Double Hairpin Complexes Allow Accommodation of All Four Base Pairs in Triple Helices Containing Both DNA and RNA Strands*

We investigated the binding of an antisense oligode- oxynucleotide to a stem-loop structure corresponding to the mini-exon sequence of the protozoan parasite Leish- mania amazonensis. This oligomer was designed to an-neal to the single-stranded region adjacent to the bot- tom of the hairpin and to fold back on itself, giving rise to a “double-hairpin” complex that involved a local tri- plex. This imposed the recognition, by the third strand, of a “purine” strand containing 6 interspersed pyrimi- dines out of 15 nucleic acid bases. The sequence of the complementary oligonucleotide was derived from the so-called pyrimidine motif; the third strand of the anti-mini-exon oligomer was parallel to the purine strand of the target. Electrophoretic mobility shift assays and footprinting studies demonstrated that such an anti- sense oligomer was able to bind to both the DNA and RNA versions of the Leishmania hairpin. These double hairpin complexes allowed the formation at pH 6.0 of a triple-stranded structure, despite the presence of 4 A:T*G and 2 G:C*T triplets out of 15. prepared by in vitro transcription with T7 RNA polymerase as described previously (20). Electrophoretic Mobility Shift Assay— Radiolabeled mini-exon oligo- nucleotide 35La or 35rLa (0.1 (cid:109) M ) was incubated with a complementary oligonucleotide (2 (cid:109) M ) for 24 h at 4 °C in a 50 m M sodium acetate (pH 6.0), containing 10 m M magnesium acetate. The samples were then mixed (v/v) with the dye solution (10% glycerol, 0.05% bromphenol blue, 0.05% xylene cyanol) and loaded on a 15% nondenaturing polyacryl- amide gel. Overnight migration (4 °C) was achieved at 0.75 mA/cm in the above buffer. UV-monitored Melting Curves— Complementary oligonucleotides (0.5 (cid:109) were incubated for 14 h at 0 °C in a 50 m M sodium acetate buffer (pH 6.0) containing 30 m M magnesium acetate in a 1-cm path length quartz cuvette. The temperature was then increased at a rate of 0.5 °C/min, while the UV absorption was monitored at 260 and 320 nm on a UVIKON 940 spectrophotometer. Footprinting was performed by min Samples were then analyzed on a 20% polyacrylamide gel. S1 mapping experiment were carried out the enzyme (50 , and Oligonucleotides were incubated at 4 for 4 in of 200 units of S1 nuclease. After precipitation by the were analyzed on a 20% denaturing polyacrylamide gel.

The common feature of messenger RNAs from trypanosomatids is the presence at their 5Ј end of a short sequence named "mini-exon" acquired during maturation of premessenger RNAs through a trans-splicing mechanism (1). This mini-exon motif, which is absolutely required for translation, is therefore a very attractive target for the design of antisense oligonucleotides as anti-parasitic agents (2). Potentially, a single complementary sequence will prevent the synthesis of all parasitic proteins. Indeed, oligonucleotides targeted to the mini-exon sequence of Trypanosoma brucei (3,4) or of Leishmania amazonensis (5) were shown to inhibit in vitro translation in cellfree extracts. Moreover, an acridine-linked 9-mer, complementary to the 5Ј end of T. brucei mRNAs specifically killed cultured procyclic forms of this parasite in vitro (6). More recently, it was reported that an anti-mini-exon phosphorothioate 16-mer, either free or associated to low density lipoproteins, displayed leishmanicidal properties against amastigotes of L. amazonensis grown in murine macrophages (7,8).
It was previously shown that the mini-exon sequence of L. amazonensis could fold into a hairpin secondary structure (9), which weakened the binding of antisense oligonucleotides (10). Rather than competing with the intramolecular hairpin, we considered the possibility to bind an oligomer to the folded structure. This can be achieved in different ways: (i) binding to the stem via a triple-stranded structure (11) or to the loop, leading to half pseudoknot structure (12), (ii) using an oligonucleotide that bridges the single-stranded parts, upstream and downstream of the stem (13,14), or (iii) selecting oligomers from a random population that recognize the folded target (15)(16)(17). The resulting complexes might actually stabilize the hairpin and interfere with mRNA translation. Alternatively, we previously described a strategy that allows accommodation of a stem-loop structure into a so-called "double hairpin" complex (18,19). The antisense oligonucleotide forms a short Watson-Crick duplex with a single-stranded sequence at the bottom of the hairpin and then folds back to give rise to a triple-stranded structure with both this short duplex and the stem of the hairpin. This approach was demonstrated using a model stem-loop purposefully designed to promote a triple helical structure; an antisense oligopyrimidine was targeted to a hairpin made exclusively of purines on the 5Ј side and consequently of pyrimidines on the 3Ј side, thus leading to the formation of canonical T:A*T and C:G*C ϩ triplets (where the colon denotes Watson-Crick base pairing and the asterisk denotes Hoogsteen hydrogen bonding with the third strand). The resulting complex involved 16-base triplets in which the two pyrimidine strands were connected by a (T) 4 loop.
The formation of triple helices is restricted to homopurine homopyridine sequences (2). Unfortunately, nucleic acid bases are not appropriately distributed for triplex formation in the hairpin derived from the L. amazonensis mini-exon. Therefore we had to design an oligonucleotide that aimed at forming a double hairpin complex involving a triple-stranded structure with a target sequence comprising all four bases. In particular two Cs of two G-C pairs should be read by the third strand. We demonstrated that an antisense oligomer composed of the four bases was able to bind to either a DNA or a RNA folded hairpin corresponding to the L. amazonensis mini-exon sequence through the formation of G:C*T and A:T(U)*G triplets, in addition to the canonical ones.

MATERIALS AND METHODS
Oligonucleotide Synthesis-The oligonucleotides used throughout this study (see sequences in Fig. 1) were prepared "trityl on" using conventional phosphoramidite chemistry. They were purified in one step by reverse phase high pressure liquid chromatography; an acetonitrile gradient in a 100 mM triethylammonium acetate buffer (pH 7.0) was used for elution. Purity was evaluated by electrophoresis of radiolabeled oligonucleotides on a 20% polyacrylamide gel containing 7 M urea, using [␥-32 P]ATP (37.5 MBq/mmol from DuPont NEN). The oligoribonucleotide 35rLa was prepared by in vitro transcription with T7 RNA polymerase as described previously (20).
Electrophoretic Mobility Shift Assay-Radiolabeled mini-exon oligonucleotide 35La or 35rLa (0.1 M) was incubated with a complementary oligonucleotide (2 M) for 24 h at 4°C in a 50 mM sodium acetate (pH 6.0), containing 10 mM magnesium acetate. The samples were then mixed (v/v) with the dye solution (10% glycerol, 0.05% bromphenol blue, * This work was supported in part by the Direction des Recherches, Etudes et Techniques, and by the Pôle Médicament Aquitaine. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. ‡ Present address: EMBL, Gene Expression Programme, Meyerhofstrasse 1, 69117 Heidelberg, Germany. 0.05% xylene cyanol) and loaded on a 15% nondenaturing polyacrylamide gel. Overnight migration (4°C) was achieved at 0.75 mA/cm in the above buffer.
UV-monitored Melting Curves-Complementary oligonucleotides (0.5 M) were incubated for 14 h at 0°C in a 50 mM sodium acetate buffer (pH 6.0) containing 30 mM magnesium acetate in a 1-cm path length quartz cuvette. The temperature was then increased at a rate of 0.5°C/min, while the UV absorption was monitored at 260 and 320 nm on a UVIKON 940 spectrophotometer.
Footprinting Experiments-For chemical footprinting 32 P 5Ј end-labeled oligonucleotides were incubated for 14 h at 4°C in a 50 mM sodium acetate buffer (pH 6.0) containing 30 mM magnesium acetate, either in the absence or in the presence of the desired complementary oligonucleotide, to allow the formation of complexes prior to the addition of 22.8 mg/ml potassium permanganate. At the end of a 30-min incubation at ϩ4°C, samples were ethanol precipitated and dissolved in 1 M piperidine (20 l). Hydrolysis was performed by 30 min heating at 90°C. Samples were then analyzed on a 20% denaturing polyacrylamide gel. S1 mapping experiment were carried out in the enzyme buffer (50 mM sodium acetate, pH 5.7, containing 200 mM NaCl, 10 mM ZnCl 2 , and 5% glycerol). Oligonucleotides were incubated at 4°C for 4 h in the presence of 200 units of S1 nuclease. After precipitation by ethanol the samples were analyzed on a 20% denaturing polyacrylamide gel.
RNase H Digestion-Complexes formed with 35rLa RNA were characterized by digestion with Escherichia coli RNase H. After preincubating 32 P-labeled 35rLa (2 nM) with a complementary oligonucleotide (1 M) at 4°C for 24 h in a 50 mM sodium acetate buffer, 10 mM magnesium acetate, and 150 mM sodium chloride, RNase H (0.15 u/l) was added, and the reaction was allowed for 30 min at 10°C. Samples were then analyzed on a 20% denaturing polyacrylamide gel.

RESULTS AND DISCUSSION
We studied the interaction between a 35-nucleotide-long sequence, corresponding to the mini-exon region of L. amazonensis, which can fold into a hairpin structure, and several complementary oligomers (Fig. 1). The mini-exon oligonucleotide, available either as DNA (35La) or RNA (35rLa) was used as a target for a 29-mer (29DE), whose 5Ј end constituted an anchor complementary to the single-stranded part located at the bottom of the stem of the folded mini-exon, leading to a 10-base pair duplex. The sequence of the 3Ј part of 29DE was chosen to optimize the formation of a triple helical structure, 15 triplets long. Although the 5Ј end of 35La contained 6 pyrimidines, we considered it as the ''purine strand'' for the putative triple helical complex. We had to accommodate 6 purine-pyrimidine pairs, two G-Cs, and four A-Ts interspersed within pyrimidinepurine pairs. This was achieved by G reading A-T and T reading G-C, which led to the least disturbing triplets A:T*G (21-24) and G:C*T (22-24) for purine-pyrimidine inversions in DNA triple strands. This design resulted in parallel orientations of the target purine strand and of the hypothetical third strand of the anti-mini-exon sequence. We also synthesized a control 29-mer (29Cont) with the same 10-nucleotide anchor as 29DE in the 5Ј part but a different 3Ј sequence, which was chosen to minimize the number of canonical triplets (Fig. 1).
Complex between the DNA Mini-exon Sequence and 29DE-We first investigated the binding of oligonucleotides to the 35-mer 35La corresponding to the DNA version of the L. amazonensis mini-exon sequence using electrophoretic mobility shift assay. As previously reported (9), this oligomer migrates as two bands on a nondenaturing polyacrylamide gel, corresponding to the minor linear form and to the major folded form (Fig. 2a). Upon the addition of the oligonucleotide 29DE, a band of intermediate mobility appears to the expense of the two bands characteristic of 35La. This revealed the formation of a 35La-29DE complex. Neither the oligonucleotides 10Cont, nor 29Cont, which can form 10 base pairs with 35La, gave rise to such a distinct retarded band; no modification of the electrophoretic profile was seen in the presence of 10Cont (not shown), whereas a smear was observed with 29Cont, indicating a weaker association than that of 29DE. Therefore, the binding of the latter oligonucleotide to the target hairpin involves more than 10 Watson-Crick base pairs, suggesting a contribution of the 3Ј part of 29DE to the interaction between the two oligomers.
As previously reported (9), the UV-monitored melting of 35La displayed a cooperative transition related to the unfolding of the hairpin characterized by a T m value of 48°C under our experimental conditions (see ''Materials and Methods''). The addition of a stoechiometric amount of 29DE led to a broad transition, indicative of a loose complex with poorly cooperative interactions (not shown). A similar melting curve was obtained with the control 29-mer. Therefore T m measurement did not allow to discriminate between the two types of complexes, 35La-29DE and 35La-29Cont.
Footprinting Analysis of 35La-29DE Complex-We investigated the structure of the 35La-29DE complex with chemical (KMnO 4 ) or enzymatic probes (S1 nuclease). The complex was labeled either on the target (35La) or on the anti-mini-exon strand (29DE). Footprinting performed with potassium permanganate showed a high reactivity of thymines 9 -15 of 29DE in the complex, compared with 29DE alone (Fig. 3a). In contrast, T residues from positions 18 -26 were significantly less susceptible to modification. This indicates that the 3Ј part of the oligonucleotide 29DE is protected through interaction with the target, whereas the T stretch is highly accessible. This might correspond to a loop structure, as confirmed by S1 mapping. Whereas this nuclease specific for single-stranded nucleic acids gave an all or nothing pattern of cleavage for 29DE alone, this oligonucleotide was specifically cleaved in the region corresponding to nucleotides 12-15 in the presence of 35La, i.e. in the T stretch that was highly reactive to KMnO 4 (Fig. 3b). A similar experiment performed with the oligomer 29Cont did not show such a pattern; however, a slightly increased sensitivity of residues located 3Ј to the anchor region was observed, indicative of a weak 35La-29Cont complex, in agreement with melting experiments (Fig. 3c).
Drastic changes of the KMnO 4 reactivity pattern were also observed for 35La upon addition of 29DE (Fig. 4). A reduced sensitivity of the T residues was generally observed in the anchor region; whereas T VI and T VIII were fully protected, T X was still available for the reaction. This latter residue faced the 5Ј-terminal nucleotide of the anti-mini-exon oligomer 29DE. This means that the 10-base pair duplex was actually formed in the 35La-29DE complex, but despite potential cooperative in-teraction between the stem structure of 35La and the doublestranded anchor site, the junction was prone to transient opening. The reactivity pattern of T residues located in the stem of 35La was also informative; T XIV was protected, whereas the sensitivity of T XVI was exacerbated (Fig. 4). Assuming that the 3Ј part of 29DE constitutes the third strand of a triple-stranded complex, T XVI could be next to the triple strand-double strand junction. A previous study (18) performed with a model sequence, designed to generate a double hairpin complex, has shown hyper-reactivity of the G residue located at the duplextriplex junction, indicative of a distorted structure. A similar effect was also described for a linear triplex made from three independent strands (25,26). In the present case the junction coinciding with a bulge might potentiate the conformation change. An additional effect could also contribute as discussed below for 10Cont.
The oligomer 10Cont, able to form a perfect duplex in the anchor region of 35La, yielded a different pattern of cleavage. As expected, this oligonucleotide protected T VI and T VIII , (slightly less than 29DE), whereas T X pairing with the 5Јterminal A of 10Cont was still reactive (Fig. 4). But T XIV , located in the stem remained accessible, in contrast to what was observed in the presence of 29DE. Surprisingly, T XVI was more reactive in the presence than in the absence of 10Cont, although it is located 6nt away from the 5Ј end of the antimini-exon oligomer (Fig. 4). This might suggest a conformation change of the target hairpin upon hybridization of the 10-mer, as observed for other complexes involving stem-loop structures (53). S1 mapping of the target also revealed conformational changes upon binding of 29DE. In the absence of any added oligomer, 35La was cleaved by S1 nuclease at T XX , i.e. in the loop, and at T XXVI , which faces the bulge (Fig. 5). Both sites were protected in the presence of 29DE. Interestingly, the addition of 10Cont, which enhanced the reactivity of the bulged T to KMnO 4 , did not change the S1 sensitivity of the opposite strand at T XXVI . Neither was any change seen at T XX (Fig. 5).
Therefore, 35La-29DE complex involves definitely more than the formation of 10 base pairs in the anchor domain. From the footprinting assays whose results are summarized in Scheme 1, it can be described as a double hairpin structure similar to the one reported previously for a model sequence (18); the 5Ј end of 29DE is paired with the single-stranded region of 35La located at the bottom of the stem, whereas the 3Ј moiety of the antimini-exon sequence forms the third strand of a triple-stranded structure, the T residues 11-14 being a connecting loop.
Binding of 29DE to an RNA Target-Because in the antisense strategy RNA is the physiological target of complementary oligonucleotides, it was of interest to monitor the binding of 29DE to 35rLa, an RNA hairpin homologous to the 35La DNA. We first studied the formation of 35rLa-29DE complex by band shift assay on a nondenaturing polyacrylamide gel. The RNA hairpin migrates as two bands, the most prominent one corresponding to the folded form. The oligonucleotide 29DE induced the appearance of two bands, the major one moving between those corresponding to the linear and the folded forms of 35rLa (Fig. 2b). A weak band also appeared above the one corresponding to the linear 35rLa, suggesting the formation of a second type of complex that was not detected with the DNA target. No retarded bands were observed with either 10Cont or 29Cont (not shown). Therefore, as for the DNA target, the interaction between 35rLa and 29DE extends beyond a 10-base FIG. 4. Permanganate footprinting of 35La anti-mini-exon oligonucleotides complexes. The 32 P 5Ј end-labeled oligomer 35La was incubated at pH 6.0 with KMnO 4 in the absence or in the presence of oligonucleotides 10Cont and 29DE, as indicated at the top of the lanes. The 35La sequence is shown to the right.
FIG. 5. S1 mapping of 35La anti-mini-exon oligonucleotide complexes. 32 P 5Ј end-labeled 35La was incubated at pH 6.0 with S1 nuclease in the absence or in the presence of the oligomers indicated at the top of the lanes. The sequence of 35La is given to the right. Filled and open symbols correspond to increased and decreased sensitivity, respectively, in the complex compared with isolated sequences. KMnO 4 (E, q), S1 (૾, ૺ), and RNase H (å). The target is shown as RNA (35rLa), although footprinting experiments (KMnO 4 , S1) were performed on DNA (35La). pair duplex formed by the 5Ј end of 29DE and the complementary region of 35rLa.
Ribonuclease H is able to cleave an RNA strand hybridized to a complementary DNA strand, but it is unable to digest RNA involved in a canonical pyrimidine-purine.primidine triplestranded structure (27,28). 1 We used E. coli RNase H to probe the complexes formed by 35rLa with either 10Cont or 29DE. The 10-mer, which forms a regular double-stranded heteroduplex, induced the cleavage of the RNA hairpin from C v to T x , with a marked preferential triad, A VII /T VIII /A IX , at pH 6.0 as well as at pH 7.3 (Fig. 6). In contrast, these cleavage sites were no longer detected in the presence of 29DE at pH 6.0 (Fig. 6b), indicating that the 3Ј part of 29DE protected partially 35rLa from RNase H activity. This was likely related to the formation of a triple-stranded structure as raising the pH up to 7.3, conditions which are known to distabilize C:G*C ϩ triplets, restored the cleavage pattern characteristic of the 35rLa-10Cont double-strand (Fig. 6b). However, three unexpected cleavage sites were observed at pH 6.0 for the 35rLa-29DE complex, the two prominent ones being at C III and C V (Fig. 6a). Assuming the likely formation of a triple-stranded complex under these conditions, these positions correspond to G:C*T triplets. In DNA triple helices these triplets are known to have a low stability (22,23,24). It is tempting to speculate that the noncanonical dG:rC*dT triplets adopt a partly open or a distorded structure that is detected by the enzyme. But, very likely the enzyme does not locally displace the third strand because this would have led to a cleavage pattern of the anchor binding site similar to that observed with the 10-mer 10Cont.
Therefore, RNase H proved to be a very sensitive probe because the E. coli enzyme was able to recognize a single deoxyribo-ribo (D-R) pair (namely G-C "mismatched" with T) in a nonsubstrate surrounding, i.e. triple-stranded structures. Interestingly, human and bovine RNases HI have been shown to cleave in the vicinity of a single R-D pair comprised in doublestranded DNA (29,30). Moreover, as the experiments were performed at 10°C, it means that the nonperfect triplex was stable under these conditions or that the enzyme stabilized it. In any case, RNase H was still able to bind to the complex, despite the presence of the third strand in the major groove. This indicates that the enzyme interacts in the minor groove of the heteroduplex, as usually observed for proteins that do not display sequence specificity.
Conclusion-Triple helix formation can be achieved in two different modes termed the pyrimidine motif and the purine motif, depending on the third strand (31). Both are restricted to the recognition of duplexes made of purine in one strand and pyrimidine in the second one, leading to the formation of C:G*C ϩ and T:A*T triplets for the pyrimidine motif, of C:G*G and T:A*A (or T:A*T) triplets for the purine one. Up to now it was not possible to recognize every double-stranded sequence even though the use of modified nucleic acid bases (32,33), intercalating agent-oligonucleotide conjugates (34,35), triplex dimerization (36) or "strand switching" (37,38) extends the number of sequences that can give rise to stable triple-stranded structures.
We recently developed a new approach that aimed at recognizing hairpin loop structures via the formation of doublehairpin complexes (18,19). Such complexes involve a local triple-stranded structure between the folded target and the antisense sequence, as demonstrated with a model sequence corresponding to the pyrimidine motif. As a contribution to the extension of the triplet alphabet, we investigated the formation of a double hairpin complex in which the purine strand of a target duplex contained 6 pyrimidines out of 15 bases. Systematic investigation of all possible triplet combinations had shown that for the DNA pyrimidine motif, inverted TA and GC pairs can be read by G and T or C, respectively, in the third strand (21)(22)(23)(24). The optimal residue to recognize an inverted pair may vary with the surrounding bases; however, our antisense sequence was designed according to these previous reports to minimize the destabilization induced by a mismatched triplet, leading to the potential formation of T:A*T, G:C*T, C:G*C ϩ , and A:T*G triplets. Although these triplets are not isomorphous, our results summarized in Scheme 1 unambiguously showed that a double hairpin complex involving a nonperfect triple-stranded complex was able to form at pH 6.0 with a DNA hairpin.
Triple helix formation by oligonucleotides containing T, C, and G has been previously described for an oligopurine-oligopyrimidine target (39). But this is to our knowledge the first time that a triple helix-forming oligomer allows accommodation of all four bases pairs in a complex with a significant stability. This should be related in part to the entropic contribution resulting from the linkage between the Hoogsteen and one Watson-Crick strand, as previously demonstrated with circular or clamp oligonucleotides (40,41) The stability of such complexes could be further increased by the use of either mod- ified nucleic acid bases like 5-methyl cytosine, which allows work at neutral pH (42,43), or by conjugation to intercalating agents (44 -47).
Our ultimate goal was to extend the antisense strategy to structured RNA targets. It was recently reported that RNA strands are excluded from triplexes with the purine motif (48). On the other hand, other studies, performed with different sequences corresponding to the eight possible combinations of DNA (D) and RNA (R) strands, concluded that two types of triplexes, namely D:R*D and R:R*D cannot form with an RNA purine strand (49 -51). The formation of double hairpin complexes with an RNA stem-loop target and an antisense oligodeoxynucleotide generates these unfavorable triplexes. The results reported here demonstrated that the double hairpin complex strategy allowed formation of a triple-stranded structure with an RNA second strand. This constitutes an alternative to the use of intercalating agent like berenil, ethidium, or 4Ј,6-diamidino-2-phenylindole, which have been shown to promote the formation of unstable polydT*polyrA:polydT (52). This offers the possibility of blocking biological process by selective targeting of RNA hairpins.