytoplasmic appearance of unspliced and singly spliced viral RNA. Rex has been shown to mediate its function through direct interaction with a highly ordered secondary structure in the 3-untranslated region of all human T-cell lymphotropic virus type I mRNAs termed the Rex response element (3-RxRE). Part of the 3-RxRE sequence is also present in the 5-end of viral transcripts (5-RxRE), and we demonstrate that Rex binds to this RNA with essentially the same affinity and specificity as to the 3-RxRE. We have analyzed the secondary structures and binding sites of Rex within the 5-and 3-RxREs by enzymatic probing and chemical modification interference and show that multiple Rex molecules bind within a stem-loop, which is similarly structured in the two RxREs. Our experiments confirm the presence of a previously characterized Rex binding site but also identify a common motif within an extended region that comprises an additional Rex binding site. This suggests that Rex oligomerizes on the RxREs similarly to what has been observed for binding of the human immunodeficiency virus type 1 Rev protein to the Rev response element.
Human T-cell lymphotropic virus type I (HTLV-I) 1 is the etiologic agent of adult T-cell leukemia and of a chronic neurologic disorder known as tropical spastic paraparesis or HTLV-I-associated myelopathy (for a review, see Ref. 1). HTLV-II is closely related to HTLV-I when comparing genome structure and nucleotide sequence, but unlike HTLV-I, no certain correlation to diseases has been established. Like human immunodeficiency virus type 1 (HIV-1), HTLV-I (and -II) is a complex retrovirus and as such its replication cycle is divided into an early nonproductive and a late virion-producing stage (for a review, see Ref. 2). During the early stage of infection, the regulatory proteins Tax and Rex are expressed from doubly spliced mRNA. Tax up-regulates transcription of viral as well as a number of cellular genes (1). Rex acts at the posttranscrip-tional level to induce the appearance of unspliced and singly spliced viral mRNA in the cytoplasm (3)(4)(5). These RNA species encode the structural and enzymatic polyproteins Gag, Pol, and Env, and the 9-kilobase unspliced RNA serves in addition as the viral genome.
It is not yet fully clarified how HTLV-I Rex mediates the accumulation of incompletely spliced mRNAs in the cytoplasm. Recently, the leucine-rich activation domain of Rex has been shown to constitute a nuclear export signal (6 -8), suggesting that Rex directly activates the transport of intron containing RNAs into the cytoplasm. The activation domain of Rex is furthermore interchangeable with the HIV-1 Rev activation domain (9,10), which has also been characterized as a nuclear export signal (11)(12)(13). However, Rex may also directly inhibit splicing, as suggested by the observation that Rex alters the balance between unspliced and spliced RNA in the nucleus as well as in the cytoplasm of T-cells (14). In addition, it has been reported that Rex specifically prevents the unspliced RNA from degradation in the nucleus (14).
Rex function requires the presence of an RNA element, the Rex response element (RxRE), present in the 3Ј-untranslated region of all HTLV-I transcripts (15)(16)(17)(18). By computer modeling (15), mutational analysis (16,18), and RNase probing (17), this element has been shown to form a highly ordered secondary structure consisting of four stem-loops, A-D, and two stems, I and II (see Fig. 3B). Rex binds directly to the RxRE in vitro (19,20), and mutational analyses of the RxRE have demonstrated that Rex binding in vitro correlates with Rex function in vivo (18,21). Within the RxRE, the stem-loop D has been shown to be indispensable to Rex binding and function (16 -18, 21). In accordance with this, Rex binds the stem-loop D alone in vitro (22), and modification interference studies and in vitro selection have identified a two-bulge structure within the stemloop D as the primary binding site for Rex (22,23). However, positions outside the two-bulge structure are most likely involved in efficient Rex recognition, since this minimal element has very reduced Rex binding capacity in vitro and function in vivo (24).
As a consequence of the retroviral replication mechanism, part of the RxRE sequence is also present in the 5Ј-end of the viral transcripts. Therefore, we introduce the terms 5Ј-RxRE and 3Ј-RxRE for denoting the RxRE at the 5Ј-and 3Ј-ends of the HTLV-I RNA, respectively. Importantly, the sequence forming the stem-loop D is present at both ends of the unspliced transcript. However, since the stem-loop D of the 5Ј-RxRE also harbors the splice donor site used for generating the spliced mRNAs encoding Env, Tax, and Rex, these mRNAs probably do not contain a functional 5Ј-RxRE. Considering that Rex in part regulates the cytoplasmic appearance of unspliced mRNA species, a potential regulatory mechanism resides within the 5Ј-RxRE.
Insight into Rex activity also derives from studies of the related virus, HTLV-II. HTLV-I and -II Rex can functionally replace each other (15), and both viruses require the presence of a functional splice donor site in the 5Ј-RxRE for Rex function (25,26). HTLV-II Rex has further been shown to inhibit in vitro splicing of an RNA substrate containing most of the HTLV-II R and U5 region upstream of a heterologous intron and exon but only if the Rex binding site within the 5Ј-RxRE is intact (27). Furthermore, in HTLV-II both the 5Ј-RxRE (26,28,29) and the 3Ј-RxRE (30) have been shown to be functionally important. Since previous studies in HTLV-I have focused on the function of the 3Ј-RxRE, it remains an open question whether the 5Ј-RxRE plays any role in the replication cycle of HTLV-I.
In this report, we compare the RNA structures and Rex binding sites within the 5Ј-and 3Ј-RxRE in HTLV-I and conclude that both structures contain similarly positioned multiple Rex binding sites. In addition to a binding site previously defined, we identify a second Rex binding site, which comprises a homologous RNA structure motif.
The vectors for expression of glutathione S-transferase (GST)-tagged Rex and Rex M1 were made by insertion of a PCR fragment encoding Rex or Rex M1 into the EcoRI-BamHI site of pGEX/GEH. 2 Rex M1 is deficient in RxRE binding because of a substitution of three arginine residues in the N-terminal RNA binding domain with an aspartic acidleucine dipeptide (Ref. 21; data not shown). The PCR was done with either a 5Ј-GAGGATCCATGCCCAAGACCCGTCGGAGG-3Ј oligonucleotide (for Rex synthesis) or a 5Ј-GAGGATCCATGCCCAAGACA-GATCTGCCCCGCCG-3Ј oligonucleotide (for Rex M1 synthesis) as the forward primer and a 5Ј-GAGAATTCCCGTGGGGCAGGAGGGGC-CAGG-3Ј oligonucleotide as the reverse primer and either pBC12/CMV-Rex or pBC12/CMV-Rex M1 as templates. The resulting pGEX/GEH-Rex and pGEX/GEH-Rex M1 plasmids encode fusion proteins consisting of the GST protein at the N terminus followed by a recognition site for enterokinase, the Rex or Rex M1 protein, respectively, and finally a small domain of 10 residues enabling C-terminal radiolabeling of the fusion protein with heart muscle kinase.
His 6 -Rex and His 6 -Rex M1 vectors were made by PCR-amplifying the Rex encoding sequences of pGEX/GEH-Rex and pGEX/GEH-Rex M1, respectively, with the same forward primers as described above and a 5Ј-CCCAAGCTTCGTGGGGCAGGAGGGGCCAGG-3Ј-primer as reverse primer. The resultant PCR fragments were digested with BamHI and HindIII and ligated into the corresponding sites of the expression plasmid pDS-H6 (34). In this context, Rex (or Rex M1) is preceded by 6 histidine residues, allowing rapid purification. All constructs were verified by sequencing.
Expression and Purification of Protein-GST-Rex and GST-Rex M1 were expressed in Escherichia coli strain BL21 and purified essentially as described by Amersham Pharmacia Biotech. 500-ml bacterial cultures were induced at log phase with 1 mM isopropyl-1-thio-␤-D-galactopyranoside, and growth was continued for an additional 1-2 h. Bacteria were harvested by centrifugation and resuspended in 12.5 ml of ice-cold buffer A (20 mM HEPES/KOH, pH 7.6, 200 mM NaCl, 20% glycerol, 10 mM ␤-mercaptoethanol) containing 0.1 mM EDTA, 0.5 g/ml leupeptin, and 2 g/ml aprotinin. The resuspended bacteria were lysed by gentle sonication on ice, and Triton X-100 was added to a final concentration of 1%. Insoluble debris was removed by centrifugation, and the supernatant was incubated with a 250-l bed volume of glutathione-Sepharose 4B (Amersham Pharmacia Biotech) at 4°C with mild agitation for 45 min. Beads were collected by centrifugation and washed 5 times with 2.5 ml of buffer A. At this step, fusion proteins bound to beads were either stored as aliquots at Ϫ80°C in buffer A or eluted with reduced glutathione (20 mM glutathione, 100 mM Tris/HCl, pH 8.0, 120 mM NaCl) by gentle shaking at room temperature and stored at 4°C. Eluted proteins were analyzed by SDS-polyacrylamide gel electrophoresis and Bradford reactions to estimate the concentration and purity.
Electrophoretic Mobility Shift Assay-For each reaction, 10 fmol of radiolabeled RNA was renatured by incubation at 80°C for 5 min in 1 l of renaturation buffer (10 mM HEPES/KOH, pH 7.6, 100 mM KCl) followed by incubation at 37°C for 20 min. 0 -15 pmol of GST-or His 6 -tagged Rex or Rex M1 protein diluted in elution buffer was preincubated on ice for 10 min with 5 l of 2ϫ Rex binding buffer (20 mM HEPES/KOH, pH 7.6, 300 mM KCl, 4 mM MgCl 2 , 1 mM EGTA, 2 mM dithiothreitol, 20% glycerol, 0.64 g/l E. coli tRNA) and H 2 O to a total volume of 9 l before the addition of 1 l of renatured RNA. After incubation at 0°C for 10 min, the reactions were applied to 5 or 8% nondenaturing polyacrylamide gels containing 100 mM Tris borate, pH 8.3, 1 mM EDTA, and 3% glycerol, and run at 4°C followed by autoradiography. In the competition experiments, renatured unlabeled RNA was incubated with Rex prior to the addition of radiolabeled RNA.
RNase Probing and Primer Extension-One pmol of either 5Ј-RxRE or 3Ј-RxRE RNA was renatured as above in 4 l of renaturation buffer per reaction and incubated with approximately 50 pmol of wild type or mutant Rex protein in a total volume of 20 l of 1ϫ Rex binding buffer. After incubation on ice for 10 min, 2 l of water (as negative control), RNase T 1 (Amersham Pharmacia Biotech, 40 units/ml), T 2 (Life Technologies, Inc.; 5000 units/ml), V 1 (Amersham Pharmacia Biotech; 100 units/ml), or A (Sigma, 1 g/ml) was added. Following incubation on ice for an additional 20 min, the reactions were terminated by adding 130 l of 0.25 M sodium acetate (pH 5.6), 1 mM EDTA, and RNA was recovered by extraction with phenol twice and precipitated. Half of the RNA was annealed to ϳ0.02 pmol of either of two 5Ј-end labeled primers; RxRE P1 (5Ј-GTCAGGCAAAGCGTGGAG-3Ј) or RxRE P2 (5Ј-CCTAGACGGCGGACGCAG-3Ј). The primers were extended by avian myeloblastosis virus reverse transcriptase (Amersham Pharmacia Biotech; 2 units/reaction), and after precipitation the cDNAs were resolved in denaturing 6% polyacrylamide gels containing 8 M urea, 100 mM Tris borate, pH 8.3, and 1 mM EDTA. Quantification was done using a Molecular Dynamics PhosphorImager SF and Molecular Dynamics ImageQuant software. An RNA marker sequence was obtained by add-2 T. H. Jensen, unpublished results.
ing dideoxynucleotides to reverse transcription reactions with untreated RNA as template.
Chemical Modification Interference-One g of 3Ј-end labeled stemloop D RNA was modified with either diethyl pyrocarbonate (DEPC) or hydrazine/NaCl (Hz/NaCl) essentially as described by Peattie (36) and gel-purified. Approximately 0.5 pmol of renatured modified RNA was then incubated in 10 l of 1ϫ Rex binding buffer in the presence or absence of 25 pmol of GST-Rex or His 6 -Rex. After a 10-min incubation at 0°C, the reactions were loaded on an 8% native polyacrylamide gel containing 100 mM Tris borate, pH 8.3, 1 mM EDTA, and 3% glycerol and run at 4°C followed by autoradiography. Free RNA and two complexed pools of RNA from lanes with GST-Rex or His 6 -Rex were excised from the gel and extracted. Subsequently, the RNA was treated with aniline (36) and analyzed on an 8% denaturing polyacrylamide gel containing 8 M urea, 100 mM Tris borate, pH 8.3, and 1 mM EDTA.

RESULTS
Rex Binds Directly to the 5Ј-RxRE-It is well established that purified Rex protein binds directly to the 3Ј-RxRE, derived from the 3Ј-end of viral transcripts (19 -21). Part of the 3Ј-RxRE sequence is also present at the 5Ј-end of viral transcripts, and an in vitro transcript initiating from the proviral transcription initiation site (5Ј-RxRE; Fig. 1) was tested for Rex binding capacity. Two recombinant types of Rex protein containing either GST or His 6 tags were expressed and purified from E. coli and used for electrophoretic mobility shift assays. Initially, increasing amounts of GST-Rex was incubated with radiolabeled 3Ј-RxRE RNA in the presence of E. coli tRNA as nonspecific competitor and run on a nondenaturing gel ( Fig. 2A, lanes  1-8). At lower Rex concentrations, two distinct complexes are visualized ( Fig. 2A, lanes 2-4), the first one being formed with an estimated K d Ͻ 16 nM. At increasing concentrations of GST-Rex, gradually slower migrating complexes are retarded in the gel ( Fig. 2A, lanes 5-8), indicating either the existence of multiple Rex binding sites within the 3Ј-RxRE or protein-protein oligomerization. To investigate the specificity of the complex formation, we made competition experiments with increasing amounts of either sense or antisense 3Ј-RxRE ( Fig. 2A,  lanes 9 -14). Preincubation of GST-Rex with 125 nM unlabeled sense 3Ј-RxRE totally inhibited complex formation between GST-Rex and the radiolabeled probe, whereas antisense 3Ј-RxRE had no effect ( Fig. 2A, compare lanes 13 and 14 with lane  7). Purified GST protein alone had no detectable affinity for the 3Ј-RxRE (Fig. 2A, lane 15). A very similar result was obtained using the 5Ј-RxRE as probe (Fig. 2B). Thus, GST-Rex binds with similar affinity and specificity to both RxREs.
To increase the resolution of the higher order complexes containing Rex and the 5Ј-RxRE, the shorter protein His 6 -Rex was used in a similar mobility shift analysis. Four distinct complexes can be detected (denoted A-D in Fig. 2C), and the K d values for the first three complexes are roughly estimated to be ϳ10, ϳ50, and ϳ200 nM, respectively. Hence, both types of recombinant Rex bind the 5Ј-RxRE with similar affinity. The affinity of GST-Rex for the stem-loop D sequence alone ( Fig. 1) was then examined using the same conditions as for the 5Ј-and 3Ј-RxRE, resulting in formation of two distinct complexes denoted A and B (Fig. 2D). Again the association is specific as judged from the challenge with either sense or antisense 3Ј-RxRE RNA (lanes 8 -13), but the affinity is approximately 30-fold lower as compared with the RxREs (K d Ϸ 500 nM). Finally, a minimal RNA element, HB-13 (nucleotides 130 -141 joined to 160 -172), encompassing the nucleotides earlier shown to be critical to mediate Rex binding (22,23) displayed Ͼ100-fold lower affinity toward GST-Rex in gel mobility shift analysis as compared with the RxREs (data not shown).
Ribonuclease Probing of Rex Binding to the 3Ј-RxRE-The observation of multiple complexes in gel mobility shift assays suggests the existence of more than one Rex binding site within the 5Ј-and 3Ј-RxRE as is the case in HIV-1, where multiple Rev binding sites within the Rev response element (RRE) are required for Rev function (37)(38)(39). To map the higher order binding sites within the RxREs, we performed RNA footprinting analysis. In vitro transcribed 3Ј-RxRE was incubated with a 50-fold molar excess of GST-Rex to generate higher order complexes (corresponding approximately to the conditions used in Fig. 2A, lane 4). As a negative control, GST-Rex M1, which is deficient in RxRE binding (Ref. 21; data not shown) was included. The RNA was subsequently cleaved partially with ribonucleases A, T 1 , T 2 , and V 1 , and cleavage sites in the RNA were detected by primer extension. To keep the fraction of doubly cleaved molecules very low, conditions were chosen so that more than 50% of the RNA remained uncleaved. By employing two radiolabeled primers, complementary either to the very 3Ј-end of the 3Ј-RxRE (RxRE P1) or to positions 102-119 (RxRE P2), the sequence of the whole RxRE could be clearly resolved. The probing was repeated three times, yielding essentially the same result as shown in Fig. 3A.
Since the RNases have different reactivities toward singleand double-stranded RNA regions, the probing gives information about the structure of the RNA. The cleavage pattern was compared with published secondary structure models (16 -18), and a proposed secondary structure is depicted in Fig. 3B. RNases T 1 and T 2 cleave predominantly single-stranded RNA, and the majority of the cleavages obtained by these RNases, and in particular the most intense ones, were in loops or bulges or at the end of helices. Equivalently, with the double-stranded specific RNase V 1 , we mainly saw cleavages at base-paired nucleotides, although some unpaired positions adjacent to helices were also hit, which may reflect additional structure or stacking (40). Due to the tendency of RNase A to cut doublehelical regions in addition to single-stranded RNA, especially at the salt concentration used in these experiments (170 mM KCl, 30 -50 mM NaCl), the observed RNase A cleavages are not included in Fig. 3B.
Several positions became specifically protected against cleavage upon GST-Rex binding ( Fig. 3A; data summarized in Fig. 3C). Rex strongly affected cleavages in the A 121 -G 136 and U 167 -G 169 regions within the stem-loop D. Besides residues located in regions previously found to be important to Rex binding capacity (the regions G 130 -G 141 and C 161 -G 171 (21)(22)(23)), additional nucleotides (the A 121 -A 129 region) became protected upon interaction with Rex, suggesting the presence of multiple Rex binding sites. Investigation of the 5Ј-half of the 3Ј-RxRE with the primer RxRE P2 did not reveal any GST-Rexspecific RNase protections (data not shown). Probing the Secondary Structure of the 5Ј-RxRE-Although we found that Rex binds the 3Ј-RxRE and the 5Ј-RxRE equally well, it was uncertain whether the 5Ј-RxRE has the same secondary structure with respect to the important stem-loop D region. The 5Ј-RxRE does not contain the 33 nucleotides present in the 5Ј-end of the 3Ј-RxRE, where they take part in the formation of stem I and II. Consequently, the two RxREs must form different secondary structures, at least in the 5Ј-part of the molecules. To examine this, we made nuclease probing of the 5Ј-RxRE structure using RNases T 1 , T 2 , and V 1 under the same conditions as for the 3Ј-RxRE (Fig. 4A). The result from three independent experiments is summarized in Fig. 4B. Most importantly, we find that the cleavage pattern of the 110 -180 region is essentially identical in the two RxREs, which we interpret as the presence of a similar stem-loop D structure in both elements. From the cap site to position 50, the nuclease digestion pattern is also highly similar in the two RxREs (primary data not shown; compare Figs. 3B and 4B). In contrast, the regions 50 -110 and 180 -190 display significant differences between the two substrates. In particular, only in the 5Ј-RxRE were high reactivities of RNase V 1 at positions C 56 , G 57 , and G 68 and of RNase T 1 at position G 99 observed. Other positions within the 5Ј-RxRE showed diminished accessibility as compared with the 3Ј-RxRE, e.g. the T 1 and V 1 cleavages at positions G 73 and U 79 , respectively. Combining the nuclease probing data with computer modeling (RNAdraw version 1.1 and RNA Fold) suggests a secondary structure of the 5Ј-RxRE as shown in Fig. 4B. Comparing this with the proposed structure of the 3Ј-RxRE (Fig. 3B) implies that the stem-loops A, B, and D are similar in the two RxREs, whereas a differently structured stem-loop C is formed. Finally, the stems I and II found in the 3Ј-RxRE obviously cannot be formed within the 5Ј-RxRE, since one of the participating RNA strands is missing.
The Rex Binding Site in the 5Ј-RxRE Overlaps the Major Splice Donor Site-The observations that the important stemloop D of the 3Ј-RxRE is also present in the 5Ј-RxRE and that recombinant Rex binds the two RxREs equally well make it conceivable that Rex contacts the same positions in both RNAs. To test this possibility, we performed nuclease probing of His 6 -Rex binding to the 5Ј-RxRE using the same conditions as for  Fig. 2. The RNases T 1 , T 2 , A, and V 1 were used to analyze the RNA structure and positions in the RNA protected by Rex. Cleavages were monitored by primer extension using the primer RxRE P2 to assay the region from Ϫ33 to 80 and primer RxRE P1 to assay the region from 70 to 190. Since RNases T 1 and T 2 predominantly cleave single-stranded RNA and V 1 preferentially cuts in RNA helices, the cleavage pattern can be used to predict the secondary structure of the 3Ј-RxRE. A, autoradiogram of primer extension covering the region 100 -190 of the 3Ј-RxRE. RNases and the presence of either wild-type GST-Rex (wt) or GST-Rex M1 (M1) are indicated above the lanes. A plus sign indicates control reactions treated as the nuclease samples but in the absence of RNase. The A, G, C, and U lanes refer to the RNA sequence and were generated by dideoxynucleotide sequencing of the untreated 3Ј-RxRE RNA template. Bars on the left denote protected regions. Nucleotides are numbered at the right from the beginning of the R region of the long terminal repeats. The overall lower intensity seen in lane 6 as compared with lane 5 was carefully normalized prior to quantifying the degree of protection. B, schematic representation of the RNase accessibility imposed on the secondary structure of the 3Ј-RxRE. The stem-loops A-D and the stems I and II are indicated. Observed cleavages are indicated by open circles (RNase V 1 ), squares (RNase T 1 ), and triangles (RNase T 2 ), respectively. The number of symbols at each nucleotide reflects the relative intensity of cleavages. The border between U3 and R (U3/R) and the 5Ј-splice site (5ЈSS) used in the 5Ј-RxRE are marked by large arrows, and the sequence complementary to the U1 small nuclear RNA at the 5Ј-splice site is shown in italic type. Data from RNase A treatment are not included, since they are inaccurate with respect to structural prediction. C, summary of Rex protections of the 3Ј-RxRE. Positions protected from RNase V 1 (open circles), RNase A (filled circles), RNase T 1 (squares), and RNase T 2 (triangles) cleavage are indicated, whereas unaffected digestions are excluded. ϩ, ϩϩ, and ϩϩϩ represent 1.5-2-, 2-3-, and Ͼ3-fold protection, respectively. Protection was quantified using an PhosphorImager as the intensity of cleavage in the presence of Rex M1 divided by the intensity of cleavage in the presence of Rex. the 3Ј-RxRE (Fig. 4A). The result of three independent experiments is summarized in Fig. 4C. As can be seen, Rex protects essentially the same positions in the 5Ј-RxRE as in the 3Ј-RxRE, only with slight changes in intensities (compare Figs. 3C and 4C; the 5Ј-RxRE was not assayed with RNase A). Interestingly, in the 5Ј-RxRE the protected region overlaps the major splice donor site, which is located after G 118 , and which is used generating all spliced HTLV-I mRNAs. As with the 3Ј-RxRE, no effect on the RNase cleavage pattern upon Rex binding could be detected in the 5Ј-part of the 5Ј-RxRE using RxRE P2 as primer, even in the presence of a 200-fold molar excess of His 6 -Rex to RxRE RNA (data not shown).
Modification Interference Analysis of the Interaction between Rex and the Stem-loop D-To substantiate the nuclease probing results, we investigated which nucleotides within the stemloop D were critical to Rex binding by modification interference. 3Ј-end-labeled stem-loop D RNA was subjected to limited chemical modification with either A-and G-specific DEPC or Cand U-specific Hz/NaCl under denaturing conditions, gel-purified, and incubated with GST-Rex or His 6 -Rex. The reactions were run on a nondenaturing polyacrylamide gel to separate unbound stem-loop D and stem-loop D⅐Rex complexes. The conditions were chosen such that two complexes were formed, corresponding to complex A and B in Fig. 2D. The RNA was recovered from the gel and treated with aniline, which cleaves the RNA strand at modified residues. Comparing the intensities of cleavages of the unbound and the two complexed pools of RNAs shows which nucleotides cannot be modified without affecting the capacity to bind Rex (Fig. 5A). The differences in cleavages were quantified and are shown schematically in Fig.  5B for the GST-Rex experiment. In agreement with our RNase protection assays, modification of the positions G 123 , A 128 -C 131 , G 135 , and G 136 reduced the affinity of the stem-loop D to GST-Rex. Moreover, positions U 132 -A 134 , A 140 , A 142 , C 163 -C 165 , and A 170 -A 175 , which were either not detectable or unaffected in the RNase protection analyses, were important to GST-Rex binding in this assay. Modification of certain nucleotides interfered strongly with formation of both complex A and B (Fig. 5B,  closed symbols), whereas other positions showed stronger interference with B complex formation as compared with complex A (Fig. 5B, open symbols). The observed modification interferences are likely to reflect that GST-Rex is prevented from contacting the particular nucleotides, although we cannot rule out the possibility that some of the modifications disrupt an RNA structure important for GST-Rex binding. Modification interference experiments using His 6 -Rex gave similar results but with one noticeable difference (data not shown). Contrary to the negative interference observed at positions A 140 , A 142 , and C 163 with regard to GST-Rex binding, DEPC-modification of G 139 -A 142 , as well as Hz/NaCl modification of C161-C163, caused enhanced His 6 -Rex binding. It is possible that modification of these residues reduces the stability of the helical structure formed by G 139 -G 141 and C 161 -C 163 and that this only affects GST-Rex, which binds the RxRE with a slightly lower binding affinity than does His 6 -Rex. DISCUSSION The repetitive nature of the termini of the HTLV-I genome implicates that part of the RxRE in the 3Ј-long terminal repeat is also present within the 5Ј-long terminal repeat. In this report, we show that HTLV-I Rex binds the RxRE structures derived from both ends of the viral transcript with equal specificity and affinity. The appearance of multiple complexes in the mobility shift assays with the two RxREs suggested the presence of more than one Rex binding site within each RxRE, and nuclease probing and modification interference studies revealed an extended Rex binding region as compared with previous reports.
The two RxREs have similarly structured stem-loops A, B, and D, as judged from our nuclease probing, whereas a different stem-loop C is formed in the two RNAs. The stem-loop D sequence has previously been shown to bind Rex in vitro (22), albeit with reduced affinity (24). Using gel mobility shift analysis, we found that the affinity is reduced approximately 30fold as compared with the intact RxREs, suggesting that the overall structure of the RxRE is important to effective Rex association. This notion is also implicated by in vivo results showing that two tandem copies of the stem-loop D are required to confer Rex responsiveness (24). The absence of any protections against RNase cleavages in regions outside the stem-loop D, even in the presence of a large excess of Rex protein, suggests, however, that the adjacent structures primarily play an indirect role in Rex binding, probably through stabilization of the stem-loop D structure.
In accordance with the similarities in structure and Rex binding capacity, the ribonuclease probing of Rex binding to the 3Ј-and 5Ј-RxRE yielded similar protection patterns. Rex impeded cleavages in the regions encompassing A 121 -G 136 and U 167 -G 169 , of which only the G 130 -G 136 and U 167 -G 169 regions FIG. 5. Modification interference assay of Rex binding to the RxRE stem-loop D. 3Ј-End-labeled stem-loop D RNA was modified under denaturing conditions with either DEPC or Hz/NaCl and gelpurified. DEPC reacts with purines and preferentially adenosine, while Hz/NaCl removes cytosine bases slightly more efficiently than uridine bases. A, ϳ0.5 pmol of modified RNA was renatured and incubated with 25 pmol of GST-Rex followed by separation of free and two complexed pools of RNA on an 8% nondenaturing gel. After recovery from the gel, the RNA was treated with aniline and analyzed on an 8% denaturing gel. have previously been observed to be critical to Rex binding in a modification interference study (22) and by in vitro selection (23). The A 121 -A 129 region, which has not been demonstrated to be implicated in Rex binding earlier, was most strongly affected by Rex binding in our assay. Since the RNA footprinting was performed under conditions where at least two Rex molecules bind to the RxRE in a native gel, the extended footprint in the A 121 -A 129 region most likely represent a second Rex binding site. This also may not have been detected in either the modification interference study by Bogerd et al. (22) (since only one complex was analyzed) or the in vitro selection experiment by Baskerville and co-workers (23), since not all nucleotides affected in our analyses were randomized, and therefore they could not be assayed. Despite titration of Rex in the ribonuclease protection experiments, we only found Rex-specific protection in the stem-loop D of the RxRE, indicating that only this region contains Rex binding sites. The multiple Rex molecules bound in the higher order complexes observed at the highest Rex concentrations in the mobility shift analyses, are, therefore, most likely either bound in this region or complexed merely through protein-protein interactions. The latter possibility is supported by the observation that Rex is able to form oligomers in the absence of RxRE (41).
The existence of at least two distinct Rex binding sites within the stem-loop D is substantiated by our modification interference analysis. Using the entire stem-loop D as a probe, we compared unbound RNA to RNA from first-and second-order complexes, denoted A and B, respectively. These experiments supported our finding of two separate domains composed of G 123 /C 172 -U 174 and G 135 -G 136 /C 164 -C 165 that were equally important for formation of complex A. Nucleotides that displayed stronger interference with B complex formation as compared with complex A were mainly located between these regions (A 128 -A 134 and A 170 -G 171 ). Although we cannot exclude the possibility that some of the interference observed may occur at the level of RNA folding, the nuclease protection experiments suggest that both sites are occupied by Rex protein. Our data is most easily explained by the existence of at least two Rex binding sites within the stem-loop D, but they do not allow an exact determination of the number of Rex molecules in complexes A and B. The observed interference at two sites for complex A formation suggests that at least two Rex molecules bind cooperatively to multiple sites or that a single Rex molecule binds both sites in this complex. In agreement with this interpretation, we found that a minimal RNA, HB-13, containing only the distal binding site, binds Rex with about 5-and 100-fold lower affinity as compared with stem-loop D and RxRE, respectively. As indicated in Fig. 6, RNase protection and/or binding interference observed in this study coincide with two repeated structural motifs in the stem-loop D, containing two G-C base pairs flanking a bulge that contains a uridine as the 5Ј-residue (boxed in Fig. 6). One motif located at the U 137 -C 138 bulge was earlier demonstrated to be critical to Rex binding (22,23), while we suggest that the second motif formed by the nucleotides G 123 , U 124 , G 130 , C 172 , and C 173 constitutes a second Rex binding site not characterized previously. Between the two motifs, a similar motif is located (marked with a dotted box in Fig. 6). We have observed minor changes in RNase cleavage pattern upon Rex binding within this element as well; thus, it may represent a weaker additional Rex binding site. A secondary role for this bulge structure was suggested by Baskerville and co-workers (23) based on in vitro selection.
The importance of multiple Rex binding sites within the RxREs is reinforced by the previous observation, that Rex, when fused to a heterologous RNA binding domain, requires the presence of at least two copies of the corresponding target RNA sequence for minimal functionality, and a further induction is obtained using four copies (42). Moreover, alignment of 20 HTLV-I and three HTLV-II isolates and two isolates of the related simian T-cell leukemia virus type I shows no sequence variation with respect to the two putative Rex binding motifs, indicating a functional role of the nucleotides concerned (data not shown). The requirement of multiple binding sites resembles the functional necessity of HIV-1 Rev protein oligomerization on the RRE (37,39). However, the two viral systems seem to differ at several important points. While only one localized binding site is detected for Rev in the RRE by modification interference analyses (35,43), the RxREs appear to contain at least two distinct sites important for Rex binding. Moreover, only one copy of the RRE is present in the HIV-1 RNA genome, whereas two Rex binding stem-loop D structures most likely are formed in the full-length HTLV-I transcript. Finally, as a result of Rev oligomerization, the RRE harbors up to ϳ10 Rev binding sites that can be detected by gel mobility shift analysis (39), while we only observe binding of about four Rex molecules within a limited subregion of the RxREs. Thus, in the context of viral replication, the presence of two RxREs could potentially compensate for the lower number of Rex molecules bound to each RxRE as compared with the Rev-RRE situation.
The Rex binding region in the 5Ј-RxRE overlaps with the consensus sequence of the 5Ј-splice site used for generating all spliced viral mRNAs (Fig. 6). This raises the possibility that the regulated expression of spliced and unspliced RNA species in part may depend on a direct regulation of splicing by Rex binding to the 5Ј-RxRE. Although considerable evidence favors a model in which Rex and the equivalent HIV-1 Rev protein facilitate nuclear export of unspliced and singly spliced viral RNA (7,44,45), the two mechanisms are not mutually exclusive. A direct function of Rex in splicing is supported by several observations. First, the 5Ј-splice signal in the 5Ј-RxRE is re-FIG. 6. The Rex binding elements imposed on a putative secondary structural model of the primary viral transcript. The nascent transcript contains differently structured RxREs at the opposite ends, the 5Ј-RxRE and the 3Ј-RxRE, respectively. Within these elements an identical stem-loop D is formed, containing putative repeated Rex binding motifs (boxed nucleotides; see "Discussion" for details). In the 5Ј-RxRE, the Rex binding region overlaps with the splice donor consensus sequence indicated by boldface letters. ported to be essential to Rex responsiveness in vivo even in the absence of a downstream 3Ј-splice site (25). Second, Rex induces accumulation of unspliced viral RNA in the nucleus as well as in the cytoplasm of T-cells in an RxRE-dependent manner, concurrently with a decrease in the amount of spliced viral RNA (14). Third, the very homologous HTLV-II Rex protein can in vitro specifically inhibit an early splicing step of a pre-mRNA consisting of the HTLV-II 5Ј-RxRE positioned upstream of an adenovirus-derived 3Ј-splice site (27). Since the recognition of the 5Ј-splice site by splicing factors is an important step in committing pre-mRNA to splicing, it is possible that binding of Rex to the 5Ј-RxRE interferes with the recognition of the downstream RNA as an intron. Consequently, full-length RNA escapes nuclear retention and is exported to the cytoplasm, mediated by the nuclear export signal of Rex.