A second exon splicing silencer within human immunodeficiency virus type 1 tat exon 2 represses splicing of Tat mRNA and binds protein hnRNP H.

An equilibrium between spliced and unspliced primary transcripts is essential for retrovirus multiplication. This equilibrium is maintained by the presence of inefficient splice sites. The A3 3'-splice site of human immunodeficiency virus type I (HIV-1) is required for Tat mRNA production. The infrequent utilization of this splice site has been attributed to the presence of a suboptimal polypyrimidine tract and an exonic splicing silencer (ESS2) in tat exon 2 approximately 60 nucleotides downstream of 3'-splice site A3. Here, using site-directed mutagenesis followed by analysis of splicing in vitro and in HeLa cells, we show that the 5' extremity of tat exon 2 contains a second exonic splicing silencer (ESS2p), which acts to repress splice site A3. The inhibitory property of this exonic silencer was active when inserted downstream of another HIV-1 3'-splice site (A2). Protein hnRNP H binds to this inhibitory element, and two U-to-C substitutions within the ESS2p element cause a decreased hnRNP H affinity with a concomitant increase in splicing efficiency at 3'-splice site A3. This suggests that hnRNP H is directly involved in splicing inhibition. We propose that hnRNP H binds to the HIV-1 ESS2p element and competes with U2AF(35) for binding to the exon sequence flanking 3'-splice site A3. This binding results in the inhibition of splicing at 3'-splice site A3.

Because the unique transcript produced from the integrated proviral cDNA of retroviruses serves as the genome for newly synthesized virions and also for the production of mRNAs by alternative splicing, retrovirus multiplication depends upon an equilibrium between spliced and unspliced primary transcripts. To ensure this equilibrium, retroviral RNAs generally have splice sites that are used with low efficiencies. In human immunodeficiency virus type I (HIV-1), 3Ј-splice sites (3Јss) 1 and several central 3Јss (A3, A4a, A4b, A4c, and A5) compete with each other (Fig. 1A). Site A3 is required for production of tat mRNAs, sites A4a, b, and c for production of rev and env mRNAs and site A5 for production of nef and env mRNAs (Fig. 1A) (1). Metazoan 3Јss consist of three critical elements: the branchpoint sequence (2, 3), a polypyrimidine tract (PPT) sequence (4,5) and an AG dinucleotide at the 3Ј-end of the intron (for reviews, see Refs. 6 -9). HIV-1 branchpoint sequences are highly divergent in comparison to the metazoan consensus sequence (10 -12), and HIV-1 PPTs are suboptimal (short and interspersed by purines) (13)(14)(15). The affinity of factor U2AF for the PPT depends upon the presence of a long stretch of U residues (9,16). Factor U2AF consists of two proteins, U2AF 65 and U2AF 35 , with molecular weight of 65 and 35, respectively (17). Introns with suboptimal PPTs, like those in HIV-1 RNA, require binding of U2AF 35 at the intron-exon junction for stable interaction of U2AF 65 with the PPT (9,18,19). Furthermore, suboptimal 3Јss are often the subject of positive or negative regulation by cis-regulatory elements, which are frequently located in the 3Ј exon. Exonic splicing enhancers (ESEs) increase the utilization of upstream 3Јss by binding of nuclear components that favor the association of spliceosomal components. Several identified ESEs were found to bind serine arginine-rich proteins (20 -24). In contrast, exonic splicing silencers (ESSs) decrease the utilization of the upstream 3Јss. To date, the nuclear components that bind to the ESSs have been identified for only a limited number of ESSs: serine argininerich proteins for the bovine papillomavirus type-1 (25), heterogeneous nuclear ribonucleoprotein H (hnRNP H) for the rat ␤-tropomyosin exon 7 ESS (26) and the Rous sarcoma virus NRS (27), hnRNP A1 for the fibroblast growth factor receptor 2 K-SAM ESS (28) and for an ESS in exon v5 of CD44 (29). To date, one ESE (30,31) and three ESSs (15, 30 -33) have been identified in HIV-1 RNA. One of the ESSs (ESS2) is located within tat exon 2 and regulates the utilization of the A3 3Јss ( Fig. 1A) (14,30,32). The second ESS (ESS3) is located within tat-rev exon 3 and regulates the utilization of the A7 3Јss (15,30,31). The third ESS (ESSV) was recently discovered downstream from 3Ј-splice site A2 (33). HnRNP A/B proteins were found to selectively bind both the HIV-1 ESS2 and ESSV ele-ments, and this binding is necessary for the inhibitory properties of these ESS elements (33,34). The hnRNP proteins are a family of nuclear proteins that package nascent pre-messenger RNAs early after their transcription by RNA polymerase II (for review, see Ref. 35). HnRNPs are involved in several steps of mRNA production, including transcription regulation, modulation of alternative splicing, and mRNA stabilization and localization (for review, Ref. 35). HnRNP A1 was previously found to modulate the choice of alternative 5Јss (36 -40). Its involvement in inhibition of 3Јss utilization by binding to ESSs was found more recently, and, in this case, the mechanism of inhibition is still unknown. HnRNP H was found to bind both ESE (c-src ESE (41) and ESS (26)) elements.
We recently proposed a model for the secondary structure of the HIV-1 RNA region containing the five central 3Јss (A3, A4a, A4b, A4c, and A5) (42). Splice site A3 is contained in the terminal loop of a conserved stem-loop structure (SLS2), and its suboptimal PPT is in the helix. To complete this previous study, we looked for the interdependence between sequence and secondary structure at site A3 and generated mutations in both the PPT and the opposite strand of SLS2. Analysis of these mutants revealed the presence of a second ESS element acting to repress splicing at 3Ј-splice site A3 (ESS2p). Demonstration of the presence of this ESS2p element and the possible involvement of protein hnRNP H as a mediator of its inhibitory property are presented in this report.

EXPERIMENTAL PROCEDURES
Plasmid Construction-Plasmid pBRU3 (43) was used as the source of cDNA sequences from the HIV-1 BRU/LAI strain (GenBank TM accession number K02013). PCR amplifications were done according to Nour et al. (44). Three DNA fragments, H1, H2, and H3, corresponding to the HIV-1/BRU cDNA regions 1-385, 5172-5408, and 5172-5637, respectively (numbering according to ref. 45 Table I). All the constructs were made in plasmid pBluescriptKSII ϩ cleaved with the BamHI and PstI nucleases. PCR-amplified fragments were digested at restriction sites generated by the primers (see Table I).
In plasmid pLD-C3, fragments H1 and H2 have been ligated together using their BglII restriction sites. Similarly, in plasmid pLD-L3.U1, fragments H1 and H3 have been ligated together. Plasmid pLD-C2 contains the HIV-1/BRU RNA regions 1-385 and 4669 -5053 (10). Plasmid p⌬PSP, used for the transfection experiments, was constructed in the following way. Infectious HIV-1 plasmid pNL4-3 (GenBank TM accession number M19921) was cleaved with SpeI and BalI nucleases to generate an 11.9-kb fragment. This fragment was ligated together with oligonucleotides 5Ј-CTAGACGCGTTTGG-3Ј and 5Ј-CCAAACGCGT-3Ј, which had been previously annealed, to form a double-stranded linker. This created an HIV-1 plasmid deleted between nt 1511 and 4551.
Site-directed Mutagenesis-To generate plasmid pSJ-C3-1227, the inserted BamHI-PstI fragment of plasmid pLD-C3 was cloned into phage M13mp9 and site-directed mutagenesis was performed according to Kramer et al. (46), using oligonucleotide O-1227 (see Table I). The mutagenized BamHI-PstI DNA fragment was reinserted into plasmid pBluescriptKSII ϩ . All other site-directed mutagenesis analyses were performed by the PCR method, using the Stratagene QuikChange TM site-directed mutagenesis kit. The resulting plasmids are shown in Table I. p⌬PSP variants were constructed by replacement of the HIV-1/pNL4-3 EcoRI-Bsu36I fragment (positions 5743-5955) by the corresponding HIV-1/BRU EcoRI-Bsu36I fragment (positions 5325-5537) from plasmids pLD-L3.U1, pSJ-L3.U1-1227, or pSJ-L3.U1-1228. To create the p⌬PSP-1695 variant plasmid, p⌬PSP was cleaved with EcoRI and SalI nucleases and ligated together with oligonucleotides L3Asense and L3A-antisense (see Table I), which had been previously annealed, to form a double-stranded mutated linker with EcoRI and SalI sticky ends. To create variants p⌬PSP-1691, p⌬PSP-1693, and p⌬PSP-1606, the EcoRI/XhoI fragment from p⌬PSP WT was cloned in plasmid pBluescriptKSII ϩ cleaved by the same enzymes and the mutations were generated by the PCR method, using the appropriate primers (Table I). The EcoRI/BamHI fragments from these mutant constructs in pBluescriptKSII ϩ (nt 5743-8465) were ligated into p⌬PSP, which was cleaved with EcoRI and BamHI. All plasmids were sequenced to confirm the expected base changes or deletion.
Enzymatic Probing of RNA Secondary Structure-Synthesis of nonradioactive transcripts for RNA secondary structure analysis and enzymatic digestions with V1 RNase and S1 nuclease were performed as previously described (42). Positions of enzymatic cleavages were identified by primer extension analysis with the avian myeloblastosis virus reverse transcriptase (Life Science) using primer O-608 (Table I). Oligonucleotide primers were 5Ј-end-labeled with [␥-32 P]ATP 3000 Ci/ mmol (Amersham Pharmacia Biotech). Annealing of primers and primer extension was made as previously described (42).
In Vitro Splicing Assays-Prior to transcription with T7 RNA polymerase, the pLD-C3 and pLD-L3.U1 constructs and all their derivatives were linearized with the PstI nuclease, whereas pLD-C2 construct and all its derivatives were linearized with the EcoRI nuclease. For splicing assays, uniformly labeled capped transcripts were synthesized and in vitro splicing assays were performed with HeLa cell nuclear extracts from the Computer Cell Culture Center S.A. (Belgium), using 100,000 Cerenkov cpm (ϳ40 fmol) of RNA transcript per assay (10). The reaction mixture was prepared on ice and then incubated at 30°C for 120 min. Spliced products were deproteinized with proteinase K, phenol-extracted, and analyzed on a 5% polyacrylamide sequencing gel. Splicing efficiency was estimated by scanning the gel with a Molecular Dynamics PhosphorImager using ImageQuaNT software, version 3.3. The M/P ratio (amount of mature RNA versus the amount of residual precursor) was determined for each transcript, taking into account the estimated radioactivity and the number of uracil residues per molecule.
In Vivo Splicing Assays-HeLa cells were transfected by the modified calcium phosphate coprecipitation technique with 12 g of plasmid DNA as described above (47). Total cellular RNA was isolated from transfected HeLa cells 48 h post-transfection, and 3 g of RNA was reversed-transcribed and PCR-amplified with forward oligonucleotide primer BSS (5Ј-GGCTTGCTGAAGCGCGCACGGCAAGAGG-3Ј; nt 700 -727) and reverse primer SJ4.7A, which spans sites D4 and A7 (5Ј-TTGGGAGGTGGGTTGCTTTGATAGAG-3Ј; nt 8369 -8381 and 6032-6044) (47). After verification of the presence of amplified spliced products by PAGE, amplification products (100 ng) were radiolabeled by performing a single round of PCR with the addition of 10 Ci of [␣-32 P]dCTP, and the products were analyzed by electrophoresis on a 6% polyacrylamide 7 M urea gel.
Electrophoretic Mobility Shift Assays-DNA matrices were produced by PCR amplification using adapted primers (Table I), one of them generating a T7 or an SP6 promoter. Transcriptions were carried out on 0.9 pmol of PCR product in a 10-l reaction mixture containing 20 mM MgCl 2 , 10 mM NaCl, 40 mM Tris-HCl, pH 7.9, 10 mM DTT, 0.1 mg/ml bovine serum albumin, 10 units of RNasin, 4 mM each rATP, rCTP, rGTP, 0.4 mM UTP, 8 Ci of [␣-32 P]UTP (800 Ci/mmol) (ICN), and 70 units of T7 RNA polymerase (USB Pharmacia Biotech). Transcripts were then treated as previously described (10). For RNP complex formation, the following RNA binding mixture was used: 20 mM HEPES (pH 7.9), 3 mM MgCl 2 , 0.1 M KCl, 0.2 mM EDTA, 0.5 mM DTT, 0.25 mM phenylmethylsulfonyl fluoride, 20% glycerol, 1 g/l yeast tRNA, and 0.5-4 l of HeLa cell nuclear extract (Computer Cell Culture Center S.A., Belgium) in a total volume of 10.5 l. For competition experiments, the RNA binding mixture was preincubated for 15 min at 30°C with 20 -200 ng of unradiolabeled competitor RNA, synthesized from 1.8 pmol of PCR product with the T7-MEGAshortscript kit (Ambion). All EMSAs (with or without competitor RNA) were performed with 100,000 cpm (6 pmol) of 32 P-labeled transcript added to the RNA binding mixture. Incubation was performed for 15 min at 30°C, heparin (5 g/l) was then added as a nonspecific competitor, and the incubation was continued for 10 min at room temperature. To test for the effect of antibodies directed against individual hnRNP proteins on the stability and the electrophoretic mobility of the shifted RNA-protein complexes, before the heparin treatment, 1 l of a polyclonal anti-hnRNP H antiserum directed against an hnRNP H peptide located at the C terminus of the molecule (a generous gift of D. Black, University of California) (41) or 1 l of the anti-hnRNP A1 monoclonal antibody 4B10 (a generous gift of G. Dreyfuss, University of Pennsylvania) were added and incubation was continued for 15 min at 30°C. RNP complexes were then fractionated onto a 6% polyacrylamide (38:2) gel with 5% (v/v) glycerol in 1 mM EDTA, 45 mM Tris borate (pH 8.3), and free and bound RNAs were visualized by autoradiography. The amount of shifted versus free RNA or shifted versus total RNA was evaluated with a Phos-phorImager using the ImageQuaNT software.
To test for the presence of protein hnRNP H or hnRNP A1 in the fractionated RNA-protein complexes, the gel slices containing the RNAprotein complexes were soaked in 10 l of SDS-PAGE loading buffer for 1.5 h at 37°C and boiled 5 min. Both the piece of gel and the eluate were loaded onto a 10% SDS-polyacrylamide gel (thickness of 1.5 mm). After 2 h of electrophoresis at 20 V/cm, the fractionated proteins were transferred to a Hybond C nitrocellulose membrane (Amersham Pharmacia Biotech) used for immunoblotting. The membrane was blocked with 5% nonfat milk in phosphate-buffered saline solution containing 0.1% Tween 20 (48) and probed with anti-hnRNP H or anti-hnRNP A1 antibodies. The bound antibodies were detected with peroxidase-conjugated anti-rabbit or anti-mouse IgG antibodies, respectively, and visualized by the Amersham Pharmacia Biotech ECL detection system.
UV Cross-linking Reactions, Immunoprecipitation, and Immunoblotting-Radiolabeled RNAs were produced by SP6 transcription using the suitable PCR products as matrices (Table I) . Incubation in nuclear extract was for 15 min at 30°C as for EMSA, except that 500,000 cpm (50 fmol) of radiolabeled transcripts was used. Formation of RNP com-plexes was verified by EMSA. Reaction mixtures containing RNP complexes were transferred to 96-well plates for irradiation at 4°C with 254-nm UV light 1 cm from the source for 10 min. RNA components were then directly digested in the 96-well plates by addition of 50 units of T1 RNase in each well, and incubation was for 30 min at 50°C. Protein G-Sepharose beads (20 l; Amersham Pharmacia Biotech) were precoated with 1 l of anti-hnRNP H antiserum or 1 l of anti-hnRNP A1 antibody 4B10 for 2 h at 4°C.
The digested cross-linked products were then incubated with the precoated beads for 2 h at 4°C in 400 l of immunoprecipitation buffer (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 0.1% (v/v) Nonidet P-40) containing 0.1 mg/ml bovine serum albumin. Subsequently, beads were washed three times with the immunoprecipitation buffer containing 0.25% Nonidet P-40. After centrifugation, the beads were resuspended in 20 l of SDS-PAGE loading buffer (80 mM Tris-HCl, pH 6.8, 3% SDS, 100 mM DTT, 10% glycerol, 0.1% bromphenol blue) and boiled for 5 min for elution of the immunoprecipitated proteins. The proteins were resolved by 10% SDS-PAGE. Each sample of the eluted proteins was divided in two parts: One part was fractionated on a gel used for autoradiography, and the amount of radiolabeled protein was estimated by PhosphorImager scanning. The other part of eluted proteins was fractionated on a gel used for immunoblotting. Immunoblotting with anti-hnRNP H or anti-hnRNP A1 antibodies was performed as described above. The intensities of the signals obtained after detection were estimated with a Bio-Rad Gel Doc apparatus, using Molecular Analyst software.

RESULTS
Modulation of the in Vitro Utilization of Site A3 by Its Downstream Sequence-To test for the effects of the sequence and the RNA secondary structure of the splice site A3 PPT on usage  of this splice site, RNA transcripts of the WT or variant pLD-C3 constructs (Fig. 1AII, C3 transcripts) were produced and used for in vitro splicing assays. WT and variant C3 transcripts contained the HIV-1/BRU RNA portion from positions 1 to 385 fused to the HIV-1/BRU RNA region from positions 5172 to 5408, plus 52 nt arising from the pBluescript vector at the 5Ј terminus (Fig. 1AII). Hence, D1 5Јss and the A3 3Јss are present in these transcripts, but the previously identified ESS2 element is absent (Fig. 1AII). Splicing efficiency of transcript C3 WT in a HeLa cell nuclear extract was relatively low as determined by the M/P ratio (0.03) between spliced (M) and unspliced RNA (P) (Fig. 1C). In agreement with previous results (14), after two A-to-U substitutions in the PPT of transcript C3 WT (variant C3-1227, Fig. 1AIII), the in vitro utilization of site A3 was increased by a factor of about 7.5 (M/P ϭ 0.22) (Fig. 1C). According to the secondary structure established for the HIV-1 RNA region containing site A3 (42), the two A-to-U substitutions in variant C3-1227 destabilized the SLS2 helix 2. This was verified experimentally by V1 and S1 nuclease probing, and the results are schematically represented in Fig. 1B.
To discriminate between sequence and secondary structure effects on splicing efficiency, two other mutant C3 RNAs (C3-1228 and C3-1695) were produced. In variant C3-1228, two U-to-A substitutions were introduced in the 3Ј-strand of helix 2, such that a helix with similar stability to that of helix 2 in the WT RNA was formed. As shown in Fig. 1B, formation of the expected helix 2 was demonstrated experimentally by use of V1 and S1 nucleases. Fig. 1C shows that no spliced RNA was detected after a 120-min incubation of RNA C3-1228 in a HeLa cell nuclear extract. Interestingly, this absence of splicing was also observed for variant C3-1695, where only the two U-to-A substitutions in the downstream sequence were introduced (Fig. 1, AIII and C). This suggested that the absence of splicing observed for variant C3-1228 was not due to the restoration of the WT RNA secondary structure but to the alteration of the sequence located downstream from site A3. Accordingly, the U-to-A substitution at position 5362 (variant 1691) was sufficient to abolish splicing (Fig. 1, AIII and C). In contrast, when the U residue was substituted for an A residue only at position 5366 (variant 1693), splicing efficiency was slightly higher as compared with WT RNA (M/P of 0.05 instead of 0.03). Finally, when the two U residues at positions 5362 and 5366 were substituted by C residues (variant 1606), splicing efficiency was markedly increased (M/P of 0.10 versus 0.03). The in vitro splicing efficiencies of the WT and variant C3 RNAs were tested several times, and reproducible results were obtained. These results suggested that splicing efficiency at site A3 is modulated by the sequence within tat exon 2 immediately downstream of splice site A3.
Because utilization of site A3 was previously shown to be negatively regulated by the ESS2 element located about 60 nt downstream from site A3 (14,32), it was important to test whether the effects of the A3 downstream sequence on site A3 efficiency were also observed in the presence of ESS2. For this purpose, the above mutations were also tested in plasmid pLD-L3.U1, which contains the entire tat exon 2 (Fig. 1AII). As found for the C3 series, mutations 1606 and 1693 increased in vitro splicing efficiency at site A3 in the L3 series, whereas mutations 1691 and 1695 drastically decreased splicing efficiency at this site (Fig. 1D). No significant differences in splicing at 3Ј-splice sites A4a, A4b, A4c, and A5 were detected for the various transcripts studied (Fig. 1D). Therefore, our results indicate that, both in the presence or the absence of ESS2, the utilization of site A3 in the in vitro splicing assays was modulated by the identity of its downstream sequence.
Site A3 Utilization Is Also Modulated by Its Downstream Sequence in an HIV-1 RNA Context-To verify that the results obtained for the WT and variant C3 and L3.U1 RNAs were not artifacts of in vitro splicing assays, we tested the effects of the mutations (1227, 1228, 1606, 1691, 1693, and 1695) on splicing of HIV-1 RNA in HeLa cells (42,47). For this purpose, we used a plasmid (p⌬PSP) (see "Experimental Procedures") that contains the HIV-1/pNL4-3 proviral genome deleted between nt 1511 and 4550 in the D1-A1 intron. This construct contains all the HIV-1 splicing sites, and the relative usage of these splice sites in cells transfected with this construct is similar to virusinfected T cells (1). 2 To test for the effect of substitutions in the A3 PPT and/or the downstream sequence, a fragment of ϳ200 nt containing splice site A3 was replaced by the corresponding HIV-1/BRU sequence. Either the WT (plasmid p⌬PSP-WT) or the variant HIV-1/BRU sequences (plasmids p⌬PSP-1227, p⌬PSP-1228, p⌬PSP-1606, p⌬PSP-1691, p⌬PSP-1693, or p⌬PSP-1695) were used. HeLa cells were transfected with the various p⌬PSP plasmids and mRNAs produced by splicing of the mini-HIV-1 primary transcript were analyzed by RT-PCR (47). As a consequence of the two A-to-U substitutions in the A3 PPT (plasmid p⌬PSP-1227), the relative yields of both Tat 1 and Tat 2 mRNAs, both of which are spliced at site A3, were strongly increased. On the other hand, the relative yields of nef mRNAs (e.g. Nef 3 and Nef 4) were decreased ( Fig. 2A). However, the increase of Tat 1 and Tat 2 was reduced for plasmid p⌬PSP-1228, which contained, in addition to the two A-to-U substitutions in the PPT, two U-to-A substitutions in the downstream sequence. Little or no product spliced at site A3 was detected for plasmid p⌬PSP-1695, which carried the two downstream U-to-A substitutions without optimization of the PPT. A single U-to-A substitution at position 5362 in the downstream sequence (plasmid p⌬PSP-1691) also decreased the yields of Tat 1 and Tat 2 mRNAs but to a lesser extent. As found by in vitro splicing assays, the U-to-A substitution at position 5366 (plasmid p⌬PSP-1693) slightly increased the level of Tat 1 and Tat 2 mRNAs, whereas the two U-to-C substitutions (plasmid p⌬PSP-1606) strongly increased the yields of Tat 1 and Tat 2 mRNAs with a corresponding decrease of nef and rev mRNAs (Fig. 2B). Thus, the effect of the downstream mutation in the presence of an optimized PPT was less deleterious in HeLa cells than in vitro. However, in the presence of the WT PPT, variations of site A3 efficiency as a consequence of downstream mutations were similar in vitro and in HeLa cells, demonstrating the importance of the downstream sequence on site A3 utilization in the context of HIV-1 RNA.
A 9-nt Sequence Located Downstream from Site A3 Has an ESS Activity-Because the U-to-A mutations downstream from site A3 strongly decreased splicing efficiency at this site, whereas U-to-C substitutions increased utilization of splice site A3, we determined whether the WT A3 downstream sequence contained an ESE or an ESS element. To answer this question, we generated a template (pSJ-C3⌬ESS2p) used to synthesize pre-mRNA deleted between nt 5360 and 5368 (Fig. 3A). As shown in Fig. 3B, this deletion resulted in an approximate 6-fold increase in in vitro splicing at site A3 (M/P ϭ 0.56 versus 0.09). This increase suggested that an exonic splicing silencer (ESS) was present downstream from site A3. This potential silencer was designated ESS2p (proximal ESS2) to distinguish it from the previously identified ESS2 element acting on site A3.
To demonstrate ESS activity of the ESS2p silencer element in a heterologous context, we inserted this element downstream from another HIV-1 3Јss, site A2 (pLD-C2 construct).  3A). In the variant C2ESS2p RNA, the distance between site A2 and the inserted ESS2p element was identical to that between site A3 and the ESS2p element in the WT A3 site context (Fig. 3A). As shown in Fig. 3B, insertion of the 9-nt ESS2p element decreased splicing at site A2 by a factor of ϳ5 (M/P ϭ 0.27 versus 0.05). Splicing was blocked before the first step of the reaction, because production of the first exon and the lariat intermediate were also strongly decreased (Fig. 3B). From these data, we concluded that the 9-nt sequence downstream from site A3 (nt 5360 -5368) was sufficient to inhibit splicing at site A2 and, therefore, that this sequence had the properties of a splicing silencer. (25,28,29,49), the interaction of nuclear components with ESS2p was investigated by electrophoretic mobility shift assays (EMSA). To this end, radiolabeled E2-C3 transcripts containing the WT or the mutated HIV-1/BRU RNA region from positions 5359 to 5408 (3Ј-exon of the C3 RNA) were produced by in vitro transcription with T7 or SP6 RNA polymerase (Fig.  4A). To limit nonspecific RNA-protein interactions, mobility shift experiments were performed in the presence of heparin and competitor tRNAs. After incubation in a nuclear extract, complexes were formed with the three T7 RNA polymerase transcripts tested (Fig. 4B). Similar results were obtained with the SP6 RNA polymerase transcripts (data not shown). One major complex (I) was formed with the E2-C3-WT and E2-C3-C (containing the two U-to-C substitutions) transcripts at all the tested concentrations of nuclear extract. This complex had a lower electrophoretic mobility compared with the major complex II that was formed with the E2-C3-A transcript. At higher nuclear extract concentrations, a diffuse band corresponding to complexes with lower electrophoretic mobilities than complex II was also detected for transcript E2-C3-A. These data suggested that distinct RNA-protein complexes were formed, depending on the nucleotide sequence of the ESS2p element.

U-to-A Substitutions in the ESS2p Element Modify Interaction with Nuclear Components-Because effects of ESSs have been shown to be mediated by the binding of nuclear proteins
As shown in Fig. 4B, E2-C3-WT and E2-C3-C transcripts on in parallel. Size markers whose molecular weights are given on the left of the panel were also run on the gel. Positions of the C3 transcript and its spliced products are indicated on the right of the panel. The M/P ratios were calculated by estimating the radioactivity with a PhosphorImager as described under "Experimental Procedures" and are given below the lanes. D, polyacrylamide gel electrophoresis of the in vitro splicing products of the WT and variant L3.U1 transcripts. Symbols are the same as in panel C. Multiply spliced products of HIV-1 RNA were amplified by RT-PCR using the oligonucleotide primers BSS and SJ4.7A. The PCR products were analyzed by polyacrylamide gel electrophoresis (see "Experimental Procedures"). A and B correspond to two different experiments where the RT-PCR products obtained with the WT construct were compared with those obtained for variant constructs, whose names are indicated above the lanes. Nomenclature of the RT-PCR products is according to Purcell and Martin (1).

FIG. 3. The 5-AUUGGGUGU-3 sequence downstream from site A3 has an ESS activity.
A, the WT SLS2 is shown. The positions of the A3 3Јss and of the ESS2p sequence are also shown. The C2 WT and C2-ESS2p transcripts are schematically represented (same legend as for the C3 WT transcript in Fig. 1A). The 5Ј-AUUGGGUGU-3Ј ESS2p sequence, that is deleted in transcript C3⌬ESS2p was inserted in transcript C2-ESS2p, as shown. B, polyacrylamide gel electrophoresis of the in vitro splicing products of the C3 WT and C3⌬ESS2p transcripts (left panel) and the C2 WT and C2-ESS2p transcripts (right panel). Untreated C3 WT and C2 WT transcripts were fractionated in parallel (symbols are the same as in Fig. 1C). the one hand and E2-C3-A transcript on the other hand formed complexes with different nuclear components. Indeed, transcript E2-C3-A did not compete with transcript E2-C3-WT for binding of the nuclear components involved in complex I formation (Fig. 4C). Similarly, the E2-C3-WT RNA did not compete with the E2-C3-A RNA for binding of the nuclear components involved in complex II formation. Only, the diffuse band corresponding to complexes with lower electrophoretic mobility than complex II disappeared upon increasing the amount of competitor E2-C3-WT RNA (Fig. 4C). In contrast, as shown in Fig. 4D, transcripts E2-C3-WT and E2-C3-C competed with each other for binding of the nuclear components involved in complex I formation. However, the affinity of the E2-C3-WT RNA for these nuclear components appeared to be significantly higher than for the E2-C3-C RNA. This was evidenced by the displacement of only ϳ25% of the bound E2-C3-WT RNA in the presence of 200 ng of competitor E2-C3-C RNA versus 55% of the bound RNA using the same amount of E2-C3-WT RNA competitor. In addition, only a fraction of the E2-C3-WT RNA was competed by the E2-C3-C RNA.
Together, the EMSA and competition data suggested that two different proteins or sets of proteins of the nuclear extract were binding to ESS2p depending on whether pyrimidines or A residues were present at positions 5362 and 5366. This may explain the higher inhibitory activity of ESS2p after the U-to-A substitution. The data also suggested that the same protein or set of proteins was binding to the WT ESS2p sequence and the ESS2p sequence with C substitutions at positions 5362 and 5366, with, however, substantial differences in affinity for this protein or set of proteins, which may explain the increase in splicing efficiency observed after U-to-C substitutions.
hnRNP H Binds to the C3 Exon 2-Interestingly, the sequence generated by the two U-to-A substitutions in ESS2p contained the winner binding sequence for hnRNP A1, selected from a pool of randomized RNA sequences (50) (Fig. 4A). On the other hand, the WT ESS2p sequence shows some homology with the ESS element found in the rat ␤-tropomyosin pre-mRNA, which binds hnRNP H (26) (Fig. 4A). Proteins hnRNP A1 and hnRNP H have molecular masses of 35 and 55 kDa, respectively. Hence, binding of hnRNP A1 to the E2-C3-A RNA might be expected to generate an RNP complex of higher electrophoretic mobility as compared with binding of protein hnRNP H to the E2-C3-WT or E2-C3-C RNAs. This could explain the difference of electrophoretic mobility of RNP complexes depending on the ESS2p sequence. To test for the involvement of hnRNP H and hnRNP A1 in complex I and II formation, we performed immunoprecipitation assays of the UV cross-linked proteins using specific antibodies directed against hnRNP H or hnRNP A1 proteins, respectively.
We had found that G residues added at the extremities of T7 RNA polymerase transcripts can generate additional hnRNP A1 binding sites of low affinity. Therefore, because T7 and SP6 transcripts gave similar results in EMSA, SP6 transcripts were used for immunoprecipitation assays. Because the ESS2p elements of the three RNAs studied (E2-C3-WT, E2-C3-C, and E2-C3-A) contain the same number of G residues, but variable numbers of A, U, and C residues, the E2-C3-WT, E2-C3-C, E2-C3-A, and E2-C3-⌬ESS2p RNAs were uniformly labeled by incorporation of [␣-32 P]GTP. Labeled RNAs were incubated in a HeLa cell nuclear extract followed by UV-cross-linking. After T1 RNase digestion, hnRNP A1 and hnRNP H were each immunoselected with specific antibodies bound to protein G-Sepharose beads (anti-hnRNP A1 antibody 4B10, generous gift of G. Dreyfuss and anti-hnRNP H antiserum, generous gift of D. Black). The presence of radiolabeled nucleotides crosslinked to hnRNP H or hnRNP A1 was detected by SDS-PAGE, followed by autoradiography (Fig. 5, A and B).
As shown in Fig. 5A, a high amount of cross-linked hnRNP H was detected for RNA E2-C3-WT. This amount was decreased by a factor of 3 for RNA E2-C3-C. Only trace amounts of labeled hnRNP H were detected for RNA E2-C3-A and RNA E2-C3⌬ESS2p. Transfer of proteins from polyacrylamide gel on a membrane followed by immunodetection with the specific anti-hnRNP H antibodies confirmed that equal amounts of hnRNP H were retained on Sepharose beads in all assays (Fig. 5A,  lower gel). Hence, the levels of hnRNP H radioactivity should reflect the affinities for hnRNP H of the various RNAs tested. Experiments were repeated several times, and reproducible results were obtained All the transcripts tested were found to bind hnRNP A1 (Fig.  5B). However, binding to transcript E2-C3-A was ϳ5-fold higher than that observed for transcripts E2-C3-WT, E2-C3-C, and E2-C3⌬ESS2p. This was in agreement with the generation of a consensus hnRNP A1 binding site upon U-to-A substitutions at positions 5362 and 5366 in ESS2p. Hence, there was a good correlation between the splicing inhibitory property of ESS2p and its high affinity for either hnRNP H or hnRNP A1 proteins.
The above experiments revealed that hnRNP H interacted with E2-C3-WT RNA and hnRNP A1 interacted with the E2-C3-A RNA after incubation of these RNAs in a nuclear extract. We then used two approaches to confirm the presence of protein hnRNP H in complex I and protein hnRNP A1 in complex II seen in Fig. 4B. First, effects of treatments of complexes I and II with antibodies directed against these proteins were tested (Fig. 6, A and B). After incubation in the presence of anti-hnRNP H antibodies, complex I disappeared and a complex of lower intensity and lower electrophoretic mobility was observed (indicated by an asterisk in Fig. 6A). In contrast, a large part of unmodified complex II was resistant to RNases in the serum and no supershift was detected (Fig. 6A). When the same kind of experiment was performed with the anti-hnRNP A1 antibody, complex I was resistant and only a trace amount of supershift was detected, whereas complex II disappeared with the appearance of a small amount of supershifted material (Fig. 6B). Results were in agreement with the idea that anti-hnRNP H antibodies bind to complex I and dissociate part of this complex. These antibodies did not bind to complex II. Similarly, the anti-hnRNP A1 antibody binds to complex II and dissociates part of this complex, whereas complex I remains undissociated upon incubation with this antibody. Hence, data in Fig. 6 support the hypothesis that protein hnRNP H is in complex I and protein hnRNP A1 is in complex II.
This hypothesis was also supported by direct analysis of proteins hnRNP H and hnRNP A1 in complexes I and II. In these experiments a two-step gel electrophoresis was used followed by immunoblotting with anti-hnRNP H (Fig. 7, B and D) or anti-hnRNP A1 (Fig. 7C) antibodies. As illustrated in Fig. 7, C and D, protein hnRNP A1 but not protein hnRNP H was detected in complex II formed with RNA E2-C3-A. The trace amounts of protein hnRNP A1 seen in the control lane may be due to the presence of endogenous RNA in the extract bound to this protein. Protein hnRNP H was detected in complexes I formed with the E2-C3-WT and E2-C3-C RNAs (Fig. 7B). As a control, the band of gel at the same level as complex I in the lane corresponding to the E2-C3⌬ESS2p RNA was analyzed. It contained no protein hnRNP H (Fig. 7B), which reinforced the idea that ESS2p is required to bind protein hnRNP H. The above immunoprecipitation data suggested a lower affinity of protein hnRNP H for E2-C3-C RNA as compared with E2-C3-WT RNA. Accordingly, a lower amount of protein(s) hnRNP H was detected for the E2-C3-C RNA compared with the E2-C3-WT RNA in the Western blot analysis illustrated in Fig. 7B. Two closely spaced bands were present when the total proteins of the nuclear extract (NE) or proteins in complex I were analyzed by the anti-hnRNP H antibodies. This is in contrast to the immunoprecipitation experiments, in which only one band was detected with anti-hnRNP H antibodies (Fig. 5A). However, protein heterogeneity may be masked in this case by the presence of cross-linked ribonucleotides. One possible explanation for the double band in Fig. 7B is the presence of isoforms   FIG. 5. Cross-linking of proteins hnRNP A1 and H to the WT and mutated C3 and C2 exon 2. A and B, the E2-C3-WT or variant RNAs and E2-C2-WT and variant RNA were incubated in 4 l of nuclear extract. After RNA-protein cross-linking at 254 nm as described under "Experimental Procedures" and digestion with T1 RNase, proteins hnRNP A1 and H were each immunoselected with a specific antibody bound to G-Sepharose beads (anti-hnRNP H antiserum, panel A upper gel, and anti-hnRNP A1 monoclonal antibody 4B10, panel B upper gel). The eluted material was fractionated by 10% SDS-PAGE, and radiolabeled proteins hnRNP H and A1 were detected by autoradiography. As a control of hnRNP H and A1 immunoselections, the proteins in the polyacrylamide gel were transferred to a nitrocellulose membrane, probed with anti-hnRNP H (panel A, lower gel) or anti-hnRNP A1 (panel B, lower gel) antibodies and followed by ECL detection. The ratios (a/b) of radiolabeled hnRNP protein (a) to total immunoselected hnRNP protein (b), indicated below the lanes, were calculated as described under "Experimental Procedures." of hnRNP H, which bind to E2-C3-WT and E2-C3-C RNAs as monomers or heterodimers. Based on the peptide used to produce the anti-hnRNP H antibodies, the homologous hnRNP F protein (51) should not be detected by these antibodies (41). The two detected proteins may simply differ by the degree of post-translational modification.
All the data obtained from EMSA, immunoprecipitation assays, supershift assays, and two-step gel electrophoresis were in good agreement and were consistent with the hypothesis that protein hnRNP H binds to the WT ESS2p and to the ESS2p mutant with 2C residues, whereas protein hnRNP A1 binds to the ESS2p mutant with 2A, explaining the difference of splicing efficiency.
The ESS2p Sequence Is Sufficient to Generate an hnRNP H Binding Site-As shown above, insertion of ESS2p downstream from site A2 limited site A2 utilization. It was important to verify that the insertion of ESS2p downstream from this site generated an hnRNP H binding site, which could explain the observed inhibition. To this end, UV-cross-linking and immunoselection experiments were performed as described above with the E2-C2-WT transcript corresponding to the 5Ј portion of the 3Ј-exon of the C2 RNA and with the same RNA portion where the 5Ј-AUUGGGUGU-3Ј ESS2p sequence had been inserted (transcript E2-C2-ESS2p, Fig. 5, A and B). In agreement with recent data of Bilodeau et al. (33), both the E2-C2-WT and E2-C2-ESS2p transcripts cross-linked at high levels to hnRNP A1 (Fig. 5B). Only trace amounts of hnRNP H were crosslinked with transcript E2-C2-WT, whereas a significant amount of cross-linked hnRNP H was observed for transcript E2-C2-ESS2p (Fig. 5A). Because the immunoblotting experiment confirmed that similar amounts of hnRNP H were immunoselected (Fig. 5A, lower gel), we concluded that insertion of the ESS2p sequence was sufficient to generate a binding site for hnRNP H. These data are consistent with the hypothesis that ESS2p inhibits splicing by binding to hnRNP H.

DISCUSSION
Highly Controlled Utilization of Site A3 in HIV-1 RNA-Utilization of the A3 3Јss of HIV-1 RNA is required for tat mRNA production. It has previously been shown that splicing at 3Ј-splice site A3 is repressed by the ESS2 cis-acting element Free RNA and RNA-protein complexes were fractionated by electrophoresis on a 6% non-denaturing gel (A). Bands corresponding to complex I and the band at the same level as complex I in the lane E2-C3⌬ESS2p (A) were excised and soaked in SDS-PAGE loading buffer as described under "Experimental Procedures" before loading on a 10% SDS-polyacrylamide gel. After 2 h of electrophoresis at 20 V/cm, the fractionated proteins were transferred to a Hybond C nitrocellulose membrane used for immunoblotting with hnRNP H antibodies (B). As a control, 0.5 l of nuclear extract (NE) was fractionated in parallel with proteins contained in the gel slices. The same kind of experiment was performed with RNA E2-C3-A. Proteins in the gel slice corresponding to complex II formed with 1 l of nuclear extract (E2-C3-A in A) and in a gel slice at the same level as complex II in a lane where no RNA was added in the nuclear extract (control lane in C and D) were excised and treated in SDS-PAGE buffer before the second step of electrophoresis. The fractionated proteins were transferred on a Hybond C nitrocellulose membrane used for immunoblotting, successively with anti-hnRNP A1 (C) and anti-hnRNP H (D) antibodies. Positions of proteins hnRNP H and A1 are indicated. (14,30,31). Here, we show that a second inhibitory element, ESS2p, also acts to repress splicing at site A3. Because ESS2p inhibits splicing in the absence of ESS2, it indicates that these two inhibitory elements act independently on site A3. To our knowledge, this is the first example of a 3Јss that is negatively regulated by two independent ESS elements. Why is there such stringent control over site A3 utilization? The A3 PPT is suboptimal compared with efficient 3Јss of cellular pre-mRNAs. However, splice site A3 contains a larger number of U residues and is less interrupted by purines compared with the PPTs of sites A4a, b, and c and A5. Although no ESS element has been shown to act on sites A4a, b, and c and A5, splice site A4b acts as a repressor of splice site A5, and removal of splice site A4b results in a large increase in splicing at the A5 site. It was proposed that the inhibition results from competition between factors binding to branch-site sequences and factors binding to the splice site A4b AG dinucleotide (12). The equilibrium for utilization of the competing five A3 to A5 sites may therefore depend upon both this competition for splicing factors and negative regulation of the most optimized splice site (A3) by ESS elements. In addition, the Tat protein encoded by mRNAs spliced at site A3 has been found to be cytotoxic or apoptotic to infected and uninfected cells (52)(53)(54)(55)(56)(57)(58)(59)(60). Thus to maintain efficient replication, it is possible that HIV-1 controls production of this toxic viral protein by maintaining tight control of tat mRNA production.
Implication Consistent with the hypothesis that binding of hnRNP H to ESS2p is responsible for the inhibitory property, splicing at site A3 in HIV-1 RNA was increased by a factor of about 2 in HeLa cells as the result of the two U-to-C substitutions. HnRNP A/B have been implicated in the regulation of HIV-1 splice sites A2 (33), A3 (through ESS2 (34) and A7 (13)). Our data are the first to implicate hnRNP H in the regulation of HIV-1 RNA splicing.
How does protein hnRNP H regulate site A3 utilization by binding to ESS2p? As mentioned above, introns with suboptimal 3Јss require the binding of the U2AF 35 subunit to the intron-exon boundary for a stable association of factor U2AF with the PPT (18,19). For such introns, U2AF 35 binding was found to encompass the terminal AG dinucleotide of the intron and about 10 nt at the exon flanking the 3Ј-splice site (18) (Fig.  8A). At site A3, the 10-nt sequence at the exon 5Ј extremity contains the ESS2p element. Thus, it is likely that factor U2AF 35 competes with hnRNP H for binding to the intron-exon junction (Fig. 8B). Such a mechanism could also explain the increased splicing efficiency observed after the two U-to-C substitutions in ESS2p. Indeed, these two mutations, which decrease affinity of hnRNP H for ESS2p, may favor U2AF 35 binding and therefore increase splicing efficiency (Fig. 8C). Interestingly, the ESS element of the rat ␤-tropomyosin pre-mRNA, which also binds hnRNP H, is also located only 5 nt downstream of the regulated 3Јss (26). Thus, the proposed mechanism for repression of splicing at site A3 may also explain the inhibitory property of the ␤-tropomyosin ESS.
According to the data presented in this paper, binding of hnRNP H to ESS2p may be complex: (i) two different proteins recognized by the anti-hnRNP H antibodies were detected in complex I (Fig. 7B) and in the present stage of the study we do not know whether they bind individually or as an heterodimer, binding of a dimer of proteins to E2-C3-WT and E2-C3-C RNA would be in agreement with the very low electrophoretic mobility of complex I, (ii) also we cannot exclude the possibility that two types of complex I, which differ by internal structural rearrangements of the partners without marked changes in the electrophoretic mobility, are formed with the WT ESS2p, because ESS2p with two U-to-C only displaced part of the complex I formed with WT ESS2p (Fig. 4D). Further experiments are underway to answer these questions.
Some Point Mutations in ESS2p Increase Silencer Activity-When the UUGGGU sequence of ESS2p was converted into UAGGGA or UAGGGU, splicing inhibition by ESS2p was strongly reinforced both in vitro and in transfected HeLa cells. We showed that ESS2p with the UAGGGA sequence binds hnRNP A1. This is in agreement with previous data on protein hnRNP A1 binding sites, because the UAGGGA sequence was selected as the hnRNP A1 winner binding site (50). The stronger inhibitory activity of the variant ESS2p element with the UAGGGA sequence is consistent with the high affinity of hnRNP A1 for the UAGGGA sequence (K D of 1 nM) (50) and with the higher abundance of protein hnRNP A1 in nuclei and nuclear extract compared with hnRNP H (for review, see Ref. 62). According to the model of inhibition proposed above, hnRNP A1 should compete strongly with factor U2AF 35 for binding to the ESS2p sequence with two U-to-A substitutions FIG. 8. Model for the inhibition of site A3 utilization by binding of hnRNP H or A1 to ESS2p. The intron is represented by a thin line, the PPT by a thick line, the exon by a black rectangle, and the ESS2p element by a white rectangle. The intron 3Ј-terminal AG dinucleotide is indicated. A, binding of the two U2AF subunits to the suboptimal 3Јss (18). B, model for splicing inhibition by WT ESS2p indicating that hnRNP H (may be as a dimer) and the U2AF 35 subunit compete for binding to the exon 5Ј extremity. The thickness of the arrows corresponds to the affinity of the proteins for ESS2p. C, competition between hnRNP H and U2AF 35 is maintained, although the binding of hnRNP H is less strong for ESS2p with U5362C and U5366C substitutions. D, competition takes place between hnRNP A1 and U2AF 35 with only the U5362A substitution or with both U5362A and U5366A substitutions. (Fig. 8D). Because the U-to-A substitution at position 5362 was also sufficient for a stronger inhibitory effect in HeLa cells, this suggests that this substitution is sufficient to convert the ESS2p binding site for hnRNP H into a binding site for hnRNP A1. In connection with our observation, a UAGG motif is present in the ESS element of the K-SAM exon of FGFR-2 pre-mRNA (63). This ESS element was found to bind protein hnRNP A1, and this protein is thought to be involved in the splicing inhibition mechanism (28). In contrast to the strong effect of the U-to-A substitution at position 5362, the U-to-A substitution at position 5366 alone did not increase inhibition by ESS2p. It appeared to have a slightly negative effect on splicing inhibition. This suggests that hnRNP H and/or hnRNP A1 binds to the UUGGGA sequence but with a slightly lower affinity compared with binding of hnRNP H to the WT UUGGGU sequence and hnRNP A1 to the UAGGGA sequence.
The ESS2p Sequence Required to Bind hnRNP H Is Strongly Conserved in HIV-1 Strains-Sequence comparison of 61 different strains of HIV-1 virus belonging to the M, N, or O groups of HIV-1 strains and of strains of the related group of SIV virus, revealed a strong conservation of the UUGG sequence of ESS2p in these various strains (64). The sequence found in the most studied members of each group of HIV-1 or SIVcpz strains are aligned in Fig. 9. As shown in the alignment, the UU*GG sequence at the 5Ј-end of ESS2p is strongly conserved even though U5362 (marked with an asterisk) corresponds to a wobble position in the Vpr coding sequence. Substitution of U5362 by a C or an A would not alter the coding capacity of the RNA. However, only a very limited number of strains have a U-to-C substitution at this position (5 among 61 examined). Furthermore, no U-to-A substitution occurred at this position, which would convert the hnRNP H binding site into an hnRNP A1 binding site. Hence, according to this sequence comparison, the capacity of ESS2p to bind hnRNP H is strictly conserved in all the strains compared, including the SIVcpzGAB virus. This is a strong indication that ESS2p is important for the propagation of HIV-1 and SIVcpz related viruses.
Interestingly, recent data of Bourara et al. (65) suggest that HIV-1 RNA may be edited in chronically infected human H9 cells. Among five reported editing sites, one is within ESS2p. The first G residue in the ESS2p sequence UUGGGU appears to be converted to an A residue. This conversion may abrogate binding of hnRNP H and therefore interfere with splicing regulation in HIV-1-infected cells.