Mechanism of dimerization of bicoid mRNA: initiation and stabilization.

Dimerization of bcd mRNA was shown to be important for the formation of ribonucleoprotein particles and their localization in Drosophila embryo. The cis-element responsible for dimerization is localized in a stem-loop domain (domain III) containing two essential complementary 6-nucleotide sequences in a hairpin loop (LIIIb) and an interior loop (LIIIa). Such an RNA element can potentially generate single or double "hand-by-arm" interactions leading to open and closed complexes, respectively. The former retains the possibility of forming multimers, whereas the latter does not. We showed previously that dimerization proceeds through a two-step mechanism, which includes a transition from the reversible initiation complex into a very stable one. Here we have addressed the nature of the initial interactions and the mechanism of transition. We engineered a series of different RNA fragments with the capacity to form defined open dimers, multimers, or closed dimers. We compared their thermodynamic and kinetic behavior and mapped nucleotides involved in intermolecular interactions by enzymatic and chemical footprinting experiments and chemical modification interference. Our results indicate that the initiation step leads to a reversible open dimer, involving a more limited number of intermolecular base pairs than expected. The two loops play distinct roles in this process, and the structure of loop IIIb is more constrained than that of loop IIIa. Thus, loop IIIa appears to be the driving element of the recognition process. The initial open dimer is then converted into a stable closed dimer, possibly through a kinetically controlled mechanism.

Dimerization of bcd mRNA was shown to be important for the formation of ribonucleoprotein particles and their localization in Drosophila embryo. The cis-element responsible for dimerization is localized in a stem-loop domain (domain III) containing two essential complementary 6-nucleotide sequences in a hairpin loop (LIIIb) and an interior loop (LIIIa). Such an RNA element can potentially generate single or double "hand-by-arm" interactions leading to open and closed complexes, respectively. The former retains the possibility of forming multimers, whereas the latter does not. We showed previously that dimerization proceeds through a twostep mechanism, which includes a transition from the reversible initiation complex into a very stable one. Here we have addressed the nature of the initial interactions and the mechanism of transition. We engineered a series of different RNA fragments with the capacity to form defined open dimers, multimers, or closed dimers. We compared their thermodynamic and kinetic behavior and mapped nucleotides involved in intermolecular interactions by enzymatic and chemical footprinting experiments and chemical modification interference. Our results indicate that the initiation step leads to a reversible open dimer, involving a more limited number of intermolecular base pairs than expected. The two loops play distinct roles in this process, and the structure of loop IIIb is more constrained than that of loop IIIa. Thus, loop IIIa appears to be the driving element of the recognition process. The initial open dimer is then converted into a stable closed dimer, possibly through a kinetically controlled mechanism.
RNA loop-loop interactions are commonly used to trigger initial recognition between two RNA molecules and to modulate a high diversity of biological functions (for a review, see Ref. 1). The frequent selection of hairpin loops as recognition motifs is explained by their intrinsic properties. Their accessibility and structural versatility allow them to adopt a variety of conformations, thus enabling an appropriate presentation of nucleotides that initiate the recognition process. This initiation step generally involves Watson-Crick pairing of a few nucleotides, preferentially G-C pairs. Usually, efficient recognition requires rapid bimolecular binding rates, regardless of the RNA pairing scheme. The initial loop-loop complex generally serves as an intermediate for subsequent stabilization involving different possible paths, such as isomerization, helix propagation, and formation of additional interactions (1)(2)(3).
Hairpin loops recognition is particularly well documented in the case of natural antisense RNAs, which play a variety of regulatory roles in bacteria and their extrachromosomal elements (1,4,5). Another well studied case is the dimerization of genomic RNA of HIV-1, 1 which is initiated by a loop-loop interaction involving an autocomplementary six-base sequence in a hairpin loop (reviewed in Ref. 1). The above mentioned examples mainly concern recognition between two hairpin loops and are referred to as "kissing" interactions. Interaction might also involve complementary sequences located in a hairpin loop and an interior loop ("hand-by-arm" interaction). This type of interaction can potentially generate "closed" dimers, involving a double symmetric hand-by-arm interaction, or multimers. The latter mode is used by Bacillus subtilis bacteriophage phi29 RNA, in which cyclic hexamers of the 120-base prohead RNA are formed and are needed for efficient in vitro packaging of the DNA genome (6,7). Another example of handby-arm interaction is provided by dimerization of bicoid (bcd) mRNA, the product of which specifies the head and thorax pattern in Drosophila melanogaster embryos. Dimerization of bcd mRNA was shown to be important for the formation of ribonucleoprotein particles containing bcd mRNA and their localization in the embryo (8,9). The cisacting element that triggers dimerization is part of the 3Ј-UTR of the mRNA, which folds into five well defined domains (Fig.  1E). 2 It corresponds to an RNA stem loop (domain III) that contains six-base complementary sequences located in the apical loop, LIIIb, and the interior loop, LIIIa (Fig. 1D). Recently, we showed that bcd mRNA dimerization is a two-step mechanism involving a rapid conversion of the initial reversible looploop complex into an almost irreversible one (9). This was supported by dissociation kinetics and the observation that sense or antisense oligonucleotides are unable to inhibit dimerization once stabilization has begun. On the other hand, we found that the isolated domain III (RNA III) formed reversible dimers and multimers, as accounted for by the hand-by-arm interaction possibilities, whereas the complete 3Ј-UTR RNA did not. It appeared from deletion analysis that domain I and the major part of domains IV and V are dispensable for the transition to the stable complex (9). This raises two important questions: (i) is a single or a double hand-to-arm interaction required for initiating dimerization, and (ii) how does stabilization proceed?
In the present work, we constructed RNA fragments of different lengths, which can form either two or a single hand-byarm interaction, and we compared their ability to be converted into stable complexes. We used enzymatic and chemical probing to map the interactions in the different kinds of complexes. Finally, we used chemical modification interference of the phosphate backbone and bases to identify individual positions of nucleotides that are essential in unmodified form to the dimerization process. Our results provide important clues on both the initial step of dimerization and the stabilization pathway. They also highlight the importance of loop IIIb structure as a trigger of the initial recognition and of the structural context in the fate of the reversible dimer.

Plasmid Construction and RNA Synthesis
Plasmids used in this study were described previously (9). The NlaIV-HindIII bcd fragment of the plasmid p⌬(IVϩV) (9) was subcloned into the StuU-HindIII sites of the pUT7 vector, a pUC derivative containing the T7 RNA polymerase, to create the plasmid p⌬(IϩIVϩV). PCR fragments corresponding to the isolated domain III (m,w) or (w,m) were generated from plasmid 875 (m,w) using primers 1 and 2 and from plasmid 875 (w,m) using primers 1 and 3 (Table I). RNA III issued from those PCR products starts at position 231 and ends at position 335 (using the numbering of the full-length bcd 3Ј-UTR). The IIIЈ series starting at position 231 and ending at position 347 was generated from the plasmid 875 (m,w) using primers 1 and 4 and from plasmid 875 (w,m) using primers 1 and 5. PCR products were purified by gel electrophoresis and transcribed by phage T7 RNA polymerase (10). Transcription with incorporation of ApG at the 5Ј terminus of the transcripts allowed efficient 5Ј end-labeling by T4 polynucleotide kinase with [␥-32 P]ATP (10). Unlabeled and labeled RNAs were purified by polyacrylamide-urea gel electrophoresis, eluted overnight by 0.5 M sodium acetate, 10 mM EDTA, pH 8.0, 0.2 volume of phenol, ethanol-precipitated, and dissolved in water.

Dimerization of RNAs
RNAs were heated for 3 min at 85°C and cooled slowly at 60°C. After the addition of the 5-fold concentrated appropriate buffer, the samples were incubated for 30 min at 25°C and then chilled on ice. Standard dimerization buffer (D1) contained 50 mM sodium cacodylate, pH 7.5, 300 mM KCl, 5 mM MgCl 2 . Monomers were obtained by replacing buffer D1 by buffer M1 (50 mM sodium cacodylate, pH 7.5, 50 mM KCl, 0.1 mM MgCl 2 ). When necessary, monomeric and dimeric forms were separated by gel electrophoresis. Short fragments (RNA III series) were fractionated at 4°C on 10% acrylamide/(bis-acrylamide 1/30) gels in 45 mM Tris borate, pH 8.3, 50 M MgCl 2 , and larger fragments on 1% (w/v) agarose gel in the same buffer.

K d Determination and Derived Procedures
The apparent dissociation constants (K d ) were evaluated by mixing a negligible constant concentration of labeled wild-type RNA III (ϳ0.1 nM) with increasing concentrations of its unlabeled RNA partner (ranging from 4.6 to 270 nM). Dimerization was conducted under standard conditions in buffer D1 and fractionated by electrophoresis (see above). Radioactivity was measured with a BAS 2000 BIO-Imager. K d values were estimated as the concentration of unlabeled RNA necessary to obtain half-saturation, assuming that complex formation (homo-or heterodimers) obeys a simple bimolecular equilibrium. The treatment assumes that [free unlabeled RNA] ϭ [total unlabeled RNA]. This condition is fulfilled in the assay because the concentration of labeled wild-type RNA III is negligible compared with the concentration of its unlabeled RNA partner. K d values were the average of at least three independent experiments and have a maximum error of Ϯ50%. For kinetics of association, a negligible constant concentration of labeled wild-type RNA III was mixed with a constant concentration (50 nm and 100 nM) of its unlabeled RNA partner in buffer D1, and aliquots were loaded onto the gel at different times (9). For dissociation experiments, the complex was formed at 5 nM (a concentration that disfavors multimerization of wild-type RNA III), and subjected to a 50-fold dilution. Aliquots were analyzed as a function of time (9). Time 0 was obtained by running an aliquot of the dimerization mixture immediately before dilution. It was also verified that no dimer was formed when the RNA was incubated for 30 min at the final concentration obtained after dilution.

Probing and Footprinting Experiments
Most experiments were conducted on 5Ј end-labeled fragments (1.2 M) of the RNA III series.
Enzymatic Hydrolyses-Incubation was for 2 to 8 min at 25°C in the presence of 0.2 unit of RNase T1, 0.02 unit of RNase T2, and 0.2 or 0.05 unit of RNase V1 in buffer D1 or M1. Reactions were stopped by phenol/chloroform extraction followed by ethanol precipitation. Dimethysulfate (DMS) Modification (C(N3))-Reaction was for 5 min at 25°C with 2.5 or 5 l of DMS in buffer M1 or D1. After precipitation with 0.3 M sodium acetate and 3 volumes of ethanol, samples were treated with 10 l of hydrazine 10% for 5 min at 0°C and precipitated.
Diethylpyrocarbonate (DEPC) Modification (AN7)-Reaction was for 12 min at 25°C with 2.5 or 5 l of DEPC in buffer M1 or D1. After precipitation with 0.3 M sodium acetate and 3 volumes of ethanol, samples were treated with 10 l of hydrazine 10% for 5 min at 0°C and precipitated. The modified RNAs were then incubated with 10 l of aniline for 10 min at 60°C for the cleavage reaction. Cleavage products were analyzed by gel electrophoresis on a 10% polyacrylamide/8 M urea gel. RNase T1 and alkaline ladders of end-labeled RNAs were run in parallel to facilitate band assignment (11). In the case of RNase T1 footprinting on larger fragments (RNA ⌬I series), cleavage was detected by reverse transcriptase extension of a primer complementary to nucleotides 389 -400. Fractionation was on 10% polyacrylamide/8 M urea gel.

Phosphate Modification Interference
The appropriate 5Ј end-labeled fragment of the RNA III series (0.35 M) was alkylated on its phosphate groups by ethylnitrosourea (ENU). Modification was for 1 min at 90°C with 0.5 volume of ENU-saturated ethanol in 50 mM sodium cacodylate, pH 6.5, 50 mM KCl, 1 mM EDTA, in the presence of total tRNA (2 mM) as carrier. An incubation control was done in parallel using the same conditions except that ENU was omitted. The reaction was stopped by precipitation with 0.3 M sodium acetate and 3 volumes of ethanol. The modified RNA was then dissolved in water, mixed with its unlabeled partner, and submitted to the dimerization procedure. Dimeric and monomeric species were fractionated and RNAs eluted from the gel by overnight incubation at 4°C in 0.5 M sodium acetate, 10 mM EDTA, pH 8.0, 0.2 volume of phenol. After ethanol precipitation, ethylated phosphates were cleaved by incubation in 10 l of Tris-HCl 0.1 M, pH 9.0, at 50°C for 10 min. End-labeled RNA fragments were sized on a 10% polyacrylamide/8 M urea gel. Cleavage positions were identified by running in parallel the RNase T1 and alkaline ladders of the end-labeled RNA (11).

Modification Interference at Watson-Crick Positions
The appropriate fragment (1.2 M) of the RNA IIIЈ series was modified with DMS (A(N1) and C(N3)) or 1-cyclohexyl-3-(2-morpholinoethyl)carbodiimide metho-p-toluene sulfonate (CMCT) (G(N1) and U(N3)). Modification with DMS was for 1 min at 90°C with 0.001 volume of DMS in 50 mM sodium cacodylate, pH 7.5, 50 mM KCl, 1 mM EDTA in the presence of total tRNA (2 mM) as carrier. Modification was stopped by the addition of ␤-mercaptoethanol (100 mM) and ethanol precipitation. Modification with CMCT was for 1 min at 90°C with 0.025 volume of CMCT (70 mg/ml) in 50 mM sodium borate, pH 7.5, 50 mM KCl, 1 mM EDTA in the presence of total tRNA (2 mM) as carrier. Modification was stopped by ethanol precipitation. Incubation controls were run in parallel in the absence of chemicals. The modified RNAs were mixed with their unmodified partner and subjected to the dimerization procedure. Dimeric and monomeric species were fractionated as above. In this case, modified nucleotides were analyzed by primer extension with reverse transcriptase, using as a primer an oligonucleotide complementary to nucleotides 335-347. Conditions were adapted from Brunel et al. (12).

RNA III Preferentially Forms One Hand-to-Arm
Interaction-We previously showed that wild-type RNA III (RNA III(w,w)) forms dimers that dissociated rapidly, whereas larger tested RNAs do not dissociate within 1 h (9). In addition, RNA III multimers were formed in a concentration-dependent manner and dissociated rapidly. RNA III(w,w), which contains both loops LIIIa and LIIIb available for interactions, can form either a double hand-by-arm interaction (closed dimer, Fig. 1A) or single interactions leading to multimers (Fig. 1B). Because the initial recognition of bcd mRNA can potentially be driven by one or two hand-to-arm interactions, we compared the intrinsic behavior of RNA III(w,w) and derivate variants mutated in one of the two loops. Mutations introduced were identical to those already used (8,9) (Fig. 1D). RNA III(m,w) carried the substitution of AGUGAC for AAGCCC 282 in loop IIIb and a wild-type sequence loop IIIa. Conversely, RNA III(w,m) carried the substitution of GUCACU for GGGCUU 322 in loop IIIa and a wildtype sequence loop IIIb. Each mutated RNA was unable to dimerize by itself. However, the addition of RNA III(w,w) permitted the recruitment of the mutant RNA in a heterodimer through the nonmutated sequence loop, thus forcing dimerization through one hand-to-arm interaction (Fig. 1C, "open" dimer).
This system allowed us to evaluate the apparent dimerization dissociation constant (K d ) of different combinations of RNA III. Experiments were done by mixing a constant negligible amount of labeled wild-type RNA III(w,w) with increasing con-centrations of unlabeled RNA III(w,m) or RNA III(m,w). Because RNA III(w,w) is unable to dimerize at the chosen concentration, it was possible to study the properties of open dimers exclusively. The K d of RNA III(w,w), which could not be determined precisely, because of the formation of multimers, was estimated to be Յ10 nM. The K d values determined for both heterodimers formed with either RNA III(m,w) or (w,m) fell in the same range (Fig. 2C).
Then we compared association and dissociation of the complexes formed by RNA III(w,w) alone or in association with RNA III(w,m) or III(m,w), as a function of time. The formation of wild-type dimers and heterodimers was already completed within 30 s of incubation at both 50 and 100 nM ( Fig. 2A). This rapid association contrasts with the slower association observed for the complete 3Ј-UTR for which half-association required 12 min at a concentration of 125 nM (9). As expected, trimers and low amounts of higher species were also observed in the case of RNA III(w,w). For dissociation experiments, the complex was formed at 5 nM (a concentration that disfavors multimerization of wild-type RNA III) and subjected to a 50fold dilution (9). Half-dissociation occurred within the minute range in every case (Fig. 2B), again contrasting with an absence of dissociation in the case of the complete 3Ј-UTR (9). However, the heterodimers were totally dissociated, whereas the wild-type RNA yielded a constant fraction (30%) of dimer resistant to dissociation (Fig. 2B). This finding suggests that RNA III(w,w) forms two types of dimers. The major fraction, which displays kinetic properties similar to the heterodimers, The part of domain III containing the two complementary loops IIIa and IIIb is schematized, and the possible interactions are indicated by double-headed arrows. Wild-type RNA (w,w) can potentially form double or single hand-by-arm interactions, leading to closed dimers (A) or multimers (B), respectively. Introduction of a mutation (denoted by a star) in one of the two loops ((w,m) or (m,w)) prevents homodimerization but allows formation of open heterodimers with RNA(w,w) using single hand-by-arm interactions. Note that the "(w,w)" partner still retains the ability to recruit another RNA(w,w), thus enabling multimerization. This process can be controlled by RNA(w,w) concentration, and only dimers were obtained at concentrations below the probably corresponds to dimers utilizing a single loop-loop interaction (open dimers). These dimers are available to further multimerization. The minor fraction that forms stable dimers might correspond to closed dimers.
Probing the Conformation of RNA III and Its Two Variants-We probed the conformation of monomeric RNA III(w,w), III(w,m) and III (m,w) with enzymatic and chemical probes. RNase T1 cleaves unpaired G residues, RNase T2 preferentially cleaves unpaired As (and Us to a weaker extent), and RNase V1 cleaves double-stranded regions. DMS and DEPC alkylate C(N3) and A(N7), respectively. RNA(w,w) was probed under low salt conditions (buffer M1), which favor the monomeric form. RNA III(w,m) and III(m,w), which are intrinsically unable to self-dimerize, were tested under the same conditions but also under the high ionic conditions (buffer D1) used to promote dimerization. Representative gels are shown in Fig. 3, and the results obtained for the three RNAs are summarized in Fig. 4A.
The experimental data fully account for the proposed secondary structure shown in Fig. 4A. In particular, loops IIIa and IIIb were highly accessible to enzymes and chemical reagents in the three RNA tested. The cleavage and modification profiles of the three RNAs looked very similar, with only a few changes at or close to the mutation sites. The expected changes directly linked to the mutations are observed (e.g. new RNase T1 cut at G278 and G280 instead of G279 in RNA III(m,w), disappearance of cleavage at G318 and G319 in RNA III(w,m)) (Fig. 3). Surprisingly, RNase T1 cleavages at G309 and G310, in the terminal part of helix IIIb, were dramatically increased in RNA III(w,m) under low salt conditions (Fig. 3A). These cuts probably arise from secondary cleavage resulting from a downstream primary cut (most likely G317) in the loop IIIa mutant context. Nonetheless, this is a unique reactivity change not directly located at mutation sites, and taken together the data show that mutations do not perturb the structural organization of the RNA. In addition, probing the RNA III variants at both salt conditions did not reveal any significant difference (data not shown). Finally, cleavage and modification patterns were very similar to those obtained for wild-type or mutated domain III in the full-length bcd 3Ј-UTR, 2 leading to the conclusion that domain III forms an independent structural domain and that RNA III is representative of what happens in the whole RNA.
Footprinting Loop-Loop Interactions in RNA III Complexes-The involvement of loop IIIa and IIIb nucleotides in intermolecular interactions was first tested by RNase T1 footprinting in the three types of complexes. Heterodimers were formed in buffer D1 by mixing 5Ј-labeled RNA III(w,m) or III(m,w) and unlabeled RNA III(w,w), so that only the RNA forming one loop-loop interaction could be visualized. A typical experiment is shown in Fig. 3A, and the results are summarized in Fig. 4B. Upon its association with RNA III(w,w), RNA III(m,w) underwent a loss of cleavage at G317, G318, and G319, whereas G278 and G280 remained fully accessible in the mutated loop IIIb. Conversely, G279 was protected in RNA III(w,m) upon association with RNA III(w,w), whereas G317 remained accessible. Notably, cleavage at G309 and G310, which was enhanced as a result of loop IIIa mutation (see above), was reduced upon formation of the heterodimer with RNA III(w,w) (Fig. 3A). This protection might result from a stabilization induced by the high ionic concentration used for dimerization rather than dimerization itself. Homodimers were formed by mixing 5Ј-labeled RNA III(w,w) and unlabeled RNA III(w,w). A clear protection in both loops (G279 in loop IIIb and G317, G318, and G319 in  1 and 2, respectively). Lane C is an incubation control in the absence of RNase T1. Lanes L and T1, alkaline and RNase T1 ladders, respectively. B, modification of C(N3) by DMS. Incubation was for 5 min in the absence (lane C) or presence of 1 and 5 l (lanes 1 and 2, respectively) of DMS. C, modification of A(N7) by DEPC. Incubation was for12 min in the absence (lane C) or presence of 2.5 and 5 l (lanes 1 and 2, respectively) of DEPC.
loop IIIa) was observed, as expected from the implication of both loops in multimers (Fig. 1B).
We then used DMS and DEPC to get a more precise footprint of the loop-loop interactions and to test the participation of the two 6-nucleotide complementary sequences (GGGCUU 322 in loop IIIa and AAGCCC 282 in loop IIIb) in intermolecular base pairing. Modification with these two chemicals, followed by direct detection on end-labeled RNAs, allows probing of cytosine at a Watson-Crick position (N3) and adenine at its Hoogsteen position (N7), respectively. Note that the nonreactivity of A(N7) can be caused either by direct involvement in hydrogen bonding or by base stacking (13). As expected, a strong reduction of reactivity of C280, C281, and C282 to DMS was observed in RNA III(w,m) upon association with RNA III(w,w) as well as in RNA III(w,w) multimers (Figs. 3B and 4B). This result accounts for base pairing between these three C residues and complementary G residues (317-319) of loop IIIa. Besides, the reactivity of C274 was unchanged. More unexpectedly, C318 remained fully reactive to DMS (even a little more) in RNA III(m,w)/RNA III(w,w) heterodimers and in RNA III(w,w) multimers (Figs. 3B and 4B). This result argues against base pairing between C318 and G279, suggesting that base pairing is more limited than expected, at least in the open complexes. Otherwise, the reactivity of A276, A277, and A278 to DEPC was strongly reduced (Fig. 3C and 4B). Because our results suggest that base pairing is probably limited to three base pairs, CCC 282 /GGG 317 , the reduction of reactivity of AAA 278 to DEPC might be due to a stacked conformation rather than to base pairing. As these A-residues were highly reactive in the free RNA, it appears that the stacked conformation results from a conformational adjustment triggered by formation of the limited intermolecular interactions.
Chemical Modification Interference-Chemical modification interference was used to obtain more information about the functional groups required in the unmodified form for initial recognition. In these experiments, 5Ј end-labeled RNA III with the different loop combinations was submitted to limited modification (less than one statistically distributed modification/ molecule) under denaturing conditions and subjected to the dimerization procedure in the presence of unmodified and unlabeled RNA III(w,w). Dimerization-competent molecules and noncompetent molecules were fractionated by polyacrylamide gel electrophoresis, purified, and analyzed. Modifications present in the monomeric species but absent from the dimeric one are those that negatively interfere with dimerization. Conversely, modifications that are more abundant in the dimeric than in the monomeric form positively influence dimerization. Representative experiments are shown in Fig. 5, and the results are summarized in Fig. 6.
First, we used ethylnitrosourea (ENU) interference to map nonbridging oxygens of the phosphate groups, which are required to be unmodified for initial recognition (Fig. 5AB). The analysis of the single loop-loop interaction between labeled modified RNA III(w,m) and unlabeled RNA III(w,w) revealed a single window of negative interference, corresponding to complete inhibition of dimerization upon ethylation of any phosphate group 3Ј to nucleotides 277-284. A weak but reproducible positive interference was also observed at phosphates 273-275. On the other hand, the interaction between loop IIIa of RNA III(m,w) with loop IIIb of RNA III(w,w), was strongly reduced by ethylation of phosphate groups 3Ј to nucleotides 317-324. ENU interference was also tested on RNA III(w,w). Much weaker interference windows could be detected in loops IIIa and IIIb, coinciding with those observed in the open heterodimers (Fig. 5, A and B). The absence of strong interference in the wild-type RNA was not surprising, because the modifi-FIG. 5. Chemical modification interference. A, phosphate ethylation interference by ENU. Modification was conducted on 5Ј end-labeled RNA III(w,w), (m,w), or (w,m) as indicated at the top of the gel. Dimerization was then allowed with unlabeled RNA III(w,w), and the resulting monomers and dimers were fractionated by gel electrophoresis. B and C, CMCT and DMS interference. The same protocol was used as described in A, except that the tested RNA IIIЈ(w,m) was unlabeled, and modified positions were revealed by primer extension. M and D, RNA extracted from the monomer and the dimer bands, respectively; T, total population of modified RNA. C, incubation control. Lanes L and T1 correspond to alkaline and RNase T1 ladders, respectively. Negative interference is indicated by arrowheads. A complete or almost complete disappearance of a band revealed a "strong" interference (black arrow), and only a slight decrease indicated a "weak" interference (gray arrow). Positive interference is denoted by an asterisk.
cation of one essential group in one loop is expected to inhibit dimerization mediated by this loop, whereas the other loop is still available for dimerization.
We then tested Watson-Crick positions by using DMS, which modifies A(N1) and C(N3), and CMCT, which modifies U(N3) and G(N1). These modifications (except C(N3)) can be revealed only by primer extension with reverse transcriptase (13). Thus, modification was conducted on the RNA IIIЈ series (RNA IIIЈ(w,m) and IIIЈ(m,w)), encompassing nucleotides 231-347, whereas the unmodified partner (RNA III(w,w)) stops at position 335. Extension from primer 335 to 347 allowed us to analyze the modified species exclusively. A large window of negative interference was observed in loop IIIb ( 277 AAGC-CCGGG 285 ) (Figs. 5B and 6), which exceeds the length of the complementary sequence and largely overlaps the nucleotides found protected in footprinting experiments (Fig. 4). Moreover, methylation of A276, C274, and A273 in the 5Ј part of loop IIIa was found to favor dimerization. Strikingly, the interference pattern mostly coincides with that obtained with ENU. Otherwise, we failed to detect any convincing and reproducible interference in loop IIIa. One possible reason is the close proximity of the primer used for reverse transcription, which impaired clear interpretation. To circumvent this difficulty, we used a DMS modification on 5Ј-labeled RNA III(m,w) to test the participation of the N3 position of C320 in association with unlabeled RNA III(w,w) (result not shown). No interference could be detected, in agreement with the observation that C320 was not base-paired in the one-loop dimer (see above).
Two important results emerged from this study: (i) the number of nucleotides in which modification interferes with dimerization is much larger than the number of nucleotides actually involved in base pairing; (ii) the two loops are not equally affected by chemical modifications (loop IIIb being more susceptible than loop IIIa), suggesting distinct roles for the two loops in promoting initial recognition.
The Wild-type Stabilized RNA Dimer Forms a Double Handto-Arm Interaction-From the experiments on RNA III, it became evident that a single arm-to-hand interaction is sufficient to promote dimerization and that this interaction is reversible. However, the fact that the wild-type RNA III was able to produce a fraction of stable dimers suggested that the stabilization might be related to its capacity to form a double closed interaction. By contrast, large RNA fragments yielded stable dimers and were unable to form multimers (9). Probing experiments conducted on the complete 3Ј-UTR (RNA 875Ј) failed to reveal clear differences between monomeric and dimeric forms, even within the two autocomplementary loops. 2 Two reasons could be invoked to explain this result. First, RNA 875Ј yielded only 50% dimerization, and second, the interpretation was spoiled by intrinsic limitations (nonreactivity of some nucleotides, pauses in reverse transcription). Thus, the exact number of nucleotides involved in base pairing could not be determined precisely. However, the results suggested that base pairing was not as extended as expected, because AA 278 and UU 322 at the edges of loops IIIb and IIIa displayed the same reactivity at their Watson-Crick positions in both monomeric and dimeric species.
Previous deletion experiments indicated that domain I and the major part of domains IV and V were dispensable for the transition to the stable complex (9). Thus, we constructed a truncated RNA, named RNA ⌬(I,IV,V), containing domains II and III, a short part of helix IVa closed by a tetraloop (Fig. 1E). As expected, RNA ⌬(I,IV,V) formed stable dimers (Fig. 2B), indicating that all of the information for stabilization is contained in this fragment. This fragment formed high numbers of dimers and did not multimerize even at high concentration (Fig. 7A). Moreover, it was able to form heterolength dimers with RNA ⌬I with a high efficiency (up to 95%), whether RNA ⌬I contained wild-type loops IIIa and IIIb (w,w) or one mutated loop ((m,w) or (w,m)). We took advantage of this possibility to footprint loops IIIa and IIIb of only one partner within the different types of heterodimers. Heterolength dimers between wild-type RNA ⌬(I,IV,V) and the different species of RNA ⌬I ((w,w), (m,w), or (w,m)) were formed, and RNase T1 was used to provide an unambiguous signature of the involvement of loops IIIa and IIIb in loop-loop interactions. RNase T1 cuts were then revealed by primer extension using a primer complementary to nucleotides 389 -400, which are present in RNA ⌬I and not in RNA ⌬(I,IV,V). A complete protection of the wild-type loop of the two RNA ⌬I variants resulted from their association with RNA ⌬(I,IV,V) (Fig. 7B), thus providing a signature for the two different interactions. Interestingly, both loops IIIa and IIIb were protected in RNA ⌬I(w,w) associated with RNA ⌬(I,IV,V) (Fig. 7B). Because these RNAs cannot form multimers, the double protection necessarily reflects a close interaction. This is the first evidence that the two interactions occur simultaneously on the same RNA molecule. These results suggest that stabilization proceeds through conversion of a reversible open dimer into a closed dimer.

Initial Recognition Involves Only a Limited Number of Nu-
cleotides-Here we show that the isolated domain III (RNA III) provides a valuable model to understand the first steps of dimerization of the complete bcd 3Ј-UTR. We used different combinations of wild-type RNA III and variants in one of the two loops with the capacity to form either open or closed interactions. We found that wild-type RNA III preferentially forms open reversible interactions, thus allowing formation of multimers that increase with RNA concentration. Notably, multimers and open heterodimers were undistinguishable, insofar as their thermodynamic and kinetic properties, as well as their probing and footprinting profiles, are concerned. One unexpected result is that base pairing between the two loops is not as extended as believed previously. Indeed, the chemical foot- print indicated that on the six potential base pairs (AAGCCC 282 /GGGCUU 322 ), only three are stably formed (CCC 282 /GGG 319 ). This is reminiscent of antisense RNAs, in which initial recognition involves only a limited number of nucleotides (for review see Ref. 1). The identified initial interaction involves G-C pairs, a feature often encountered in the loops of natural antisense RNAs that trigger recognition. Notably, the dimerization initiation site of HIV-1 genomic RNA contains a six-base complementary sequence with a conserved central GC, a feature that was also selected among degenerated pools of RNAs capable to homodimerize (14). These observations suggest that G-C base pairs are frequently used as the nucleation point of loop-loop interactions.
The Two Loops Play Distinct Roles-The most striking information provided by interference experiments is the unbalanced response of loops IIIa and IIIb. Indeed, loop IIIb appeared much more sensitive to modification than loop IIIa. The large window of interference to both phosphate ethylation and base modification covers the greater portion of loop IIIb, for the most part exceeding the nucleotides involved in the initial base pairing, whereas positive interference was observed in the 5Ј part of the loop. Such a high susceptibility to phosphate ethylation appears to be a common theme in loop-loop interactions (i.e. the loops of the antisense RNA CopA and its target CopT (2)). A similar ethylation interference pattern (with negative interference in most part of the loop and positive interference in the 5Ј part) was observed in the 9-nucleotide loop of the HIV-1 dimerization initiation site containing the 6-nucleotide self-complementary sequence (10). However, in these systems, base modification interference was restricted to nucleotides involved in initial pairing. Such an extent of interference by DMS and CMCT was rather unexpected and likely suggests that small perturbations of the loop topology are sufficient to inhibit looploop recognition. By contrast, loop IIIb appeared poorly, if at all, sensible to DMS and CMCT modification, whereas interference by ENU was observed at discrete positions.
These differences highlight the different structural constraints and specific roles of the two loops in initial recognition. Thus, loop IIIb should adopt a precise and unique topology that is required for recognition of loop IIIa, which in turn does not appear to be structurally constrained. Most likely, the only requirement for loop IIIa recognition is to be accessible and sufficiently flexible to be available for base pairing with bases displayed by loop IIIb. A similar mode of recognition governs the codon-anticodon interaction, in which the structure of the tRNA anticodon loop (and not the number of base pairs) plays a crucial role in correctly positioning bases in a pre-formed A helical conformation (15). This holds also true for antisense/ target RNA recognition, where the proper conformation of loops is required for initial recognition. Notably, the target or antisense loop generally contains a conserved YUNR motif proposed to form a U-turn motif similar to that of the anticodon loop (16,17). Thus, we can assume that structurally not yet identified features, such as noncanonical interactions and ion binding, are required for correct presentation of nucleotides. Only RNA III ⌬I was probed by primer extension because the primer used did not hybridize to the shorter fragment. Note that RNase T1 cleavage was less pronounced in wild-type RNA III (⌬I) alone than in the variant monomers because some homodimerization occurred at the concentration used.
Stabilization Involves Conversion of Open to Closed Interactions-The next question concerns the nature of the elements that trigger the conversion. We first hypothesized that sequences in domain II, or in the large hinge connecting domains II, III and IV, might be involved in this process (9). However, large deletions in the hinge region (or replacements by stable stem loops) did not alter the capacity to promote stabilization (results not shown). Furthermore, no conserved sequences were found in these regions. The observation that a fraction of wild-type RNA III yielded stable dimers, whereas forced open heterodimers could not, was a first indication that stabilization could be triggered by formation of closed interactions (this work). Thus, we constructed an RNA fragment lacking domains I, IV, and V, which were previously showed to be dispensable for stabilization. Importantly, this RNA formed very efficiently stable dimers, while unable to form multimers. Then we showed, using RNase T1 footprinting, that both loops IIIa and IIIb were engaged in closed interactions in the stabilized dimer. This was the first experimental evidence for the actual existence of the double hand-to-arm interaction in a dimer. Our results also indicate that formation of the single loop interaction is rapid and reversible in the isolated RNA III (this work), whereas larger fragments form dimers much slower (9). Indeed, half-association was reached in 2.5 and 12 min for RNA ⌬I and RNA 875Ј, respectively, at a concentration of 125 nM (9), whereas it was completed in less than 30 s for RNA III at 50 -100 nM.
Taken together, our findings provide new insight into the two-step mechanism, allowing us to propose the following model. The first step involves the recognition of a few nucleotides in loop IIIa of one molecule by loop IIIb of another molecule, leading to a reversible open dimer (Do). The following step corresponds to the conversion of Do into a stable closed dimer (Dc), with both loops engaged in symmetrical hand-to-arm interaction. Our data suggest that the mechanism of the Do to Dc transition may be controlled kinetically (Fig. 8). It can reasonably be assumed that the first step (formation of Do) would obey a bimolecular association, depending on concentration and diffusion. Otherwise, the conversion of Do into Dc can be assimilated to a monomolecular reaction, independent of RNA concentration (Fig. 8). Thus, the fate of Do (multimerization or stabilization) would be driven by the association rate of the two steps. If the association rate of the formation of Do (k ϩ 1) is higher than the rate of conversion of Do into Dc (k ϩ 2), the reaction would be displaced toward multimer formation. Conversely, if k ϩ 2 is higher than k ϩ 1, the reaction would tend toward the monomolecular conversion. This situation might occur when k ϩ 1 decreases as the length of RNA increases. Although this simplified model describes most of our observations, further experiments will be required for testing its validity (i.e. by precise determination of rate constants using more accurate techniques).
Whether the conversion of the initial open dimer, Do, into the stable closed dimer, Dc, is accompanied by an extension of base pairing is presently unknown. Indeed, the reactivity of C320(N3), which was still very high in open dimers of RNA III (this work), could not be tested in larger RNAs due to reverse transcriptase pausing, whereas G279, the putative partner of C320, was unreactive in both monomer and dimer species. 2 Otherwise, one might assume that additional interactions (i.e. helix/helix or bulge/helix) also contribute to the high stability of the dimer. Further investigation will be required to answer this question.
Thus, bcd mRNA has developed a highly tuned mechanism to dimerize. Although the role of dimerization is still unclear, one might assume that one goal is to generate an intrinsic duplication of the numerous regulatory sites that are present in the 3Ј-UTR 2 (18) or to create a new cis-acting element. The mechanism proposed from this study differs from that described for the prohead RNA of phage phi29, although they both utilize hand-to-arm interactions. The phi29 RNA does not form closed dimers but is used to build very stable hexamers that are held together by open interactions involving four complementary nucleotides in the lateral and apical loops (6, 7). One similar point might concern the first recognition step, which results, in both cases, in the formation of open dimers. However, although the bcd mRNA dimer is spontaneously converted into a stable dimer by the formation of a closed interaction, the phi29 RNA dimer is driven to the multimerization reaction, which stops at the level of hexamers, probably because of structural constraints. Clearly, this example highlights how similar recognition features can be used to achieve different biological purposes.