The role of Mg(II) in DNA cleavage site recognition in group II intron ribozymes: Solution structure and metal ion binding sites of the RNAmiddle dotDNA complex

Group II intron ribozymes catalyze the cleavage of (and their reinsertion into) DNA and RNA targets using a Mg2(+)-dependent reaction. The target is cleaved 3' to the last nucleotide of intron binding site 1 (IBS1), one of three regions that form base pairs with the intron's exon binding sites (EBS1 to -3).We solved the NMR solution structure of the d3' hairpin of the Sc.ai5γ intron containing EBS1 in its 11-nucleotide loop in complex with the dIBS1 DNA 7-mer and compare it with the analogous RNA-RNA contact. The EBS1-dIBS1 helix is slightly flexible and non-symmetric. NMR data reveal two major groove binding sites for divalent metal ions at the EBS1-dIBS1 helix, and surface plasmon resonance experiments show that low concentrations of Mg2(+) considerably enhance the affinity of dIBS1 for EBS1. Our results indicate that identification of both RNA and DNA IBS1 targets, presentation of the scissile bond, and stabilization of the structure by metal ions are governed by the overall structure of EBS1-dIBS1 and the surrounding loop nucleotides but are irrespective of different EBS1-(d)IBS1 geometries and interstrand affinities.

Group II introns are large ribozymes and mobile genetic elements capable of catalyzing their own splicing reaction (1)(2)(3). During splicing, the intron RNA excises itself from an RNA transcript in two sequential phosphotransesterification reactions that yield the two ligated exons and the excised intron in a lariat structure. Both steps of splicing are reversible, which enables the intron to reinsert into intronless sites on RNA or DNA, a process that is referred to as reverse splicing or retrohoming, if genomic DNA is the target of reinsertion (4 -8). The most extensively studied example of the retrohoming pathway is the one of the Lactococcus lactis Ll.LtrB group IIA intron, which requires an intron-encoded protein (IEP) 2 (9,10) encoded in an open reading frame in domain IV of the intron. During retrohoming, the IEP unwinds the DNA locally to allow hybridization of the spliced lariat intron RNA and the target DNA. The intron catalyzes the reverse splicing by cleaving the target strand and ligating its own termini to the flanking DNA. The opposite strand is cleaved by the IEP endonuclease domain, and the reverse transcriptase domain of the IEP transcribes the complementary cDNA from the intron RNA template. The removal of the RNA and the synthesis and ligation of the DNA, which replaces it, are catalyzed by host proteins and complete the insertion process. As mobile genetic elements, group II introns resemble non-LTR retrotransposons (11), and they perform splicing in a way very similar to that of the eukaryotic spliceosome (12,13). These parallels gave rise to the idea that group II introns might be ancestors of both the spliceosome and non-LTR retroelements, suggesting pivotal evolutionary importance of the group II introns for the shaping of eukaryotic genomes (14,15).
In both splicing and reverse splicing, exon-intron recognition is mediated by base pair formation between the exon binding sites (EBS) of the intron and the corresponding intron binding sites (IBS) on the exons (16). In the case of group IIB introns, there are three such contacts. EBS1 (with 5-7 nucleotides, the longest of the three sequences) and EBS2 bind the 5Ј-exon (17), whereas EBS3 forms a single base pair with the 3Ј-exon (Fig. 1, A and C). Additionally, another base pair within the intron, the so-called ␦-␦Ј interaction, helps to stabilize the intron-exon contacts by positioning the sequentially distant EBS1 and EBS3 close to each other (Fig. 1C) (18,19). The EBS1⅐IBS1 interaction confers high specificity to the site of reinsertion of the intron, thus preventing insertion into sites from which the intron cannot splice again. However, it has been shown that EBS sequences are not conserved within different group II introns (17,20,21). For this reason, any desired sequence can be bound and cleaved by the intron in trans as long as the EBS and IBS sequences are complementary (22)(23)(24)(25). This characteristic endows group II introns with a remarkable potential for gene therapy applications (26).
Group II introns consist of six domains (DI-DVI) radiating from a central wheel (Fig. 1A). DI, containing the EBS sequences, is the largest and constitutes an autonomous folding entity to which other domains dock in the folding process (27)(28)(29). Together with DV, it forms the minimal structure required for catalytic activity of the intron (30,31). Mg 2ϩ ions play a critical role for both structure and function of group II introns and large ribozymes in general (32)(33)(34)(35). Formation of a stable tertiary structure of the group II intron is dependent on the presence of divalent metal ions (28,36,37). Moreover, several metal ion binding sites have been located in the active site (38 -40), and a two-metal ion mechanism (41,42) has been suggested to underlie intron catalysis (43,44). In-cell studies establishing a correlation between the intracellular Mg 2ϩ concentration and the frequency of splicing and retrohoming buttress the relevance of Mg 2ϩ for group II intron catalysis (45)(46)(47)(48). The importance of the identity of the divalent metal ions bound to the intron is underscored by the finding that the presence of Mn 2ϩ can lead to a shift of the cleavage site (49) and that already low amounts of Ca 2ϩ decrease the turnover rate by 50% in the Sc.ai5␥ intron (50).
Although a wealth of genetic and biochemical investigations have shed light on group II intron function, the information on tertiary structure is sparse. The group IIC intron of Oceanobacillus iheyensis is the only entire group II intron for which crystal structures are published (44,(51)(52)(53)(54).
In this paper, we present the first structure of the complex between the d3ЈEBS1 hairpin and the dIBS1 DNA, using EBS1⅐dIBS1 of the intron Sc.ai5␥, found in mitochondrial transcripts of Saccharomyces cerevisiae, as a model construct. We focus on a detailed analysis of the metal ion binding properties of the complex as determined by NMR spectroscopy and surface plasmon resonance (SPR). Because the same catalytic mechanism underlies intron-catalyzed DNA and RNA cleavage, we compare our data with the structure and metal ion binding of the analogous d3ЈEBS1⅐IBS1 homoduplex construct (55) and discuss common features relevant for stable binding of the target and for the recognition of the cleavage site.

EXPERIMENTAL PROCEDURES
NMR Sample Preparation-In d3ЈEBS1⅐dIBS1 (Fig. 1B), nucleotides 5-25 of the hairpin correspond to the sequence of the d3Ј hairpin from DI of the Sc.ai5␥ group II intron (Fig. 1, A and B) except for nucleotides 15 and 17 in EBS1, which are adenines in the wild type sequence. In order to have a suitable starting sequence for in vitro transcription (56) and a more stable hairpin stem, four base pairs were added to the stem (box in Fig. 1B). The dIBS1 sequence is a deoxyribonucleotide 7-mer corresponding to the wild type sequence of dIBS1 except for T-to-G mutations in position 61 and 63 matching the mutations of EBS1. The resulting G⅐C base pairs are required to achieve a stable enough duplex formation for NMR investigation (see Table 4; see also Refs. 55 and 57). RNA was transcribed in vitro according to standard procedures (58) with T7 RNA polymerase produced in our laboratory. Isotope-labeled RNA was obtained by transcribing with uniformly 15 N, 13 C-labeled NTPs (Silantes GmbH) or with selectively deuterated NTPs (Cambridge Isotope Laboratories Ltd.). The RNA was purified by polyacrylamide gel electrophoresis using acrylamide/bisacrylamide concentrations of 15-18% and recovered from the gel by electroelution (Elutrap Electroelution System, Whatman, UK) and annealed by dissolving it in an excess of water at 85°C and rapidly cooling in icy water after 2 min of incubation. RNA was washed with 1 M KCl, pH 8, and H 2 O and concentrated by ultrafiltration in Vivaspin devices (Sartorius Stedim Biotech S.A.). The dIBS1 deoxyribonucleotide 7-mer was purchased HPLC-purified from Microsynth (Balgach, Switzerland) and desalted by gel filtration on illustra TM NAP-10 columns (GE Healthcare). The concentration of d3ЈEBS1 and dIBS1 was determined by UV-visible spectroscopy using extinction coefficients ⑀ 260 of 303.3 mM Ϫ1 cm Ϫ1 for d3ЈEBS1 and 63.9 mM Ϫ1 cm Ϫ1 for dIBS1. dIBS1 was added to d3ЈEBS1 to an excess of 10% to avoid the presence of unbound d3ЈEBS1. All samples contained between 0.5 and 0.8 mM d3ЈEBS1⅐dIBS1 as well as 110 mM KCl and 10 M EDTA. Prior to the acquisition of NMR data, each sample was lyophilized and dissolved in 100% D 2 O (Armar Chemicals) or 90% H 2 O, 10% D 2 O, and the pH was adjusted to 6.4 in D 2 O, corresponding to a pD of 6.8 (59) or to a pH of 6.8 in 90% H 2 O, 10% D 2 O.
NMR Spectroscopy-All spectra were recorded on a Bruker Avance 500-MHz spectrometer with a 5-mm CRYO QNP probe head with z-gradient coil, a Bruker Avance 600-MHz spectrometer with a 5-mm CRYO TCI inverse triple-resonance probe head with z-gradient coil, or a Bruker Avance 700-MHz spectrometer with a 5-mm CRYO TXI inverse triple-resonance probe head with z-gradient coil. Non-exchangeable proton res-FIGURE 1. Location and secondary structure of EBS1 and dIBS1. A, scheme of the proposed secondary structure of a group IIB intron. Base pairs between EBS1-3 (purple) and exonic IBS1-3 (green) as well as the ␦ and ␦Ј bases (orange) mediate correct recognition of the 5Ј-and 3Ј-exon both in splicing and reverse splicing events. The six domains of the intron (DI-DVI) and the four main branches of DI (Ia-Id) are labeled. Sites of intron-catalyzed cleavage are marked with black arrows. B, d3ЈEBS1⅐dIBS1, the RNA⅐DNA hybrid construct used in this study. The sequences of nucleotides 5-25 of the RNA (d3ЈEBS1) containing EBS1 (purple) and of dIBS1 DNA (green) are derived from the Sc.ai5␥ intron found in the cox1 gene of S. cerevisiae mitochondria The base pairs marked with light green/dark purple letters were mutated from A⅐T to G⅐C for the sake of stability (57). The nucleotides 1-4 and 26 -29 (boxed) are added to the wild type sequence. C, the spatial arrangement of the EBS⅐dIBS and ␦-␦Ј base pairs ensures binding of both exons in the correct orientation for cleavage. Interactions are exemplified for a double-stranded DNA target (gray).  15 N resonances are indirectly referred to 2,2-dimethyl-2-silapentanesulfonic acid proton resonances (61). All processing was done in TopSpin version 3.0, and assignments were carried out with the program Sparky. Residual dipolar couplings were determined by recording a series of J-modulated [ 1 H, 13 C]-HSQC spectra (62) that were recorded in the presence and in the absence of ϳ17 mg/ml filamentous Pf1 bacteriophages (ASLA Biotech Ltd.) used for alignment. Peak volumes were determined using the program CCPNmr Analysis (63) and fitted in with the program gnuplot.
Base pair formation was validated by the presence of characteristic interstrand [ 1 H, 1 H]-NOESY cross-peaks. In calculations, hydrogen bonds within base pairs were maintained by applying distance restraints between donor hydrogen and acceptor and between donor and acceptor atoms and by enforcing planarity.
From the extended RNA and DNA chain, 200 starting structures were calculated by restrained molecular dynamics (rMD) with CNS version 1.21 (66,67), applying all but residual dipolar coupling restraints. A high temperature stage of 40 ps at 20,000 K was followed by two cooling stages of 90 ps in torsional space and 30 ps in cartesian space. The 20 structures of lowest energy were subjected to a refinement by 88 ps of rMD cooling from 3000 to 50 K. For this step, XplorNIH version 2.3 (68, 69) was used, and 21 1 H-13 C residual dipolar couplings were included. The axial and rhombic component of the alignment tensor were estimated using PALES (70) and determined by an extensive grid search (71) to be Ϫ27.3/0.08. Throughout the refinement, the force constant for residual dipolar couplings was gradually increased from 0.01 to 1 kcal mol Ϫ1 Hz Ϫ2 . In the resulting 200 refined structures, some of the structures contained one or two NOE violations from the 19th conformer on. Accordingly, only the 18 conformers of lowest energy that satisfied all given restraints were subjected to further analysis. The structure ensembles were analyzed using MOLMOL (72), and the electrostatic surface potential was determined with the PDBPQR version 1.8 webserver (73,74) 6 ] 3ϩ protons were assigned in Sparky. All nucleic acid protons displaying such cross-peaks to the ammine protons were clustered according to their position in the solution structure calculated in the absence of [Co(NH 3 ) 6 ] 3ϩ (Table 1). For rMD calculations of the structures with bound [Co(NH 3 ) 6 ] 3ϩ , a loose distance restraint of 3-7 Å between the Co 3ϩ central ion and each nucleic acid proton displaying an NOE cross-peak to the ammine protons was added in the refinement. Because all ammine protons of [Co(NH 3 ) 6 ] 3ϩ resonate at one common frequency and therefore cannot be distinguished, the distance to the Co 3ϩ central ion was used for the restraints (77). In the resulting ensemble, the six of the 10 lowest energy conformers that had no violations of NOE or dihedral angle restraints were used for further analysis.
SPR Sample Preparation and Measurements-All data were recorded on a Biacore T100 system. d3ЈEBS1/d3ЈEBS1wt coupled to biotin via a four-uracil 3Ј-overhang was purchased PAGE-purified from IBA GmbH (Göttingen, Germany) and used as ligands. (d3ЈEBS1/d3ЈEBS1wt)-4U-biotin was immobilized on a Series S Sensor Chip SA (GE Healthcare) precoated with streptavidin on a carboxymethyldextran surface or on a carboxymethyldextran hydrogel chip (XanTec Bioanalytics, Düsseldorf, Germany), coated with neutravidin in our laboratory. The surface was pretreated with 3-5 injections of 1 M NaCl, 50 mM NaOH lasting 50 s at a flow of 30 l/min. Immobilization was carried out by injecting 200 g/ml d3ЈEBS1-4Ubiotin for 10 min at a flow rate of 5 l/min. All experiments were performed in 10 mM MOPS, 107 mM KCl (I ϭ 110 mM), 0.05% polysorbate 20, pH 6.8. The dIBS1 and dIBS1wt DNA and IBS1 and IBS1wt RNA 7-mers were used as analytes for kinetics measurements. dIBS1 and dIBS1wt were acquired and treated as described for NMR experiments. IBS1 and IBS1wt were purchased double-HPLC-purified from IBA GmbH. Each kinetics run was preceded by five startup cycles injecting the current running buffer. The system was normalized using BIA normalizing solution (GE Healthcare). The flow rate was 30 l/min. In each cycle, the adsorption and desorption were allowed to proceed for 60 s, each followed by 180 s of stabilization. At the end of each cycle, water was injected for 60 s to remove any residual analyte and Mg 2ϩ bound to the surface. For all experiments, buffer injections were used for blank subtraction, and one or more non-zero concentrations of the analyte were injected twice before and after the highest concentration to ensure that the performance of the surface did not significantly change within one experiment. All analyte samples were injected both into a flow cell where d3ЈEBS1 was immobilized and in a ligandfree reference flow cell for control and background subtraction. Measurements were repeated on a different sensor chip for confirmation. In order to compare the affinity of dIBS1 and IBS1 with d3ЈEBS1, kinetics experiments were recorded at 25°C. Mg 2ϩ titrations of dIBS1 and IBS1 binding to d3ЈEBS1 and dIBS1wt and IBS1wt binding to d3ЈEBS1wt were performed at 15 or 25°C by adding 0, 1, 2, 5, or 20 mM MgCl 2 or 1 mM [Co(NH 3 ) 6 ]Cl 3 to the running buffer and to the analyte stock. For each concentration of MgCl 2 , a separate experiment was run. In all experiments, 5-7 non-zero concentrations of the analyte were injected, being in the range of 0.25-16 M for dIBS1, 0.5-45 M for dIBS1wt, 0.0156 -8 M for IBS1, and 0.19 -45 M for IBS1wt. In order to obtain k on , k off , and K D , data were fitted and analyzed with the corresponding Biacore T100 evaluation software, assuming a 1:1 binding model.

RESULTS
Characterization of dIBS1 Binding to EBS1 by NMR Spectroscopy-To verify stable formation of the EBS1⅐dIBS1 hybrid, we used [ 1 H, 1 H]-NOESY spectra recorded in H 2 O ( Fig. 2A). When d3ЈEBS1 or dIBS1 are measured separately in solution, the imino protons of the recognition sequences (G13, G14, U18, and G19 of EBS1 and G61, T62, G63, and T64 of dIBS1) cannot be observed because these regions are largely unstructured, and the protons exchange rapidly with the solvent. The presence of resonances for each of these protons (colored labels in Fig. 2A) and of cross-peaks within and between EBS1 and dIBS1 is a clear indication that EBS1 and dIBS1 are indeed fully base-paired. Each imino proton in the d3Ј stem can be attributed to one resonance (black labels in Fig. 2A), their chemical shifts being very similar to the ones observed for the unbound d3ЈEBS1 (55), proving that the addition of dIBS1 does not interfere with the base pairing in the stem.
Sequence-specific assignment of the resonances of the nonexchangeable d3ЈEBS1 and dIBS1 protons was accomplished using standard [ 1 H, 1 H]-NOESY spectra and F1,F2-[ 13 C, 15 N]filtered NOESY spectra (60) (Fig. 2B). The chemical shifts of protons from the RNA stem are in excellent accordance with previously published ones for unbound d3ЈEBS1 (55). The sequential cross-peaks between U12 in the loop and G13 and G14 in EBS1 are very low in intensity (data not shown), probably due to an unusual geometry at residue G13. The remaining loop residues display typical spectral features of an A-form RNA except for A10 -U12, whose ribose moieties are in C2Јendo conformation according to the [ 1 H, 1 H]-TOCSY data (see Fig. 4B; also see "Experimental Procedures").
For dIBS1, the cross-peak intensity pattern in the [ nucleotides, except for C65, which displays the C3Ј-endo crosspeak pattern, display cross-peaks of intermediate intensity for both the H1Ј-H2Ј correlation and the H3Ј-H4Ј correlation, which is atypical for a pure C3Ј-endo or C2Ј-endo conformation. This can signify either conformational flexibility of the deoxyribose rings or a rare O4Ј-endo conformation. The latter however is associated with a very short H1Ј-H4Ј distance (2 Å) (65), which the NOESY data only suggest for T62 and T64 (data not shown). Additionally, the intensity difference between the H2Ј-H6/8 and the H2Љ-H6/H8 NOESY cross-peaks (Fig. 3) and a systematically higher J 1Ј2Ј than J 1Ј2Љ coupling characteristic for B-form DNA are not observed. It is thus clear that the DNA adopts neither the A-form of its EBS1 binding partner nor B-form geometry, which is the preferred one of DNA. Additionally, the rather broad DNA cross-peaks suggest conformational exchange within dIBS1 (Fig. 3).  The sequential walk between intra-and interresidue H2Љ-H6/8 cross-peaks is shown as a black line. In a standard B-form conformation cross-peak pattern, the H2Ј(i)-H6/8(i) (intraresidue) cross-peak is more intense than the H2Љ(i)-H6/8(i) cross-peak, whereas the H2Љ(i Ϫ 1)-H6/8(i) (interresidue) cross-peak is more intense than H2Ј(i Ϫ 1)-H6/8(i). The fact that no such pattern is observed rules out a stable B-form conformation of the DNA, and the rather broad appearance of the peaks points out the flexibility of dIBS1. The spectrum was recorded in D 2 O at 25°C. The Solution Structure-The ensemble of the 18 d3ЈEBS1⅐dIBS1 conformers (Fig. 5A) of lowest energy shows good convergence of the heavy atoms, represented by the low overall root mean square deviation of 1.00 Å ( Table 2). In the short helix formed by dIBS1 and EBS1 (Fig. 5B), the backbone trajectory of the dIBS1 strand varies. The stem, which is a regular A-form helix, and the EBS1⅐dIBS1 helix are nearly parallel to each other but slightly shifted in all 18 conformers. This shift is due to the uneven number of unpaired bases on the 5Ј-and 3Ј-side of EBS1 (see A10, U11, U12, and A20 in Fig. 5D). A20 on the 3Ј-end of the loop forms a bridge between the stem and the EBS1⅐dIBS1 helix by stacking in between their terminal base pairs C59⅐G19 and U9⅐G21. Opposite of A20, A10 on the 5Ј-end of the loop is stacked on U9, and in some conformers, U11 and U12 also display stacking interactions (Fig. 5D). In this arrangement, it is probable that hydrogen bond formation between A10N61 and A20N1 further stabilizes the structure. The single-stranded nucleotides not only stabilize the junction between the d3Ј stem and EBS1⅐dIBS1 but also seem to fix the position of the 5Ј-end of dIBS1. The observation of several cross-peaks between protons of C59, the 5Ј-terminal nucleotide of dIBS1, and A10, U11, U12, and A20 (Fig. 2B) agree well with the position of C59, which is placed between A10 and U11 or U12 at the 5Ј-end of the loop and A20 on the 3Ј-end (Fig. 5D).
In contrast to C59, C65, where cleavage occurs, is in an exposed position. Between U12 and G13 of EBS1, a sharp turn or kink changes the direction of the RNA backbone (Fig. 5C). This kink moves the bases of G13 and U12 far apart so that stacking interactions are only possible between G13 and G14. This explains why NOE correlations between U12H1Ј and G13H8 are extremely weak if observed at all because both protons are separated by a distance greater than 6 Å.
The Variable Non-standard Conformations of EBS1⅐dIBS1 Cause Its Low Stability-Because it was evident from TOCSY and NOESY spectra that dIBS1 does not assume any standard helical conformation and seems to be subject to conformational exchange, we evaluated more closely the geometry of EBS1 and dIBS1 in the hybrid duplex in five of the 18 lowest energy structures with visibly different backbone trajectories on the side of dIBS1 (Fig. 5B), representing possible fits to the NOE data. Importantly, no dihedral angle restraints limiting the sugar pucker of the dIBS1 nucleotides were included in the calculation (see "Experimental Procedures"). Because RNA is conformationally less tolerant than DNA, the geometries of hybrid duplexes are usually reported to be more similar to the A-form (78 -80). In agreement with this, the EBS1 strand adopts an A-form geometry even in control calculations, where only the ␣ and backbone angles are loosely restrained to the trans range, which is in line with the [ 1 H, 1 H]-NOESY and [ 1 H, 1 H]-TOCSY data. However, in contrast to EBS1, comparison of the back-  bone and sugar pucker-defining angles (Table 3) of dIBS1 with the standard angles found in A-form or B-form DNA proves that dIBS1 conforms to neither conformation in any analyzed trajectory. Another remarkable feature of the dIBS1⅐EBS1 duplex is the fact that all of the five conformers have a significantly narrower minor groove than major groove (14.8 Å versus 16.4 Å, on average), which is normally a feature of B-DNA. The deoxyribose rings of the different dIBS1 nucleotides have different sugar puckers and seem to be able to exchange between similar sugar puckers with the exception of T62 and G63 (O4Ј-endo) and C65 (C3Ј-endo), which have the same conformation in all analyzed structures (Table 3). This asymmetric structure of the EBS1⅐dIBS1 duplex adds a new variation to the continuum of helical conformations that can be observed for RNA⅐DNA helices, depending on the exact sequence and the distribution of purines and pyrimidines in each strand (81)(82)(83). SPR experiments were performed to investigate the impact of the conformational heterogeneity on EBS1⅐dIBS1 stability. At 25°C, in the absence of any divalent metal ions, the K D of EBS1⅐dIBS1 is 29 Ϯ 6 M (Table 4). This value is at the upper limit of what can be accurately measured by the instrument and hence should be considered an estimate. For comparison, the K D of the RNA⅐RNA duplex of EBS1⅐IBS1 is about 200 times lower (Table 4), due to the much lower dissociation rate of IBS1 RNA from EBS1. Given that the EBS1⅐IBS1 homoduplex is a regular A-form helix (55), the heterogeneous geometry of the EBS1⅐dIBS1 hybrid must be causing this drastically decreased affinity. To obtain more reliable data for the EBS1⅐dIBS1 interaction, we repeated the experiments at 15°C, where the affinity is higher, and determined a K D of 1.65 Ϯ 0.2 M (Table 4). We also tested the influence of low millimolar concentrations of Mg 2ϩ on the stability of the interaction ( Fig. 6 and Table 4). Strikingly, in the presence of only 1 mM Mg 2ϩ , which is in the range of the physiological intramitochondrial Mg 2ϩ concentration (84,85), k off and, consequently, K D decrease by a factor of 6 and 4.6, respectively, and in the presence of 2 mM Mg 2ϩ , the K D is about 1 order of magnitude lower than in the absence of Mg 2ϩ . This demonstrates that Mg 2ϩ is of vital importance to stabilize EBS1⅐dIBS1 by inhibiting dissociation of the two strands. Importantly, all experiments were carried out in a buffer with an equal ionic strength of 110 mM KCl sufficient to provide charge screening of the polyanionic sugar-phosphate backbone. Consequently, the stabilization induced by Mg 2ϩ is of a specific nature and not simply a charge compensation effect. Also, the affinity of the RNA⅐RNA contact is increasing upon the addition of Mg 2ϩ (Table 4). Whereas the RNA⅐RNA contact shows very similar K D in 1 and 5 mM Mg 2ϩ , suggesting that the maximum affinity has been reached, the K D of the RNA⅐DNA contact seems to stabilize only at 10 -20 mM Mg 2ϩ . Also, [Co(NH 3 ) 6 ] 3ϩ enhances the affinity of dIBS1 for d3ЈEBS1.
[Co(NH 3 ) 6 ] 3ϩ is a kinetically stable complex, which mimics a hexahydrated Mg 2ϩ ion. It thus probes for outer sphere binding events of Mg 2ϩ , which means a coordination mediated by the

Backbone torsion angles and pseudorotation angle of the EBS1 and dIBS1 residues
Values represent the average and S.D. of five of the 18 lowest energy conformers of d3'EBS1 ⅐ dIBS1 in degrees. Aithough the backbone and ribose conformation within the EBS1 strand is typical for an A-form helix, dIBS1 is very weakly defined in its ␣, ␥, and angles, and the ␦, , and P pseudorotation angles are at intermediate values between the optima of A-form and B-form.
For comparison, SPR data for the wild type sequences of d3ЈEBS1 and (d)IBS1 were collected. The wild type EBS1(d)IBS1 helix has two A⅐U base pairs (instead of C⅐G) in positions 15⅐63 and 17⅐61 (Fig. 1B; see "Experimental Procedures"), which is reflected in the drastically lower stabilities of wild type d3ЈEBS1⅐(d)IBS1 (Table 4). Just like the mutant, the wild type contact is efficiently stabilized by Mg 2ϩ addition. In fact, precise rate constants can only be measured in the presence of at least 5 mM Mg 2ϩ .
Two Metal Ion Binding Sites Are Located in the EBS1⅐IBS1 Region-Because Mg 2ϩ is critical not only for EBS1⅐dIBS1 stability but for the folding of group II introns and retrohoming in general, we localized Mg 2ϩ binding sites by a combination of Mg 2ϩ , Mn 2ϩ , and [Co(NH 3 ) 6 ] 3ϩ titrations.
To determine Mg 2ϩ binding sites, an NMR sample was titrated with millimolar concentrations of Mg 2ϩ and [Co(NH 3 ) 6 ] 3ϩ . A plot of the chemical shift differences ⌬␦ of the protons of d3ЈEBS1⅐dIBS1 in the presence of 3 mM Mg 2ϩ and 2 mM [Co(NH 3 ) 6 ] 3ϩ is depicted in Fig. 7. In the middle of the d3Ј stem, the protons of the two base pairs G4⅐C26 and U5⅐A25 react to Mg 2ϩ addition with intermediate chemical shift changes (Fig. 7A). U5H6 and G4H8 resonances also shift strongly in the presence of the larger [Co(NH 3 ) 6 ] 3ϩ molecule, indicating that the d3Ј stem contains a binding site that is accessible for both hydrated and bare Mg 2ϩ ions.
In the loop region (Fig. 7B), U11H1Ј, U12H1Ј, and A20H8 and -H2 display intermediate ⌬␦ values, whereas A10H2, G21H1Ј, and C59H6 experience strong chemical shift changes of Ͼ0.05 ppm in the presence of both Mg 2ϩ and [Co(NH 3 ) 6 ] 3ϩ . C59H6 is most affected, moving by 0.074 ppm. These findings suggest that the U9⅐G21 wobble pair that closes the loop, the adjacent single-stranded region, and C59 constitute a Mg 2ϩ binding site. In a similar titration experiment of [ 1 H, 15 N]-HSQC correlations (data not shown), the chemical shifts of A10N3 and A20N1 changed by 0.5 ppm upon the addition of 3.5 mM Mg 2ϩ , which corroborates this finding. G13 proton cross-peaks were not observable during the titration with Mg 2ϩ , but the addition of 2 mM [Co(NH 3 ) 6 ] 3ϩ has a large impact on G13H8. Also, the H1Ј of G13 and H5 of C65 experience intermediate ⌬␦ in reaction to both [Co(NH 3 ) 6 ] 3ϩ and Mg 2ϩ . This indicates a second binding site near the other end of the EBS1⅐dIBS1 helix. Within the EBS1⅐dIBS1 interaction, the A60⅐U18 base pair is most influenced by Mg 2ϩ . H1Ј and H8 of G1 display very drastic shifts in the presence of Mg 2ϩ , which result from Mg 2ϩ binding to the di-or triphosphate moiety only present on the 5Ј nucleotide (39,89). Because this binding  The experiment in 0 mM Mg 2ϩ was repeated (bottom row) after the one containing the highest concentration of Mg 2ϩ to rule out distortion of the k on , k off , and K D values due to Mg 2ϩ -induced degradation. c Rate constants are at the instrument limit. d Rate constants are outside of the instrument limit; the affinity was determined using a fit to the equilibrium (maximal) RU values.
site, which also causes the ⌬␦ of C29 and G2 protons, does not exist in the context of the whole intron, we will not discuss it any further. A Mg 2ϩ -induced chemical shift change of a proton can be the result of coordination of Mg 2ϩ at the same residue or of subtle structural rearrangements caused by Mg 2ϩ coordination in the vicinity. It can also be a mixture of both effects. However, the relative cross-peak intensities in fingerprint regions, such as the H1Ј-H6/8 and H2Ј/H2Љ-H6/8 cross-peaks in the [ 1 H, 1 H]-NOESY and the intense A10-U12 H1Ј-H2Ј C65 H3Ј-H4Ј crosspeaks in the [ 1 H, 1 H]-TOCSY (Figs. 2B, 3, and 4) remain unchanged in the presence of up to 4 mM Mg 2ϩ and 2 mM [Co(NH 3 ) 6 ] 3ϩ (data not shown). 3 This means that neither metal ion causes significant changes in the d3ЈEBS1⅐dIBS1 structure.
To locate Mg 2ϩ binding sites more precisely, titration experiments were conducted with metal ions that affect NMR parameters other than the chemical shift. Mn 2ϩ is a paramagnetic metal ion. Its binding to RNA at specific sites promotes relaxation of the protons in the vicinity, depending on the distance between the manganese and the proton nucleus (90). At low ratios of Mn 2ϩ to RNA (1:100), selective broadening of the resonances in Mn 2ϩ binding sites can be monitored (91,92) undisturbed by structural rearrangements. We therefore recorded [ 1 H, 1 H]-NOESY spectra in the presence of different micromolar Mn 2ϩ concentrations (Fig.  8). At 60 M, very few peaks are already broadened below the detection limit, thereby indicating good binding sites for Mn 2ϩ . Among the central residues in the d3Ј stem, only G4 and A3 appear to be sensitive to the presence of Mn 2ϩ . In the loop region, various protons are influenced by Mn 2ϩ , but less strongly. The cross-peaks between A10 and U11 and between U11 and U12 are not observable anymore in 60 M Mn 2ϩ . A20H8 and H1Ј also appear broader but are still observable. These findings support the idea of metal ion binding occurring at the single-stranded loop residues but imply low tendency of Mn 2ϩ to bind here. C65H5 and T64H6 at the 3Ј-end of dIBS1 as well as G14H1Ј and H8 of EBS1 are broadened to baseline, indicating strong binding near the cleavage site. G13 resonances were not observed in either the absence or the presence of Mn 2ϩ and could not be evaluated.
Finally, we performed structure calculations of [Co(NH 3 ) 6 ] 3ϩ bound to d3ЈEBS1⅐dIBS1. Apart from the chemical shift changes that [Co(NH 3 ) 6 ] 3ϩ binding induces (Fig. 7), NOEs between DNA or RNA protons (Table 1) and the 18 protons of the NH 3 ligands can be observed upon binding of the complex to the nucleic acid in [ 1 H, 1 H]-NOESY spectra. This fact is exploited to localize metal ion binding sites on DNA or RNA molecules (77,93). These NOEs were used to calculate the solution structure of d3ЈEBS1⅐dIBS1 with three [Co(NH 3 ) 6 ] 3ϩ    6 ] 3ϩ titrations on this structure (Fig. 9A). Evidently, the three calculated [Co(NH 3 ) 6 ] 3ϩ binding sites (large dark blue spheres) coincide well with the protons reacting to the addition of Mg 2ϩ (gray spheres) and Mn 2ϩ (yellow spheres) and with protons that are strongly affected by at least two different metal species (cyan spheres).
Ultimately, three metal ion binding sites of d3ЈEBS1⅐dIBS1 can be defined: the first one is located in the lower part of the RNA stem centered at the G4⅐C26 base pair and a second and third site are found in the loop region. Of the latter two, one is located at the stem-loop junction, involving the unpaired bases on both sides of EBS1 and the G19⅐C59 base pair, and the other is formed between dIBS1 and EBS1, near the 5Ј-end of the EBS1. All of the resulting binding sites are situated in the major groove of either the stem or EBS1⅐dIBS1. Fig. 9A also demonstrates that Mg 2ϩ -induced chemical shift changes alone (represented by gray spheres) often coincide well with the effect of other metal ions, indicating true binding regions. However, chemical shift changes can also be caused by structural effects on protons in the vicinity of a binding site (94), as is the case for U9H2Ј and U7H6 (see arrows in Fig. 9A).
The proposed binding sites agree well with the electrostatic surface potential of d3ЈEBS1⅐dIBS1 (see arrows in Fig. 9, B and C). In the case of the loop binding site close to G13⅐C65, the electrostatic surface potential map reveals a small, negatively charged tunnel that is formed between dIBS1 and EBS1. Probably, the Mg 2ϩ , attracted by the negative charge, binds further inside this tunnel and interacts with N7 of G13 or G14. This possibility is not reflected by the calculated position of the [Co(NH 3 ) 6 ] 3ϩ ion, which is probably due to the complex being too big to enter the tunnel. In contrast to Co 3ϩ in [Co(NH 3 ) 6 ] 3ϩ , both Mg 2ϩ and Mn 2ϩ can shed their water ligands partly or entirely and make inner sphere contacts with nucleic acid ligands. Importantly, the addition of the much smaller Mn 2ϩ ion has an effect on G14 protons, whereas [Co(NH 3 ) 6 ] 3ϩ ion addition does not, which supports the concept of Mg 2ϩ and Mn 2ϩ binding further inside the tunnel than [Co(NH 3 ) 6 ] 3ϩ .

DISCUSSION
In this study, we present the first solution structure of an EBS1⅐dIBS1 hybrid representing the recognition and cleavage site of a group IIB intron and a DNA target. In the absence of their binding partner, dIBS1 is entirely unstructured, and d3ЈEBS1 forms a stable hairpin with an unstructured loop region (55). Upon dIBS1 binding to EBS1, the two form a short hybrid helix whose position relative to the stem is determined by the stacking interactions and putative hydrogen bonds between the single-stranded nucleotides surrounding EBS1. Due to EBS1⅐dIBS1 helix formation, the loop backbone is no longer flexible and is forced to assume a sharp turn between the first nucleotide of EBS1 and U12/␦ base. These structural features are highly similar to those of the analogous RNA⅐RNA interaction of d3ЈEBS1⅐IBS1 that was previously solved in our group ( Fig. 10; compare Fig. 10A and Fig. 5). Based on this structure, we argued that the position of EBS1 in the loop and the length of the loop forcibly lead to formation of this turn upon IBS1 binding and hence to adjusting the scissile bond at the 3Ј-OH of C65 in a defined position easily accessible to the other active site components. In this paper, we demonstrate that also the dIBS1 target strand induces the same characteris-  (72). B and C, electrostatic surface representation of d3ЈEBS1⅐dIBS1 displaying three patches of negative potential. The potential is represented by a color gradient from red (Ϫ667 mV) to blue (128 mV). These panels were prepared in PyMOL with the APBSTools2 plugin (73,74). tic kinked structure of the recognition complex, although it has a much weaker affinity and different conformation when bound to EBS1 as opposed to the RNA target. Also, in the crystal structure of a substrate-bound group IIC intron (54), a very similar turn is observed between ␦ and the first EBS1 nucleotide, and ␦Ј and IBS1 bind from different sides, thus supporting a general relevance of the kink for the active site architecture. The helical geometries of EBS1⅐dIBS1 and EBS1⅐IBS1 are very different on the side of the target strand. This difference strongly suggests that the specific geometry of EBS1⅐(d)IBS1 is not relevant for cleavage site recognition, with the exception of C65. At C65, the cleavage site, the structures of dIBS1 and IBS1 are more similar. C65 in dIBS1 has the (for DNA) unusual C3Ј-endo sugar pucker, which it naturally has in IBS1 RNA. Probably, this conformation is meaningful for the alignment of the scissile bond in the active center and the coordination of the catalytic Mg 2ϩ ions, and thus it must be the same in both DNA and RNA targets. The importance of the conformation of C65 is underlined by the crystal structure of the group IIC intron of O. iheyensis, in which two metal ions are coordinated in the active site between the backbone of the 3Ј terminus of IBS1 and the catalytically relevant nucleotides of DV (54), in a position and mutual distance that would allow catalysis by a two-metal ion mechanism (43).
The RNA⅐RNA and the RNA⅐DNA contacts differ in the relative orientation of the EBS1⅐(d)IBS1 helices to the d3Ј stem (Fig. 10, B and C); EBS1⅐IBS1 is more tilted than EBS1⅐dIBS1, resulting in a different position of the cleavage site with respect to the stem (Fig. 10C). One possible interpretation is that the hybrid helix, being more flexible than the homoduplex, can arrange in a way that maximizes stacking onto the stem helix, whereas the RNA⅐RNA interaction is too rigid for this. Within the full-length intron, however, this difference might be of little consequence, because a multitude of interactions, such as the ␦-␦Ј base pair (Fig. 1C) and tertiary contacts between DV and EBS1⅐IBS1 (95) as well as contacts between the target strand and the auxiliary protein components of the IEP, influence the exact cleavage site position.
The difference in geometry between the RNA⅐RNA and the RNA⅐DNA interaction causes the latter to be markedly less stable. This is well in line with previous studies attesting to lower melting temperatures and lower thermodynamic stability to hybrid helices in comparison with RNA⅐RNA homoduplexes of corresponding sequence (96 -98). We could show by SPR measurements that Mg 2ϩ concentrations similar to the physiological concentration stabilize both the RNA⅐DNA and the RNA⅐RNA interaction strongly without altering the overall helical geometry. Control experiments did not reveal any changes in the fold or the flexibility of the unbound d3ЈEBS1 loop upon Mg 2ϩ addition (55). This is in line with the observation that Mg 2ϩ has little influence on the association of EBS1 and dIBS1. In contrast to this, Mg 2ϩ strongly decreases the dissociation rate constant. Probably, Mg 2ϩ helps to prevent dissociation of the exon-intron recognition complex until all active site components have been assembled or even throughout the cleavage reaction. In the following, the metal ion binding sites relevant for this stabilization will be evaluated. At the K ϩ and Mg 2ϩ concentrations used, it is probable that also diffuse Mg 2ϩ ions play a role in the stabilization of EBS1⅐(d)IBS1 (99); a detailed quantification of their influence, however, is beyond the scope of this work. We thus focus the discussion on the site-bound Mg 2ϩ ions.
The binding site found in the d3Ј RNA stem of d3ЈEBS1 shows a preference for [Co(NH 3 ) 6 ] 3ϩ or hydrated Mg 2ϩ because for almost all protons, 2 mM [Co(NH 3 ) 6 ] 3ϩ causes stronger chemical shift changes than 3 mM Mg 2ϩ (Fig. 7A), and a wealth of NOE correlations to [Co(NH 3 ) 6 ] 3ϩ are observed. Such outer sphere binding sites in the major groove are regularly found in RNA (86). Possibly, this binding site contributes to stability (e.g. by making the d3Ј stem more rigid). However, the two metal ion binding sites in the loop region (Fig. 9A) directly involve EBS1 and dIBS1 nucleotides and thus seem more relevant for the affinity of d3ЈEBS1⅐dIBS1 in the presence of Mg 2ϩ . These are located in the major groove at the two termini of EBS1⅐dIBS1. Neither binding site shows a clear preference for inner or outer sphere binding. In general, the NMR and SPR data do not provide an exact characterization of the mode of interaction (100, 101) of the Mg 2ϩ ion with each binding site, because all three metal ions tested are able to interact with each binding site and because both Mg 2ϩ and [Co(NH 3 ) 6 ] 3ϩ efficiently enhance the affinity of dIBS1 for d3ЈEBS1.  (55). B and C, overlay of the backbone traces of d3ЈEBS1⅐dIBS1 (light gray) and d3ЈEBS1⅐IBS1 (dark gray) aligned by the backbone atoms of nucleotides 13-17 (in EBS1) and 61-65 (in dIBS1) close to the cleavage site (root mean square deviation ϭ 1.17 Å) (B) and aligned by the heavy atoms of the stem nucleotides (1-9 and 21-29, root mean square deviation ϭ 1.45 Å), showing only the G13⅐C65 base pair directly next to the cleavage site (C). Shown are electrostatic surface potential representations of d3ЈEBS1⅐dIBS1 (D) and d3ЈEBS1⅐IBS1 (E), depicting the strongly negative potential (indicated by the dark shade) in the tunnel formed by the major groove of the EBS1⅐(d)IBS1 helix. A-C were prepared with MOLMOL (72); D-E were prepared with PyMOL with the APBSTools2 plugin (73,74), and the images of d3ЈEBS1⅐IBS1 were prepared from Protein Data Bank entry 2M23.
Metal ion binding at the 5Ј-end of dIBS1 may reduce the flexibility of the unpaired nucleotides and contribute to stabilize this end of the EBS1⅐dIBS1 helix, by accepting ligand atoms from C59, G19, and the unpaired nucleotides surrounding them. In the second loop binding site, located between EBS1 and dIBS1 close to the G13⅐C65 and G14⅐T64 base pairs, Mn 2ϩ and Mg 2ϩ seem to be able to bind deeper inside the tunnelshaped major groove than [Co(NH 3 ) 6 ] 3ϩ . This indicates that a Mg 2ϩ ion might be able to move slightly within this binding region by exchanging some of its hydration shell with nucleic acid ligands. Such partial inner sphere coordination is well in line with crystal structures of RNAs in general, which show that the vast majority of Mg 2ϩ ions are partially dehydrated (102,103). In general, the combination of the kink in the sugar phosphate backbone at G13 and the short and tunnel-shaped major groove of EBS1⅐dIBS1 seems ideal to attract metal ions because it provides a suitable shape and accumulates negative charge in a small region. The G9⅐U21 wobble pair closing the loop, which is known for its affinity toward metal ions, completes this binding platform.
Mg 2ϩ titrations of d3ЈEBS1⅐IBS1 indicate that Mg 2ϩ binds to the same regions in both constructs (55). This means that the overall structure described above, which is common to the RNA⅐RNA and the RNA⅐DNA contact (Fig. 10, B-E) is much more relevant to attract metal ions than the specific geometry of the EBS1⅐(d)IBS1 helix, including the exact width of the major groove, which is different (Fig. 10, D and  E). Moreover, this structure is supposed to form independently of the exact sequences of dIBS1 and EBS1, provided the length and position of EBS1 in the d3Ј loop are suitable (see above; see Ref. 55). The hypothesis of equivalent Mg 2ϩ binding to different EBS1⅐(d)IBS1 sequences is tentatively supported by the observation that also the wild type sequences of d3ЈEBS1⅐IBS1 and d3ЈEBS1⅐dIBS1 show much higher affinities in the presence of low millimolar Mg 2ϩ concentrations. However, localization of these binding sites and structure determination are impeded by the low affinity of the wild type recognition complexes.
It is thus reasonable to assume that similar structural features as described above are used by different group II introns to attract stabilizing metal ions to the EBS1⅐IBS1 complex. In the case of the O. iheyensis group IIC intron, a binding site for divalent metal ions is found in the d3Ј stem major groove near the single-stranded nucleotides framing EBS1 (54). Furthermore, G⅐U wobble base pairs (see above) are found at different positions within EBS1⅐IBS1 (as in RmInt1 (104), ScB1 and SoPETD (17), and EcI5 introns (105)) or at the final base pair of the d3Ј stem (as in Pl.LSU/2 (106) and Ll.LtrB (107) introns) in other group II introns, supporting the idea that metal ion binding in EBS1⅐IBS1 is a common feature.
It has been shown both in bacterial and eukaryotic cells that the efficiency of retrohoming is strongly coupled to the Mg 2ϩ concentration in the cell (47,48). In fact, the lower Mg 2ϩ concentration of the eukaryotic cell limits the retrohoming efficiency of group II introns that are of bacterial origin. Probably, group II introns residing in eukaryotic genomes have evolved to make optimal use of the available Mg 2ϩ (e.g. by promoting structures such as that of the cleavage site recognition complex described herein).