Twin Hydroxymethyluracil-A Base Pair Steps Define the Binding Site for the DNA-bending Protein TF1*

The DNA-bending protein TF1 is the Bacillus subtilis bacteriophage SPO1-encoded homolog of the bacterial HU proteins and the Escherichia coli integration host factor. We recently proposed that TF1, which binds with high affinity (K d was ∼3 nm) to preferred sites within the hydroxymethyluracil (hmU)-containing phage genome, identifies its binding sites based on sequence-dependent DNA flexibility. Here, we show that two hmU-A base pair steps coinciding with two previously proposed sites of DNA distortion are critical for complex formation. The affinity of TF1 is reduced 10-fold when both of these hmU-A base pair steps are replaced with A-hmU, G-C, or C-G steps; only modest changes in affinity result when substitutions are made at other base pairs of the TF1 binding site. Replacement of all hmU residues with thymine decreases the affinity of TF1 greatly; remarkably, the high affinity is restored when the two hmU-A base pair steps corresponding to previously suggested sites of distortion are reintroduced into otherwise T-containing DNA. T-DNA constructs with 3-base bulges spaced apart by 9 base pairs of duplex also generate nm affinity of TF1. We suggest that twin hmU-A base pair steps located at the proposed sites of distortion are key to target site selection by TF1 and that recognition is based largely, if not entirely, on sequence-dependent DNA flexibility.

The genome of the Bacillus subtilis bacteriophage SPO1 contains 5-hydroxymethyluracil (hmU) 1 in place of thymine, and discrimination between T-and hmU-containing DNA is important during phage multiplication (1,2). For one phageencoded protein, the DNA-binding and -bending protein TF1, this discrimination is essentially absolute: the high affinity of TF1 (K d order of magnitude nM) for preferred sites within the hmU-containing phage genome is greatly reduced for the corresponding T-containing DNA (3,4). We recently showed that the affinity of TF1 for hmU-DNA is matched in T-containing DNA constructs that contain sets of two consecutive mismatches separated by 9 bp of duplex, indicating that hmUcontent and appropriately placed flexible loops contribute sim-ilarly to complex formation. We proposed, therefore, that DNA in a complex with TF1 is distorted at two sites that are separated by 9 bp of duplex DNA, sites corresponding to two hmU-A base pair steps, and that TF1 identifies its target sites through recognition of sequence-dependent DNA flexibility (5). Here we examine the requirement for hmU residues across a 37-mer DNA duplex representing a preferred binding site within the phage genome (4) and specifically evaluate sequence requirements at the two inferred sites of DNA distortion.
EXPERIMENTAL PROCEDURES TF1 was overexpressed in Escherichia coli DL39 (6) and purified as described (5). Oligonucleotides with hmU content were synthesized as described (7) and purified by high pressure liquid chromatography. T-containing oligonucleotides were purchased and purified by denaturing polyacrylamide gel electrophoresis. The top strand (shared among bulge constructs) was 32 P-labeled at the 5Ј-end with T4 polynucleotide kinase. Complementary oligonucleotides were mixed at equimolar concentrations, heated to 90°C, and slowly cooled to 4°C over several hours to form duplex DNA.
Electrophoretic mobility shift experiments were carried out as described (5). Experiments with perfect duplex hmU-or hybrid T/hmU-DNA were done at room temperature unless noted otherwise. Experiments with bulge constructs were performed at 4°C to prevent dissociation of duplexes. After electrophoresis, gels were dried and protein-DNA complexes were quantified using phosphorimaging. DNA migrating between bands corresponding to free and complexed DNA was considered as free DNA (5) in calculating equilibrium dissociation constants (K d ) (5). Values of K d are reported as the average of at least three experiments Ϯ S.E. (for closely related DNA constructs yielding suboptimal complex formation, triplicate determinations were deemed unnecessary).

RESULTS
The sequence of a 37-mer DNA construct corresponding to a preferred binding site within the phage SPO1 genome (4) is shown in Fig. 1A. Because of the ability of TF1 to form nested complexes on longer DNA fragments, the length of the DNA constructs was chosen to accommodate only one TF1 dimer (5). The affinity of TF1 for the 37-mer hmU-containing duplex (K d was ϳ3 nM; Fig. 1A) was reduced 10-fold upon replacement of the two hmU-A base pair steps corresponding to the potential sites of distortion with A-hmU, G-C, or C-G steps (Figs. 1 and 2, panels B). Altering the sequence at the center of the binding site (Figs. 1 and 2, panels C) or on the flanks, distal to the proposed sites of bending (Figs. 1 and 2, panels D), did not significantly affect complex formation. The observed sequence specificity at proposed sites of distortion is consistent with a direct participation of hmU-A base pair steps in complex formation and DNA bending.
To evaluate the contribution of hmU residues to the DNA binding affinity of TF1, DNA constructs were designed in which central hmU-containing blocks of 13 or 11 bp, encompassing the proposed sites of DNA distortion (Fig. 3, constructs Th2 and Th3), or of 9 or 7 bp, excluding hmU-A steps that are candidate sites of specific distortion (Fig. 3, constructs Th4 and Th5), were embedded in T-containing surrounding sequence. Constructs in which the proposed sites of distortion lie within the hmU-containing blocks (Th2 and Th3) yielded K d ϭ 5-6 nM compared with 3 nM for all-hmU-DNA. A more than 5-fold reduction in affinity was observed for DNA with shorter hmUcontaining blocks (Th4 and Th5). A DNA duplex in which a central 13-bp T-containing block is flanked by hmU-containing sequence (Fig. 3, construct Th1) bound TF1 with 10-fold lower affinity relative to all-hmU-DNA.
For DNA with 2 4-bp hmU-containing blocks, a significantly lower affinity was observed for DNA constructs with T-A base pair steps at the proposed sites of bending (Fig. 3, constructs Th8 and Th9) than for otherwise identical constructs with hmU-A base pair steps (Fig. 3, constructs Th6 and Th7; K d was ϳ5 nM). Remarkably, TF1 was found to have essentially undiminished affinity for DNA in which the two hmU-A base pair steps are completely surrounded by T-containing DNA (K d ϭ 4 nM; Fig. 3, construct Th7/6, and Fig. 4, A and B). We note also that the affinity of TF1 for hmU-DNA and for the Th7/Th6 construct was the same at room temperature and at 4°C (Figs. 3 and 5 and data not shown). A 7-10-fold reduction in affinity was observed if only one strand contained the requisite hmU-A sequence (Fig. 3, constructs Th7/T and T/Th6, and Fig. 4C). Our interpretation of these observations is that hmU content affords an increased flexibility that is recognized by TF1 and that hmU-A base pair steps symmetrically disposed about the center of the binding site are targets for specific distortion in the protein-DNA complex.
The high affinity of TF1 for hmU-DNA (K d was ϳ3 nM) was reproduced in T-DNA that contains two appropriately positioned flexible sites generated by sets of tandem mismatches (Ref. 5 and Fig. 5A). Improved complex formation relative to perfect duplex T-DNA would also be expected for T-DNA constructs in which the energetic cost of bending is lessened through the incorporation of bulge-loops, i.e. with nucleotides that are formally unopposed on the complementary strand (Fig.  5). Constructs with sets of 1-or 2-nt bulges combined with one mismatched base pair positioned 9 bp apart indeed generated high affinity for TF1 (Fig. 5B); note that sets of 2-nt loops (one mismatch) generated suboptimal binding compared with sets of 4-nt loops (two consecutive mismatches). The high affinity of TF1 was fully restored for DNA constructs with sets of 2-or 3-nt bulges with suboptimal binding for DNA with a pair of 1-nt bulges (Fig. 5C). Replacing the T-containing top strand of bulge constructs with the corresponding hmU-containing strand did not affect complex formation (data not shown). Bulge-loops are considered to generate directed kinks in the helix axis (8,9). Consistent with the presence of such kinks, we observed a  reduction in electrophoretic mobility of bulged DNA constructs, the magnitude of which depended on the size of the bulge (data not shown). The mobility of protein-DNA complexes was similar irrespective of the nature of the DNA construct.
The TF1 double mutant protein TF1 (E15G/T32I) binds DNA with ϳ40-fold higher affinity than does wild-type TF1 (10) due to interactions outside the central 25-bp region of the binding site (5). The presence of bulge-loops in T-DNA affected the affinity of TF1 (E15G/T32I) and wild-type TF1 in similar ways, consistent with unaltered interactions for the mutant protein with a central DNA segment that includes proposed sites of bending (data not shown).
The structure of TF1 reveals the presence of flexible ␤-ribbon arms that are directly involved in DNA binding (11), 2 as does the structure of the closely related HU protein (12,13) and the recently reported structure of integration host factor (IHF) in a complex with DNA (14). At the tip of the DNA-binding arms, TF1 contains a Phe residue at position 61 in place of the Arg found in all other members of this protein family. TF1(F61R) has reduced preference for hmU-DNA over T-DNA (15) but was seen to retain high affinity for hmU-DNA (K d ϭ 9.6 Ϯ 0.7 nM). Substituting Phe-61 with Tyr had little effect on affinity (K d ϭ 6.8 Ϯ 0.9 nM). Replacing Phe-61 with a polar, uncharged residue (Ser or Gln) significantly reduces both DNA binding affinity and site-selectivity (15). Notably, the affinity for hmU-DNA of both TF1(F61S) (K d ϭ 236 Ϯ 19 nM) and TF1(F61Q) (K d ϭ 685 Ϯ 48 nM) increased more than 5-fold for T-DNA with 2 3-base bulges (K d ϭ 36 Ϯ 2 and 90 Ϯ 7 nM, respectively; Fig. 5). This is in contrast to wild-type TF1 that bound hmU-DNA and bulge-DNA constructs (1ϩ2, 1ϩ3, 0ϩ2, and 0ϩ3) with approximately the same affinity (Fig. 5). T-DNA with sets of 4-nt loops, which also reproduces the affinity of wild-type TF1 for hmU-DNA (Fig. 5A and Ref. 5), did not significantly enhance complex formation by TF1(F61S) (K d was ϳ200 nM) or TF1(F61Q) (K d was ϳ570 nM) beyond that observed with hmU-DNA. Apparently hmU content and flexible loops in T-DNA contribute similarly to complex formation for these TF1 mutant proteins, which are deficient in binding and bending of hmU-DNA (15,16), whereas DNA with a static kink generates an increase in affinity. DISCUSSION Our previous conjecture that TF1 principally identifies its preferred binding sites through recognition of sequence-dependent DNA flexure was based on an analysis of affinities for DNA constructs with sets of 4-nt loops (5). For this mode of target site selection, the prediction would be for the sequence at the sites of distortion to significantly affect complex formation. This prediction appears to be borne out both in terms of preferred base pair steps in the context of hmU-DNA (Figs. 1 and 2) but particularly by the significant effect of hmU-A base pair steps in a T-containing surrounding sequence (Figs. 3 and 4). We propose that hmU content confers a site-specific flexibility on the DNA duplex that is recognized by TF1. hmU-A base pair steps that are situated at presumptive sites of DNA bending make the principal contributions to complex formation with TF1. hmU residues at other positions that generate measurable, yet suboptimal affinities, presumably do so either due to longer range effects on DNA flexure or because of alternative placement of TF1. An effect of hmU residues on the deformability of the double helix is corroborated by the increased affinity observed for other members of the family of type II DNA-binding proteins, HU and IHF, and for the unrelated eukaryotic protein, HMG1 (17). Changes in complex formation by both major and minor groove-binding proteins and antibiotics induced by introduction of other modified bases have also been interpreted in terms of effects on DNA deformability by exocyclic substituents (18 -20).
Inferences about increased flexure of DNA containing modified bases stress the need for analysis of the associated structures and conformational energetics. Structural nuclear magnetic resonance analysis of an A-hmU base pair flanked by G-C base pairs indicates no significant deviations from classical B-form DNA (21), although nearest neighbors do influence the structure (22). 13 C or 31 P nuclear magnetic resonance relaxation experiments may provide the required insights into the dynamics of duplex DNA the size of a protein binding site, features that are only poorly extractable from measurements using techniques such as fluorescence polarization and DNA cyclization (23,24). The suggested locations of DNA bends in the TF1⅐DNA complex coincide with two hmU-A base pair steps. Considering the unique properties of the T-A step in terms of diversity of structure (25,26), it would be of particular interest to determine whether the inherent deformability of the T-A step is accentuated in hmU-A.
For a protein that bends DNA through a large angle (16), it is not surprising to find a greater affinity for DNA whose axis is kinked by unpaired bases (Fig. 5). Indeed, similar results have been previously reported for HU (27). We note that T-DNA constructs with pairs of suitably placed 4-nt loops or 3-nt bulge-loops generate the same affinity for wild-type TF1. However, different structures and energetics of deformation are indicated by the significantly higher affinity for bulge constructs of the TF1 mutant proteins TF1(F61S) and TF1(F61Q), which are impaired in DNA binding and bending. These observations are consistent with a static kink introduced by bulgeloops compared with a looser more flexible structure imposed by symmetric loops (8,9,24,28); we suggest that pre-bent DNA bulge-loop constructs also permit complex formation with certain TF1 mutant proteins, whereas binding and bending of DNA with sets of 4-nt loops (or hmU content) require a proteinmediated introduction and stabilization of DNA kinks.
Replacement of Phe-61 at the tip of DNA-binding arms with a polar, uncharged residue may render TF1 mutant proteins either incapable of producing specific DNA kinks, a process which would be aided locally by the presence of DNA bulgeloops, or deficient in securing the resulting DNA conformation, demands on which would be lessened in DNA with a predisposed orientation. If TF1, like IHF, employs a totally conserved Pro residue to intercalate between specific base pairs of the binding site (14), then the deficiency of TF1(F61S) and TF1(F61Q) in DNA binding and bending may reside in their inability to stabilize a severely bent DNA conformation. Phe-61 contributes to the striking preference of TF1 for hmU-DNA over T-DNA (15). TF1(F61R) also has high affinity for hmU-DNA but a reduced preference for hmU-DNA over T-DNA (15). In addition, IHF (which has higher affinity for a specific binding site in hmU-DNA compared with T-DNA (17)) relies heavily on indirect readout to recognize its binding sites (14). If TF1 and IHF engage their DNA targets in an isomorphous fashion, the preference for hmU-DNA characteristic of TF1 may not be due to specific interactions between Phe-61 and hmU residues of the TF1 binding site. Rather, the inability of TF1 to stabilize a T-DNA duplex that is severely kinked through intercalation of Pro residues may be alleviated by the more pliable hmU-DNA. On the other hand, the possibility that TF1 does use Phe-61 directly for inducing DNA kinks at two hmU-A base pair steps remains to be established or disproved by direct structure determination.