Solution structure and DNA-binding properties of the C-terminal domain of UvrC from E.coli

UvrC CTD is a DNA-binding domain

The loss of 5′ incision by UvrC protein and its inability to bind to single-stranded DNA (ssDNA) in the absence of the UvrC CTD (Moolenaar et al., 1998) suggests that the UvrC CTD is directly involved in DNA binding. To test this hypothesis, we performed electrophoretic mobility shift assays (EMSAs) using ssDNA, double-stranded (ds) DNA and various forms of DNA with ss–ds junctions. (Figure 2A). No binding was detected on dsDNA or ssDNA, while binding to bubble type DNA was observed when at least six unpaired bases are present. The binding strength increased with the number of unpaired bases. Weaker binding was observed on hairpin substrates, although the difference from the bubble substrate is not very large. However, a stable complex was formed only when 16–20 unpaired bases were present. A ‘fork’ with unpaired DNA on the 3′ end of the dsDNA was a poor substrate for UvrC CTD. Weak binding was detected with a fork containing 8 or 10 unpaired bases on both the 3′ and 5′ ends. Interestingly, the protein–DNA complexes formed with various substrates with an increasing number of unpaired bases appear in almost equal positions in the gel, despite their differences in size. This suggests a similar distortion or bending of the various DNA substrates upon binding. The specificity of the complex was confirmed by binding competition using an excess of unlabelled DNA (data not shown). The N-terminal His tag does not contribute to the observed binding since its removal did not change the binding affinity (data not shown).

The observed binding of UvrC CTD contrasts with results obtained using full-length UvrC on comparable substrates. No binding was observed for UvrC on DNA fragments that served as good substrates for the UvrBC complex (Zou et al., 1997; Moolenaar et al., 1998, 2000). Although experimental differences such as binding conditions (or DNA sequences) could not be excluded, we propose that the HhH domain is exposed differently due to steric hindrance or conformational changes when present in a UvrB–UvrC complex, in UvrC or in isolation. Indeed, the activities of other domains of UvrB or UvrC also change upon complex formation. For example, UvrB is unable to bind to lesion-containing dsDNA, whereas it can bind to a damaged DNA substrate with a 5′ overhang. The UvrB–UvrC complex, however, can bind to both substrates (Moolenaar et al., 1998, 2000). Furthermore, formation of a complex is essential for DNA repair since neither UvrB nor UvrC alone can perform incisions (Moolenaar et al., 1995). Together, these data indicate that the two individual proteins function differently when present in isolation or in the UvrB–UvrC complex. This might explain the absence of DNA binding of UvrC, despite the presence of a domain that is shown to be required for DNA binding (Moolenaar et al., 1998).

The observed binding preferences of the UvrC CTD to various (non-damaged) DNA substrates are in good agreement with the reported binding preference of the UvrB–UvrC complex. This complex binds to a DNA substrate containing an ss–ds junction (denoted Y substrate) with an apparent K_d of 5–10 nM (Zou et al., 1997). Studies by Moolenaar et al. (1998) extended these observations, showing that the UvrB–UvrC complex binds a probe containing a 5′ ssDNA overhang but not a 3′ ssDNA. The prerequisite for at least six unpaired bases for UvrC CTD to bind to a bubble substrate is in good agreement with the minimal size of the bubble required for incision by the UvrB–UvrC complex. However, under these conditions, the incision was strictly dependent on DNA containing a modified base (Zou et al., 1999). The weak binding to ssDNA is in accordance with the observation by Moolenaar et al. (1998) showing CTD-dependent binding to an ssDNA cellulose column (estimated to be in the micromolar range by Moolenaar et al., 1998). The inability to detect binding with dsDNA is compatible with the reported absence of binding affinity of the monomeric form of UvrC (Tang et al., 2001).

Quantitative analysis of the binding of UvrC CTD to DNA

Data on binding of UvrC CTD to various DNA substrates were fitted to determine the K_d. The apparent binding affinity for the bubble substrate was calculated to be 0.7 ± 0.2 µM. While the hairpin was a slightly worse substrate, the fork binds with a 4-fold lower affinity (Table I). We were unable to determine the K_d for ssDNA accurately, especially for dsDNA due to the low affinity of the protein for these substrates (Table I; data not shown). These binding data could be fitted to cooperative binding models with Hill coefficients >1, indicating the formation of at least two UvrC CTD molecules per DNA substrate. The Hill coefficient of 2.3 ± 0.4 for bubble DNA points towards an even higher order complex. For comparison, we show the simulated cooperative binding curves (Figure 2B) for this protein–DNA complex when composed of one, two or four UvrC CTD molecules per DNA substrate. Due to binding of UvrC CTD to additional low affinity binding sites (as illustrated by smearing at higher protein concentrations, inset of Figure 2B), we are unable to conclude from these DNA-binding studies whether two or more UvrC CTD proteins are binding to a bubble substrate.

The observed smearing indicates that during electrophoresis, the complex dissociates, suggesting a high off-rate. Dissociation rate experiments confirm the presence of a high off-rate for all the substrates tested (bubble 8, hairpin 16, fork 8, ssDNA or dsDNA). At the first time point (15 s) after the addition of unlabelled DNA, all protein–DNA complexes were competed away (Figure 2C; data not shown). Given the detection limits of the experiment, this indicates a k_off > 0.5/s.

The binding experiments show that UvrC CTD binds with an apparent K_d of ∼1 µM to specific DNA substrates containing ss–ds junctions. Although systematic analyses of DNA-binding affinities of isolated HhH domains are not available, the observed K_d is in the same range as reported for other HhH domain-containing proteins (Doherty et al., 1996; Shao and Grishin, 2000). This binding affinity is at least an order of magnitude lower than for the UvrB– UvrC complex on similar substrates (Zou et al., 1997; Moolenaar et al., 1998). The lower affinity could be caused by DNA substrates that differ in both length and sequence. However, it is more likely that other part(s) of the proteins present in the UvrB–UvrC complex also fulfil an essential role in complex formation. Indeed, isolated UvrB can bind to some DNA substrates without lesions (Moolenaar et al., 2000), and atomic force microscopy revealed that extensive UvrB-dependent DNA wrapping is observed (Verhoeven et al., 2001), underscoring the idea that other domains in the UvrB–UvrC complex also contribute to DNA binding.

UvrC CTD is a monomer in solution and binds as a dimer to bubble DNA

From binding studies, we were unable to determine accurately the protein–DNA ratio, although a 2:1 ratio seems most likely. This leads to the question of whether UvrC CTD is a dimer in solution or dimerizes upon DNA binding. From a blue native gel electrophoresis experiment (Schägger et al., 1994), we noticed that UvrC CTD is monomeric in solution (data not shown). Gel filtration experiments further confirm this since UvrC CTD elutes from the column corresponding to a mol. wt of 8.1 ± 0.8 kDa (Figure 2D). The determined molecular weights for the bubble DNA and protein–DNA complex from the elution profile were 20 ± 2 and 38 ± 4 kDa, respectively. This indicates that the complex most probably contains two UvrC CTD molecules per DNA molecule.

Three-dimensional structure

The structure of UvrC CTD (UvrC^554–610) was solved at 27°C (20 mM NaPO₄ pH 6.8, 300 mM NaCl) using two- and multidimensional heteronuclear NMR spectroscopy, making use of uniformly ¹⁵N- and ¹⁵N/¹³C-labelled protein. NMR shows that the UvrC CTD is monomer in solution (by estimating the line width of the signals in the ¹H–¹⁵N HSQC spectrum), which is in agreement with the biochemical data presented in the previous section. Nearly complete ¹H, ¹⁵N and ¹³C assignments were obtained using standard resonance assignment procedures (see Materials and methods). The global fold was established using unambiguously assigned long-range nuclear Overhauser effects (NOEs). Several long-range NOEs between the helices were obtained from a 3D-NOESY(¹³C,¹H)-HSQC spectrum and a 3D-NOESY(¹⁵N,¹H)-HSQC spectrum. The unassigned NOEs with multiple possible assignments were used in ARIA (Linge and Nilges, 1999) as ambiguous restraints. The final ensemble of structures was calculated by simulated annealing on the basis of 1326 unambiguous and 60 ambiguous distance restraints, with the inclusion of 68 dihedral angle constraints obtained from TALOS on the basis of backbone chemical shift values. A summary of the experimental restraints and structural statistics is provided in Table II. A superposition of the final ensemble of 22 simulated annealing structures is shown in Figure 3A, and a stereo view of the representative (closest to average) structure in Figure 3B. It should be noted that inclusion of the automatically assigned restraints from ARIA increases the coordinate precision from 0.57 ± 0.1 to 0.46 ± 0.1 Å for the backbone atoms (Figure 4) and from 1.17 ± 0.1 to 1.0 ± 0.1 Å for all heavy atoms (residues 36–75).

The UvrC CTD consists of two consecutive HhH motifs, called HhH-I and HhH-II, linked by a connector helix (Figure 3). In the first motif, HhH-I, the first presumed helix, H1, is composed of just a single helical turn, while in general this is a well-defined helix in most of the HhH domains determined so far. In principle, this could be due to inadequate domain selection. However, (¹⁵N,¹H)-HSQC and 3D-NOESY(¹⁵N,¹H)-HSQC experiments with an extended protein (UvrC^481–610) exclude this possibility, since neither significant differences in chemical shift values nor extra NOEs that would be indicative of formation of an α-helix were seen for the H1 residues. The ¹⁵N- and ¹³C-edited 3D-NOESY-HSQC spectra display only a few sequential and no long-range NOEs for residues 22–27. The ¹H–¹⁵N NOE data (Figure 4C) reveal high backbone mobility in this region of the protein. The hairpin loop h1 (residues 28–34) shows mainly sequential NOEs but only a few long-range NOEs, and this results in the disorder observed for this region. The hairpin includes the conserved GhG pattern (where h is valine in this case). After this h1 loop, the second helix, H2, follows (residues 35–43). The N-terminal part of H2 containing the positively charged residues Lys35 and Arg36 shows very few sequential or long-range NOEs. This could be due to the mobility of their side chains. The connector helix, Hc, consists of residues 47–52. The connector helix is oriented in a perpendicular direction with respect to adjacent helices. The last two helices, H3 and H4, consisting of residues 55–58 and 67–77, form the HhH-II motif, with a hairpin loop, h2, consisting of residues 59–66. The conserved pattern GhG is broken in h2 by a serine in the place of the second glycine (GIS). This type of replacement of a conserved sequence is often seen in the case of HhH domains.

Structural similarity to other HhH proteins

By searching the Protein Structure DataBase (PDB) with the coordinates of the UvrC CTD using the program DALI (Holm et al., 1993), structural similarity was found to HhH domains of two proteins, namely RuvA and DNA ligase (Thayer et al., 1995; Pelletier et al., 1996; Roe et al., 1998; Ariyoshi et al., 2000; Lee et al., 2000). Backbone r.m.s.ds of 1.63 and 1.47 Å were obtained between the representative UvrC CTD structure and RuvA and DNA ligase, respectively (Figure 5). The most prominent difference between UvrC CTD and these two proteins is that UvrC has just a single turn in the first helix of HhH-I, while the others have a well-defined helix. In fact, comparison of the presently known HhH domain structures reveals that there is considerable variation in the detailed structures of various HhH domains: e.g. the absence, or distortion of the first α-helix, the absence or presence of the connector helix, variation in the size of the connector helix, etc. (Velankar et al., 1999; Wada et al., 2000). The localized sequence and structural differences in HhH motifs might result in different substrate specificities.

Mapping of the binding site by chemical shift perturbation

The DNA interaction surface of the UvrC CTD protein was determined using the NMR chemical shift perturbation method (Dekker et al., 1993), in which the 2D-¹H–¹⁵N HSQC spectra of UvrC CTD were recorded with successive additions of DNA. The method detects residues that are interacting directly with DNA or that are indirectly affected by binding (Figure 6). Large chemical shift changes (>0.2 p.p.m. in ¹H and 3.0 p.p.m. in ¹⁵N) were observed upon DNA binding mainly in the hairpin regions, indicating their involvement in the interaction with DNA. The changes increased on continued addition of DNA. Except for the signals corresponding to Glu30 and the side chain of Arg36, the rest of the signals that were involved in binding disappeared when half equivalent of DNA was added. As it became difficult to determine the end point of the titration due to broad or disappearing signals, determination of the K_d and the stoichiometry of the complex became impossible. Since, the maximum observed chemical shift change is 100 Hz (which is the last point before disappearance of the amide proton of Gly31), we find a lower limit of ∼600/s for the dissociation rate constant. The titration results suggest a complex binding mechanism that includes both high- and low-affinity binding sites. The protein seems to be in intermediate to fast exchange on the NMR time scale in its complex with DNA, as judged by ¹H–¹⁵N HSQC cross-peaks that showed changes in their position and broadened resonances. In most cases, these changes allowed unambiguous assignment of the cross-peaks of the ¹H–¹⁵N HSQC spectrum by following the cross-peak position during the DNA titration. The ¹H and ¹⁵N chemical shift changes for backbone amides are shown in Figure 6A and define the DNA-binding surface for this DNA substrate with the UvrC CTD protein. The interaction site includes residues Thr28, Ile29, Glu30, Gly31, Val32 and Gly33 from the HhH-I hairpin. Positively charged residues such as Lys35, Arg36 and Arg37 along with Met39 of the second helix of HhH-I also show changes in their chemical shift values. In addition, the peaks of Gly63, Ile64 and Ser65 from the second hairpin loop HhH-II shift upon DNA addition. The maximum amide chemical shift perturbation is observed for the residues Gly31 and Gly33 of the first hairpin. The down field shift of the backbone nitrogens as well as that of the amide protons of Gly31 and Gly33 of the HhH-I hairpin indicate the formation of hydrogen bonds between amide protons and the phosphate oxygens of DNA, as also found in the case of RuvA (Ariyoshi et al., 2000). The amide chemical shift changes of the basic residues Lys35 and Arg36 may indicate an interaction with the phosphate backbone, although these changes could also be due to the interaction of the side chains with the unpaired bases as there is a substantial downfield shift of the side chain amide proton of Arg36, indicating hydrogen bond formation.

From the NMR chemical shift mapping method, we see that UvrC CTD shows a similar pattern of interaction with DNA to that observed for RuvA (Ariyoshi et al., 2000) and DNA ligase (Lee et al., 2000). RuvA is a tetrameric Holliday junction-recognizing protein, containing two copies of the HhH motif in each monomer that interact with the minor groove of the DNA. The crystal structure of the RuvA–Holliday junction complex shows that DNA interaction with the protein is mainly through hydrogen bond formation between the amides of the residues in the hairpin and the phosphate backbone of the Holliday junction DNA. Polar interactions between the side chains of the basic residues close to the hairpin loop of the HhH motif and the phosphate oxygens have also been found. Despite the absence of the well-defined first helix, the UvrC CTD can interact with the DNA through the hairpin loops and the neighbouring positively charged residues of the HhH-I.

Model for DNA binding

We have shown above that the UvrC CTD binds weakly to either ssDNA or dsDNA. However, when binding experiments were performed with looped duplex DNA, binding to DNA was found with a minimum loop size of six bases. Strong binding was not observed with hairpin or fork DNA substrates, suggesting that a ds–ss junction is not sufficient for binding. The cooperative nature (Figure 2A) of the protein binding to bubble DNA requires that at least two molecules of UvrC CTD are involved in DNA binding. To accommodate two UvrC CTD molecules at the two ds–ss junctions of the bubble DNA, the junctions need to be at a certain minimum distance. Thus, to account for the loop size requirement and the cooperative binding, we suggest a model where two properly interacting molecules of UvrC CTD bind to the two ds–ss junctions of the bubble DNA.

The NMR shift perturbation results of UvrC CTD with DNA have shown that both hairpins h1 and h2 are involved in binding. As compared with the structurally homologous HhH domain of RuvA, similar regions are interacting with DNA: the hairpin loops and the positively charged residues from the second helix of HhH-1. In RuvA, these make contacts with the DNA backbone in the minor groove at a Holliday junction (ds–ss) (Roe et al., 1998). In the deposited structure of the RuvA complex (Rafferty et al., 1996; PDB 1CUK), however, the hairpins contact DNA within the duplex region and the DNA coordinates for the junction are missing. Since our biochemical experiments have shown that UvrC CTD binds to bubble DNA, we have modelled the protein–DNA complex using the ds–ss junction coordinates of the DNA of the RNA polymerase II complex (Cramer et al., 2001; PDB 1I6H). Figure 7 shows an interaction surface of the protein in the UvrC CTD–DNA complex where the two hairpins of the HhH domain interact with the two strands of the DNA. In our NMR titration experiments with the UvrC CTD, the conserved glycine residues in the first hairpin and the positively charged residues (Lys35) at the N-terminus of the second helix of the HhH-I show the largest chemical shift perturbation. The chemical shift changes for the backbone amides of Gly31 and Gly33 are attributed to the interaction between these amides and the phosphate oxygens, similarly to that seen for RuvA. The side chain of Lys35 might interact either with the phosphate oxygens as in the case of RuvA, or with the unpaired bases in the loop. This type of interaction with the phosphate backbone is likely to be non-sequence specific, but may require a distortion in the backbone. Studies of the pre-incision complexes of UvrB protein with substrates such as cis-Pt–DNA (Visse et al., 1994), BPDE–DNA (Zou and Van Houten, 1999) and AAF–DNA (Pierce et al., 1989) have shown that the formation of the UvrB–DNA pre-incision complex is associated with significant changes in the DNA structure. These reports suggest that those changes may include unpairing of base pairs, unwinding of the helix, destacking of the base pairs, etc. The bending of the DNA in the pre-incision complex has also been revealed by electron microscopy experiments (Shi et al., 1992) and by linear dichroism (Takahashi et al., 1992) of the DNA. These studies, along with the preference of the UvrC CTD for bubble DNA might suggest that bubble DNA resembles the intermediate state of DNA corresponding to the pre-incision complex. In summary, we have established that UvrC CTD is crucial for DNA binding, specifically to bubble DNA with a loop size of at least six bases. The structure of the domain has been solved and the DNA interaction surface characterized. Based on this, a model for the protein–DNA complex is presented that accounts for the interaction with DNA. More detailed structural studies of the UvrC CTD–DNA complex are in progress.

Plasmid

UvrC CTD (residues 554–610) and an extended UvrC construct (residues 471–610) were PCR amplified using Escherichia coli genomic DNA as a template, with primers (sequence available upon request) that were extended with an NdeI and BamHI site at the 5′ and 3′ end, respectively. Purified PCR products, after digestion with NdeI and BamHI, were cloned into the corresponding sites of pET15B (Novagen). Constructs were confirmed by sequencing.

Purification of UvrC^554–610

Escherichia coli BL21 (DE3) containing the plasmid were grown in minimum medium at 30°C, induced at A₆₀₀ = 0.6 with 0.5 mM isopropyl-β-d-thiogalactopyranoside and grown for a further 5–6 h before harvesting. For the production of ¹⁵N-labelled protein ¹⁵NH₄Cl was used, and for the ¹³C/¹⁵N-labelled sample, 0.2% (U-6)[¹³C]glucose and ¹⁵NH₄Cl was used. All purification steps were carried out at 4°C. The cells were resuspended in lysis buffer [50 mM NaPO₄ pH 8.0, 300 mM NaCl, 10 mM imidazole, 0.2% Triton X-100 (w/v), 1 mM β-mercaptoethanol, 0.2 mM phenylmethylsulfonyl fluoride (PMSF) and protease inhibitors (Sigma)] and stored at –80°C.

Frozen cells were lysed by two freeze–thaw cycles, followed by sonication using a Soniprep 150 model (MSE). The sonicate was cleared of cell debris by centrifugation at 15 000 g for 1 h. The supernatant solution containing UvrC CTD was loaded on an Ni-NTA column (Qiagen) equilibrated with buffer containing 50 mM sodium phosphate, 300 mM NaCl and 20 mM imidazole at pH 8.0, and eluted with 350 mM imidazole in the same buffer. The protein-containing fraction was dialysed against buffer containing 300 mM NaCl, as the solubility of UvrC CTD depends on the concentration of NaCl. After 2-fold dilution, it was loaded on a cation-exchange column and eluted with a 150 mM–1 M NaCl linear gradient. The fraction containing the protein was loaded on a high-load Superdex75 prep grade gel filtration column (APB) with a buffer containing 300 mM NaCl, 50 mM NaPO₄ pH 6.8. The volume of the sample was reduced by ultrafiltration (amicon) with a 1 K cut-off filter.

DNA fragments used for binding studies

dsDNA: 1, 5′-GGGCGGCGGG and 5′-CCCGCCGCCC; 2, 5′-GGG CGGCGGGGGGCGGCGGG and 5′-CCCGCCGCCCCCCGCCGCCC; and 3, 5′-GGGCGGCGGG(T)₈GGCGGGGCGG and 5′-CCGCCCCG CC(A)₈CCCGCCGCCC. Bubble DNA: 5′-GGGCGGCGGG(T_x)GGC GGGGCGG and 5′-CCGCCCCGCC(T_x)CCCGCCGCCC, where x = 4, 6, 8 or 10. Hairpin: 5′-GGGCGGCGGG(T_x)CCCGCCGCCC, where x = 4, 6, 8, 10, 12, 16 or 20. Fork: 5′-GGGCGGCGGG(T_x) with lower strand 5′-(T_x)CCCGCCGCCC, where x = 4, 6, 8 or 10. Half fork 5′-GGG CGGCGGG(T_x) with lower strand 5′-CCCGCCGCCC where x = 4, 6, 8 or 10. ssDNA: 5′-GGGCGCGGG(T_x), where x = 4, 6, 8 or 10.

EMSAs

The oligonuceotides purchased from Life Technologies were end labelled (only the G strand e.g. 5′-GGGCGGCGGG) with [γ-³²P]ATP using T4 kinase and purified using a polyacrylamide gel. An equimolar amount of unlabelled ssDNA was added to the gel-purified ssDNA for the preparation of the various substrates, if required.

Binding reactions (20 µl) were performed at 4°C in buffer containing 10 mM Tris pH 7.3, 5% glycerol (w/v), 1 mM EDTA, 1 mM dithiothreitol (DTT) and 10 µg/ml bovine serum albumin (BSA). Unless otherwise indicated, a final concentration of 1 µM UvrC CTD was used. After the addition of 2000 c.p.m. of purified DNA (specific activity 10⁹ c.p.m./µg), samples were incubated for 2 h on ice and loaded on a pre-run 6% (w/v) polyacrylamide gel (29:1) containing 0.5× TBE as a running buffer. The electrophoresis was carried out at 120 V for 3–4 h at 4°C. Gels were vacuum dried and exposed against a phosphoimager screen (Kodak) for at most 2 days. Quantification was performed using a Molecular Imager FX system using Quantity One software (Bio-Rad). For competion experiments, a 500-fold molar excess of unlabelled DNA was added to the reaction mixture prior to the addition of labelled bubble DNA.

Off-rate experiments were performed under identical conditions but in a larger volume (200 µl) and a UvrC CTD concentration that reaches ∼80–90% binding saturation. After complex formation, a 200-fold molar excess of unlabelled DNA was added (t = 0) and, at the indicated time points, 20 µl of the sample was loaded on a constantly running gel.

Determination of the K_d and the Hill coefficient

Binding affinity was determined by the quantification of the bound and unbound DNA since [DNA] << [protein]_total, [protein]_unbound = [protein]_total

All binding data were fitted using a non-linear regression method, against:

Fraction bound = 1/[1 + (K_d/[protein])ⁿ](1)

where n represents the Hill coefficient, indicating the number of proteins in complex with DNA. Since complete binding of DNA was not reached for all probes at the highest protein concentration due to distortion of the complex during electrophoresis, we also fitted the binding data with:

Fraction bound = a/[1 + (K_d/[protein])](2)

where a represents the fraction of DNA that was bound at binding saturation. K_d values did not change significantly if this correction factor was omitted.

The presence of additional low affinity binding site(s) did not dramatically influence the calculated K_d for the higher affinity binding sites, although relatively large errors were obtained due to complications with the determination of binding saturation.

K_d and Hill coefficients were determined using equation (1). To confirm that the calculated Hill coefficient can also provide information about cooperativity for these non-ideally behaving DNA complexes, we compared the binding data of the most stable complex formed [with bubble (8T)]. An accurate fit requires knowledge about the K_d, fraction bound at binding saturation (a) and the Hill coefficient (n). These values could be determined combining equations (1) and (2):

Fraction bound = a/[1 + (K_d/[protein])ⁿ](3)

Fitting of the binding data would not result in accurate determination of the variables due to the limited number of points and the steepness of the curve, together with the observed problems in determining the fraction bound at higher protein concentrations. Therefore, we took the average of three independent binding experiments and fitted these against equations (2) or (3) to determine the K_d and a. These values were then used in equation (3), and simulated curves were calculated for n = 1, n = 2 and n = 4, and plotted together with the average and SDs for the calculated fraction bound.

Molecular weight determination by gel filtration

For gel filtration, an analytical Superdex 75 HR 10/30 (APB) was equilibrated with NMR buffer at 4°C using a Pharmacia FPLC. A 100 µl aliquot of UvrC CTD (10 nmol), bubble DNA [5′-GGGCGG CGGG(T)₁₀GGCGGGGCGG and its complement, 5 nmol] or protein–DNA complex, incubated for 2 h on ice (5 and 10 nmol for DNA and UvrC CTD, respectively), were injected. For calibration, a mixture of BSA, ovalbumin, carbonic anhydrase, cytochrome c and aprotinin (Sigma) was used. Each sample was injected twice and mass was determined by fitting the elution time with the mass of the indicated proteins.

DNA titration

The oligonucleotides used for NMR measurements were purchased from RNA-TEC Belgium. The 22 bp oligonucleotides used for NMR has the sequence 5′-GGAAGATC(T)₆ACCGTACG; 5′-CCTTCTAG(T)₆TGGC ATGC.

The DNA samples were annealed and dialysed in 5 mM sodium phosphate and 30 mM NaCl pH 6.8, so that the final concentration of the buffer becomes the same as that for protein. The 2D ¹H–¹⁵N HSQC spectra of UvrC CTD with 0.024, 0.032, 0.048, 0.060, 0.068, 0.083, 0.12, 0.19, 0.22, 0.25, 0.45, 0.5, 0.60, 0.8, 1.0 and 2.0 equivalents of DNA were recorded at 20°C.

NMR measurements

NMR experiments were carried out at 27°C on a Bruker DRX 600 MHz instrument and Varian Inova 750 MHz. For the backbone resonance assignments, 3D-HNCO, CBCANH, CBCA(CO)NH, NOESY-(¹⁵N,¹H)-HSQC, TOCSY-(¹⁵N,¹H)-HSQC (as described in Cavanagh et al., 1996) spectra were recorded, and for the side chain resonance assignments, 3D-H(C)CH TOCSY and (H)CCH TOCSY spectra were recorded. NOE distance constraints were obtained from 2D-NOESY, 3D-NOESY(¹⁵N,¹H)-HSQC and 3D-NOESY-(¹³C,¹H)-HSQC spectra with mixing times of 70 and 150 ms. All NMR spectra were processed using the software package NMRPIPE (Delaglio et al., 1995) and analysed using NMRView (Johnson and Blevins, 1994).

Chemical shift-derived restraints

The chemical shifts of 97% of the nuclei were determined (BMRB accession No. 5217). The Cα, Cβ, C_O, Hα and N chemical shifts of 56 residues served as input for the TALOS program (Cornilescu et al., 1999). TALOS derives information on the Φ and Ψ backbone dihedral angles from a comparison of secondary chemical shift patterns of amino acid triplets against a database of secondary chemical shifts corresponding to known conformations. A conservative approach was chosen requiring that all 10 best matches agree for a prediction to be accepted. The TALOS predictions were converted into dihedral angle restraints as the average Φ and Ψ angles ± 2 SD or a minimum of ±10°.

Structure calculations

Distance restraints were obtained from 2D-NOESY and 3D-¹⁵N- and ¹³C-edited NOESY experiments (mixing time 150 ms). TALOS-derived dihedral angle constraints were used as described above. Structure calculations were performed (see methods in Bonvin et al., 2001) with CNS (Brünger et al., 1998) using the ARIA setup and simulated annealing protocols (Nilges and O’Donoghue, 1998; Linge and Nilges, 1999). The best 22 structures were selected based on the lowest total restraint energy.

Model for DNA binding

In the model structure of the complex, the ss–ds junction coordinate of the nucleotide was taken from the PDB 1I6H of the RNA polymerase II– DNA complex (Cramer et al., 2001). In the RNA polymerase II–DNA complex, the two loops of the first domain (RPB1) were interacting with the nucleotide junction. When we superimposed the UvrC CTD hairpin loops with the loops of the RPB1 domain that interact with the junction, an r.m.s.d. of <1.6 Å was obtained. Also, the distance between the two loops of the RPB1 was comparable with that between the two hairpin loops of the UvrC CTD.

Coordinates

The PDB accession No. for the coordinates is 1KFT.