Contacts between the 5' nuclease of DNA polymerase I and its DNA substrate.

The 5' nuclease of DNA polymerase I (Pol I) of Escherichia coli is a member of an important class of prokaryotic and eukaryotic nucleases, involved in DNA replication and repair, with specificity for the junction between single-stranded and duplex DNA. We have investigated the interaction of the 5' nuclease domain with DNA substrates from the standpoint of both the protein and the DNA. Phosphate ethylation interference showed that the nuclease binds to the nucleotides immediately surrounding the cleavage site and also contacts the complementary strand one-half turn away, indicating that contacts are made to one face only of the duplex portion of the DNA substrate. Phosphodiester contacts were investigated further using DNA substrates carrying unique methylphosphonate substitutions, together with mutations in the 5' nuclease. These experiments suggested that two highly conserved basic residues, Lys(78) and Arg(81), are close to the phosphodiester immediately 5' to the cleavage site, while a third highly conserved residue, Arg(20), may interact with the phosphodiester 3' to the cleavage site. Our results provide strong support for a DNA binding model proposed for the related exonuclease from bacteriophage T5, in which the conserved basic residues mentioned above define the two ends of a helical arch that forms part of the single-stranded DNA-binding region. The nine highly conserved carboxylates in the active site region appear to play a relatively minor role in substrate binding, although they are crucial for catalysis. In addition to binding the DNA backbone around the cleavage point, the 5' nuclease also has a binding site for one or two frayed bases at the 3' end of an upstream primer strand. In agreement with work in related systems, 5' nuclease cleavage is blocked by duplex DNA in the 5' tail, but the enzyme is quite tolerant of abasic DNA or polarity reversal within the 5' tail.

DNA polymerase I (Pol I) 1 of Escherichia coli has an intrinsic 5Ј nuclease activity that is important for the removal of RNA primers from Okazaki fragments during lagging strand replication and for the removal of damaged nucleotides in DNA excision repair (1). Homologous 5Ј nuclease domains are found in most other bacterial Pol I enzymes and in some bacteriophages, where the polymerase and 5Ј nuclease exist as separate polypeptides (2,3). The bacterial 5Ј nuclease family also shows substantial sequence similarity to the FEN-1 eukaryotic nucleases, which are involved in various DNA transactions (4 -8).
Although the 5Ј nucleases were originally called 5Ј-3Ј exonucleases, these enzymes are more accurately described as structure-specific nucleases, with specificity for the junction between a base paired region and a single-stranded 5Ј overhang or "flap" (9 -11). Cleavage by these nucleases usually takes place between the first two paired bases at the junction between the duplex and the single-stranded 5Ј tail (see Fig. 1), although some variability has been observed (10 -13). Recent work suggests that the preferred substrate of the bacterial nucleases has a single unpaired base at the 3Ј end of the primer upstream of the 5Ј nuclease cleavage site (thus facilitating formation of a ligatable nick), and it seems likely that any variability in the observed cleavage position may reflect rearrangement via branch migration of the 5Ј and 3Ј singlestranded flaps (14,15).
Alignment of the sequences of the bacterial and bacteriophage 5Ј nucleases revealed nine invariant carboxylates (2,3), the majority of which are also conserved in the FEN-1 family (6,16). The large number of conserved carboxylates raises the possibility that the 5Ј nuclease reaction, like the polymerase and 3Ј-5Ј exonuclease reactions, may be catalyzed by divalent metal ions coordinated to the enzyme via carboxylate ligands (17). Moreover, mutagenesis studies in several systems have demonstrated the importance of the carboxylates in the 5Ј nuclease reaction (3, 18 -21).
Crystal structures have been determined for three of the prokaryotic 5Ј nucleases, the intrinsic 5Ј nuclease of Thermus aquaticus (Taq) DNA polymerase I (22), bacteriophage T4 RNase H (6), and bacteriophage T5 exonuclease (23), and for two archaebacterial FEN-1 analogs (24,25). The core structures of all five enzymes are very similar with the invariant carboxylates clustered in the central portion of the molecule, some coordinated to divalent metal ions, although the precise locations of the metal ions and the ligand geometries are somewhat variable (3,26).
In the absence of any crystalline complexes with bound DNA, little information exists on how the 5Ј nucleases interact with their substrate. A plausible candidate for a DNA-binding region in the prokaryotic 5Ј nucleases is a cluster of highly conserved and mostly basic residues. The relevant region of the protein structure is not well resolved in the Taq DNA polymerase and T4 RNase H structures, but in T5 exonuclease it is located on one side of an unusual helical arch structure (23). The two FEN-1 structures have a flexible loop in place of the helical arch (24,25). In all three structures the loop or arch is large enough to accommodate single-stranded DNA, suggesting that the single-stranded 5Ј end of the DNA substrate may be threaded through this part of the protein (Refs. 23, 24, and 26, see Fig. 2B), with the duplex DNA portion held close to the base of the loop or arch by a helix-loop-helix motif present in all five of the available structures (25). The threading model was originally proposed based on observations that 5Ј nuclease cleavage requires a free 5Ј end and is blocked by significant obstructions (e.g. annealed primers) in the single-stranded tail (3,10,27). However, recent work has shown that FEN-1 can tolerate a variety of modifications, including an 11-nucleotide branch, within the 5Ј flap DNA, prompting re-examination of the threading model (28). In this work, we have investigated the interaction between the 5Ј nuclease and its DNA substrate from the perspective of both the protein and the DNA, and have attempted to identify contacts between particular protein side chains and individual phosphodiester groups on the DNA.

EXPERIMENTAL PROCEDURES
Materials-Oligonucleotides for site-directed mutagenesis and for 5Ј nuclease substrates were synthesized by the Keck Biotechnology Resource Laboratory at Yale Medical School. Those used as reaction substrates were purified by gel electrophoresis. Oligonucleotide concentrations were determined spectrophotometrically using calculated extinction coefficients (29). DNA oligonucleotides containing abasic spacers or regions with reversed polarity were kindly provided by Dr. Daniel Kaplan, and have been described elsewhere (30). Radiolabeled nucleotides were from Amersham Pharmacia Biotech. N-Ethyl-N-nitrosourea was purchased from Sigma. DNase I was from Cooper Biomedical. T4 polynucleotide kinase, DNA ligase, and restriction endonucleases were from New England Biolabs or Roche Molecular Biochemicals and were used according to the accompanying instructions.
Mutagenesis and Protein Purification-Following our published methods (3), mutations in the 5Ј nuclease were constructed and subcloned into an expression plasmid for the 323-amino acid 5Ј nuclease domain, and the mutant proteins were purified by fast protein liquid chromatography (Amersham Pharmacia Biotech). To remove low levels of contamination by intact Pol I, the purification method was modified in either of two ways. When preparing 5Ј nuclease mutants on a small scale, the peak fractions from the MonoQ HR 16/20 column were assayed for polymerase activity (31), and only those fractions that had no measurable polymerase activity were combined and applied to the phenyl-Superose column. For large scale purification, a final gel filtration step was included. The 5Ј nuclease pool from the phenyl-Superose column was concentrated by ammonium sulfate precipitation, applied to a HiLoad 16/60 Superdex 200 prep grade column (120-ml bed volume) and eluted with 50 mM Tris-HCl, pH 7.5, 1 mM dithiothreitol, 100 mM NaCl. These modifications were necessary for the present study because low-level contamination by full-length Pol I interfered with DNase I footprinting experiments, particularly with weak binding 5Ј nuclease mutants. Because the DNA binding affinity of the polymerase domain is much greater than that of the 5Ј nuclease, competition between the two was a problem even though the amount of Pol I was not sufficient to affect enzyme activity assays and was not detectable on a Coomassie Blue-stained gel. We believe that the contaminating Pol I in our preparations was derived from the chromosomal polA locus rather than from readthrough of the amber codon in our expression plasmid for the 5Ј nuclease domain (3) because the problem was not alleviated by subcloning the 5Ј nuclease mutations into an analogous expression construct that lacked the complete polymerase coding region.
Kinetic Measurements-Single-turnover measurements of 5Ј nuclease cleavage were carried out using the substrates shown in Fig. 1a. The DNA oligonucleotides were labeled with 32 P at either the 5Ј or the 3Ј end (see legend to Fig. 1). Reactions contained Ϸ5 nM DNA substrate and the 5Ј nuclease at a series of 7-8 concentrations (typically from 0.1 to 40 M), chosen so as to bracket the K D value. All reactions were carried out at ambient temperature (23°C) in a buffer containing 50 mM Tris-HCl, pH 7.6, 5 mM MgCl 2 , 100 mM NaCl, and 5 M p(dT) 10 . The reactions catalyzed by the mutant 5Ј nuclease derivatives (requiring sampling at time intervals of Ն10 s) were conducted as described previously (3). The reaction catalyzed by the wild-type 5Ј nuclease was too fast for manual sampling, and was instead carried out on a rapid quench-flow instrument (KinTek Corp., Model RQF-3) by mixing a solution containing the DNA substrate with an equal volume of enzyme FIG. 1. DNA oligonucleotides used in this study. Hairpin structures are closed using a stable tetraloop (51). The expected positions of structure-specific cleavage by the 5Ј nuclease are indicated by arrows. a, oligonucleotides used in kinetic studies of the 5Ј nuclease. The derivative with a (dC) 10 5Ј tail (n ϭ 10) is the (22 ϩ 68)-mer used in most of the 5Ј nuclease assays in this study and our previous work (3). Oligonucleotides related to the 22-mer, that were used to generate substrates with 5Ј tails of different lengths, are listed. The 22-mer and 42-mer were labeled at the 5Ј end using T4 polynucleotide kinase and [␥-32 P]ATP. The other oligonucleotides were synthesized without the 3Ј-terminal C residue and were labeled at the 3Ј end using Klenow fragment and [␣-32 P]dCTP. The 5Ј termini of the shorter oligonucleotides were all phosphorylated; additionally, the non-phosphorylated 12-mer (12-mer-OH) was tested. In each case, the labeled oligonucleotide was annealed with Ϸ2-fold excess of the 68-mer template. b, double hairpin substrates used in DNA binding studies. These substrates were both 5Ј labeled with 32 P. Restriction sites that were used to generate markers are indicated. c, analogs of the 22-mer, in a, that were 5Ј labeled and annealed to the 68-mer hairpin, to generate 5Ј nuclease substrates with methylphosphonate groups ("M") at the indicated positions.
solution (both at twice the desired final concentration in the buffer described above) and quenching with 150 mM EDTA. Samples, obtained either manually or from rapid-quench experiments, were fractionated over a denaturing 10% polyacrylamide gel, and quantitated as described previously (3). For each set of reaction conditions, the enzyme concentration was much higher than that of the substrate, so that the reaction followed pseudo first-order kinetics. The observed rate constants (k obs ) were then plotted as a function of enzyme concentration (E 0 ), and the DNA binding constant, K D , and cleavage rate constant, k c , were obtained by fitting to the equation: k obs ϭ k c E 0 /(K D ϩ E 0 ). All determinations were carried out at least in duplicate. Details of the reaction conditions for 5Ј nuclease cleavage of modified substrates are given in the relevant figure legends.
Measurement of K D by DNase I Protection-Footprinting of the 112mer double-hairpin oligonucleotide (Fig. 1b), as a function of 5Ј nuclease concentration, was carried out as described previously (15,32). After gel fractionation, appropriate fragments were quantitated by phosphorimaging. For each enzyme concentration, the percent of uncomplexed DNA was calculated by determining the ratio of the radioactivity in a DNA band within the protected region to that in a band outside the protected region, and comparing with the same ratio from a control lane with no enzyme present. The binding constant, K D , equals the enzyme concentration at which 50% of the DNA is bound, which was determined by curve fitting. For this analysis to be valid, the DNA concentration in the reaction must be substantially below K D , so that the total enzyme concentration approximates the concentration of free enzyme. This condition was fulfilled because of the relatively high K D values for the 5Ј nuclease and its mutant derivatives.
CD Measurements-CD spectra were recorded at 20°C on a solution of wild-type or mutant 5Ј nuclease (Ϸ4 -10 M in 20 mM potassium phosphate, pH 7.0, 15 mM NaCl, 15% (v/v) glycerol) in a 2-mm path length quartz cuvette (110-QS, Hellma) which was placed in a thermostated sample holder in an Aviv 62 DS CD spectrometer (Lakewood, NJ). Thermal denaturation measurements were performed in the same buffer by monitoring the ellipticity at 222 nm as a function of temperature, over the range 10 -70°C in 1°C increments (samples were equilibrated for 2 min at each temperature). The data were processed as described (33), to give T m , the temperature at which the protein is 50% unfolded.
Phosphate Ethylation Interference-The procedure was modified from published procedures (34,35). The 5Ј end-labeled 132-mer doublehairpin substrate (Fig. 1b) was used to analyze phosphate contacts close to the cleavage site. To Ϸ2 pmol of the DNA substrate in 0.1 ml of 50 mM sodium cacodylate, pH 8, was added 0.1 ml of ethanol saturated with ethylnitrosourea at 50°C. After incubation at 50°C for about 25 min, the ethylnitrosourea was removed by five extractions with 1-ml portions of water-saturated ether. The ethylated DNA solution (Ϸ0.1 ml) was adjusted to 50 mM Tris, pH 7.5, 5 mM MgCl 2 , and incubated at room temperature with wild-type 5Ј nuclease (Ϸ200 pmol). One-third of the reaction mixture was quenched with an equal volume of formamidedyes containing 30 mM EDTA, at time intervals chosen so as to obtain three samples having Ϸ25, 50, and 75% product formation, respectively. The 36-mer product was separated from unreacted 132-mer substrate in each sample by electrophoresis on an 8% polyacrylamide gel containing 40% (v/v) formamide and 7 M urea. The DNA bands were located by autoradiography, excised, eluted into 240 l of 10 mM Tris-HCl, pH 8.0, 0.1 mM EDTA, containing 0.5 M NH 4 OAc, and concentrated by ethanol precipitation. The resulting DNA samples were cleaved at the ethylated phosphate positions by heating for 30 min at 90°C in 10 l of 10 mM sodium phosphate, pH 8.0, 1 mM EDTA, containing 75 mM NaOH. A portion (up to 4 l) of each sample was mixed with 2 l of formamidedyes, and fractionated on a 12% polyacrylamide-urea gel. Size markers were generated by restriction enzyme digestion (see Fig. 1b), and by limited chemical cleavage at guanine residues (36). Data analysis is described in the legend to Fig. 3. To analyze phosphate contacts on the DNA strand opposite the cleavage site, the same procedure was carried out using the 112-mer, and analyzing samples from the pool of unreacted substrate on an 8% polyacrylamide gel containing 40% (v/v) formamide and 7 M urea.

RESULTS
Mutant 5Ј Nuclease Derivatives-To determine which amino acid side chains of the 5Ј nuclease are involved in binding the DNA substrate, we studied mutations in two groups of highly conserved residues (Table I). One was the nine carboxylates, shown to be crucial for 5Ј nuclease cleavage in our previous study (3). We also made alanine substitutions at Arg 20 , Arg 70 , Tyr 77 , Lys 78 , and Arg 81 ( Fig. 2A); in the structure of the T5 5Ј nuclease the equivalent residues define the base of the helical arch that is proposed to form part of the DNA-binding site (23) (Fig. 2B). The effects of all these mutations were studied in context of the separate 5Ј nuclease domain (residues 1-323 of Pol I) so as to avoid ambiguities due to the DNA binding properties of the other domains of Pol I.
Cleavage of our standard (22 ϩ 68)-mer substrate ( Fig. 1) by the wild-type and mutant 5Ј nuclease derivatives was measured under single-turnover conditions (excess enzyme) as a function of enzyme concentration, giving the enzyme-DNA dissociation constant (K D ) and the maximum cleavage rate (k c ), which reflects the rate of steps up to and including phosphoryl FIG. 2. Structural features of the 5 nuclease family, illustrated from the T5 nuclease structure (23). A, location of residues in T5 5Ј nuclease homologous to those mutated in this study. These are: Arg 33 (20), Arg 75 (70), Tyr 82 (77), Lys 83 (78), and Arg 86 (81), where the residue number in the E. coli enzyme is in parentheses. The nine highly conserved carboxylates are also shown in pink and red, but not labeled. Those colored red correspond to the carboxylates whose removal resulted in an increase in DNA binding affinity. The side chain of Asp 204 (corresponding to Asp 188 , mentioned in the text) is located closest to Tyr 82 . This figure was made using Ribbons (52). B, hypothetical model of T5 5Ј nuclease bound to a DNA molecule with a 5Ј single-stranded tail (23). The two green spheres represent divalent metal ions bound at the enzyme active site. This figure was kindly provided by Dr. Jon Sayers. transfer (Table II). All of the mutations tested caused substantial decreases in cleavage rate, ranging from Ϸ20-fold (R20A and R81A) to Ͼ5 ϫ 10 5 -fold (D115A). R20A and K78A were the only mutations that caused a significant decrease in DNA binding affinity; some of the carboxylate mutations, particularly D63A, D138A, and D185A, caused an increase in DNA binding affinity. For a few 5Ј nuclease mutants, the DNA binding affinity was measured by a DNase I footprint titration; this showed good agreement with the kinetic determinations and had the added advantage that the measured binding constant is related to a binding site which can be visualized on a gel (Ref. 15 gives details of the 5Ј nuclease footprint). Footprinting was clearly the preferred method for the D115A mutant protein, which had no measurable activity in the 5Ј nuclease assay; conversely, footprinting was not appropriate for mutants with relatively high nuclease activity which degraded the DNA substrate during the experiment.
Circular dichroism (CD) measurements provided two lines of evidence that the 5Ј nuclease mutations being studied did not cause any major changes in protein three-dimensional structure. First, the CD spectra of the mutant proteins were similar to that of wild-type, implying that the same secondary structure components are present in all. Second, the denaturation temperatures of the mutant proteins were all within a few degrees of the T m for the wild-type 5Ј nuclease (Table II). Mutations in surface residues on the proposed helical arch region gave T m values very close to the wild-type value, as observed previously for mutations in surface residues on the polymerase domain of Klenow fragment (33). The T m values for the carboxylate mutants were more variable; perhaps reflecting subtle changes in protein conformation and stability on altering negatively charged side chains (and metal ligands) in the central portion of the 5Ј nuclease domain.
Phosphate Ethylation Interference-We used phosphate ethylation interference to define contacts on the DNA substrate that are important in binding to the 5Ј nuclease. Our basic experimental design involved treatment of a labeled doublehairpin DNA substrate ( Fig. 1b) with ethylnitrosourea under conditions that gave no more than one ethylation of backbone phosphodiester groups per DNA molecule. The pool of ethylated molecules was then used as the substrate for cleavage by the 5Ј nuclease, and the labeled product was separated from uncleaved substrate by gel electrophoresis. Those molecules in which ethylation interfered with the 5Ј nuclease reaction should be under-represented in the product molecules and enriched in the population of unreacted substrate molecules (Fig.  3A). Chemical cleavage and gel electrophoresis then revealed the positions of ethylation in each population. Analysis of the non-cleaved substrate pool showed that phosphate ethylation was most inhibitory at the cleavage site and the immediately flanking positions (ϩ1 and Ϫ1); weaker inhibition was seen at neighboring positions on the "top" strand (ϩ2, Ϫ2, and Ϫ3) (Fig. 3, C and E). The product pool was less informative because the full-length product band obscured the most relevant posi- a Kinetic measurements were made using the (22 ϩ 68)-mer substrate (Fig. 1a). b DNase I protection was carried out using the 112-mer substrate (Fig. 1b). Small differences in binding affinity between the two substrates used in K D measurements are to be expected because of their different sequences, and because the (22 ϩ 68)-mer has a nick and the 112-mer has a single base gap upstream of the cleavage site (see also Ref. 15).
c For the three proteins tested, the CD spectra in the presence and absence of Mg 2ϩ were superimposable. d Use of the rapid-quench-flow instrument gave a more accurate measurement of the reaction rate for the wild-type 5Ј-3Ј exonuclease, and indicated that this rate is about 10-fold faster than that reported in our previous study. e The number of determinations is shown in parentheses. f ND, not determined; however, the high level of 5Ј nuclease activity argues against any significant alteration to the structure. g The reaction was too slow to measure.
tions. On the opposite strand, inhibitory ethylation positions were located 6 and 7 bases away from the cleavage site (Fig. 3, B, D, and E); however, ethylation immediately opposite the cleavage site did not affect the reaction. Methylphosphonate Modifications-To investigate contacts between individual phosphodiester positions on the substrate and side chains on the 5Ј nuclease, we used methylphosphonate-substituted DNA oligonucleotides in conjunction with mutations in the proposed helical arch region (R20A, R70A, Y77A, K78A, and R81A). Unlike phosphate ethylation, which adds a bulky alkyl group to the DNA phosphodiester backbone, the methylphosphonate modification removes a negative charge on the phosphodiester backbone without substantially increasing the size of the substituent. We tested the phosphate positions flanking the exonuclease cleavage site that corresponded to positions of inhibitory ethylations (Ϫ3, Ϫ2, Ϫ1, and ϩ1 positions), using a series of oligonucleotides, each having a methyl group on a single backbone phosphate position (Fig. 1c). A methylphosphonate at the Ϫ7 position served as a negative control. Because the methylphosphonate modification was introduced by chemical synthesis, each oligonucleotide consisted of approximately equal amounts of the R p and S p methylphosphonate diastereomers.
Cleavage of the methylphosphonate substrates by the wildtype and mutant 5Ј nucleases was measured initially at two different enzyme concentrations: an enzyme concentration at least 3-fold greater than K D (rate constant reflects k c ) and an enzyme concentration less than 0.5 ϫ K D (rate responds to k c / K D ). At both concentrations the MP Ϫ2 and MP Ϫ3 analogs gave reaction rates indistinguishable from those of the MP Ϫ7 and unmodified (22 ϩ 68)-mer controls, and were therefore not studied further. The MP Ϫ1 substrate was cleaved by the wild-type 5Ј nuclease about 5-fold more slowly than the unmodified substrate, based on initial rates (Fig. 4A). Cleavage was noticeably biphasic, presumably reflecting different rates of cleavage of the two methylphosphonate diastereomers; however, the rates of the two phases were sufficiently different to allow determination of the kinetic parameters for the fast phase (Table III). For some of the mutant proteins the two phases were more similar in rate, and this precluded a detailed kinetic analysis. Nevertheless, some interesting trends were apparent from a comparison of the cleavage of unmodified and MP Ϫ1 substrates by mutant 5Ј nuclease derivatives (Fig. 4A). The R20A and R70A mutant proteins were similar to the wild-type 5Ј nuclease, with cleavage of the methylphosphonate substrate about 10 -20-fold slower than the normal substrate. By contrast, the Y77A, K78A, and R81A proteins cleaved the methylphosphonate substrate more rapidly than they cleaved the unmodified substrate; the effect was largest for the K78A mutant.
All substrates with methylphosphonates 5Ј to the cleavage site (including MP Ϫ1 ) gave essentially the same pattern of cleavage  Fig. 1b were used, and gave similar results. The 132-mer was preferable for studying the "top" strand around the cleavage site because its 35-nucleotide 5Ј extension ensured that the relevant phosphate positions were sufficiently far from the labeled 5Ј end that the two hydrolysis products (phosphorylated and non-phosphorylated) derived from each ethylation position (34) would have very similar gel mobilities. The 112-mer substrate, which has a shorter 5Ј-tail, gave better resolution for investigating the "bottom" strand opposite the cleavage site. products as the unmodified substrate, with the 11-mer as the predominant product (Fig. 4B). 2 By contrast, a methylphosphonate 3Ј to the cleavage site (MP ϩ1 substrate) changed the position of cleavage by the wild-type 5Ј nuclease. With the unmodified (22 ϩ 68)-mer, the major product was the 11-mer; this is attributed to cleavage between the two paired bases immediately following the 5Ј single-stranded tail (Fig. 5). The 10-mer was a minor product, amounting to a few percent of the total. By contrast, with the MP ϩ1 substrate, the yield of 10-mer was about 1.5 times the amount of 11-mer. A likely explanation is that the methylphosphonate substitution inhibited cleavage at the normal position, allowing a minor cleavage site to predominate. This idea is consistent with the observation that cleavage of the MP ϩ1 substrate was 40 -50-fold slower than for the unmodified (22 ϩ 68)-mer. The R20A mutation reversed the effect of the MP ϩ1 substitution on the product distribution, so that cleavage generated about four times more 11-mer than 10-mer (Fig. 5). When 2 There were some minor differences: the MP Ϫ1 substrate gave a small amount of product longer than 11 nucleotides, presumably reflecting a preference for placing the methylphosphonate further from the cleavage site. Also, as would be expected, neither the MP Ϫ1 nor MP Ϫ2 substrates gave detectable amounts of 10-mer, normally seen as a minor cleavage product with the normal substrate. Assuming cleavage always take place between two bases in a duplex region immediately 3Ј to the single-stranded 5Ј overhang, we infer that the structures shown give rise to the 11-and 10-mer products, respectively. * marks the methylphosphonate modification.  Fig. 1a). Except for the 12-mer-OH, all were 5Ј phosphorylated.
b The kinetic determinations were carried out at least in duplicate, except for the determination of k c and K D for R81A cleavage of the 12-mer and 12-mer-OH, which came from a single experiment. However, the Ͼ10-fold difference in the rate of cleavage by R81A of the phosphorylated and nonphosphorylated 12-mer was confirmed independently.
c ND, not determined because the plot of rate versus enzyme concentration did not plateau even at the highest concentrations attainable. From the slope at low enzyme concentration, k c /K D ϭ 9.7 ϫ 10 3 min Ϫ1 M Ϫ1 , which is Ϸ10 3 -fold lower than for the shorter substrates.
FIG. 6. Effect of an upstream 3 flap on 5 nuclease cleavage. A, gel fractionation of the products of cleavage by the wild-type 5Ј nuclease of a series of substrates (shown in B) having different lengths of 3Ј flap. The strand with the 5Ј overhang was labeled at the 5Ј end and then annealed to a 2-fold excess of the template and a 4-fold excess of the appropriate upstream primer. Cleavage was allowed to proceed for the indicated times in a reaction mixture containing 10 nM DNA (labeled strand) and Ϸ50 M 5Ј nuclease (wild-type or mutant) in 50 mM Tris-HCl, pH 7.9, 5 mM MgCl 2 , 0.1 M NaCl. In these experiments, about 60% of the substrate remained uncleaved, presumably because of incomplete annealing of the labeled strand. The identity of the 15-mer cleavage product was confirmed using a chemically synthesized marker. B, product analysis of the reactions shown in A, as well as similar reactions carried out with the D188A and R20A mutant derivatives. The amount of each labeled product band, as a percentage of the total product, is plotted on a bar graph. On the left are indicated the substrate structures expected in the absence of any rearrangement. Using the reasoning explained in the text, we deduced the substrate configurations expected to give rise to the major products in each reaction, and these are shown to the right of the graphs. indicates the position of 5Ј nuclease cleavage. other mutants were tested with the MP ϩ1 substrate, D188A resembled R20A in giving predominantly the 11-mer product, but most had too little activity on the MP ϩ1 substrate to allow assignment of the position of cutting.
Length of the 5Ј Flap-Comparing a series of oligonucleotides having 5Ј overhangs of different lengths (Fig. 1a), the kinetics of cleavage were relatively insensitive to changes in length between 0 and 3 nucleotides, showing only a gradual increase in binding affinity with 5Ј flap length (Table III). With a 30nucleotide tail, the reaction rate did not plateau even at 50 M 5Ј nuclease, and the apparent k c /K D value was Ϸ10 3 -fold lower than for the substrates with shorter 5Ј tails. This could reflect either weaker binding of the 5Ј nuclease to the substrate with the 30-nucleotide tail, or slower formation of the catalytically competent complex with the DNA branch point at the nuclease active site. The presence or absence of a phosphate at the 5Ј end of a fully annealed downstream DNA strand (compare 12-mer with 12-mer-OH) did not affect cleavage by the wild-type 5Ј nuclease, or by the R70A, Y77A, and K78A mutants. By contrast, the R81A mutant cleaved the phosphorylated substrate Ϸ10-fold more slowly than the 5Ј-hydroxyl substrate; the k c for the non-phosphorylated 12-mer was similar to the value for the standard (22 ϩ 68)-mer, whereas the value for the phosphorylated 12-mer was unexpectedly low.
Effect of Unpaired Bases at the Primer Terminus (a 3Ј Flap)-The wild-type 5Ј nuclease was assayed on a series of substrates designed so that they differed only in the length of the unannealed region at the 3Ј end of the primer upstream of the 5Ј nuclease cleavage site (Fig. 6). The best substrates were those designed with one or two unannealed bases at the primer 3Ј end; these were cleaved slightly faster than the substrates with 0 (nicked substrate) or three unannealed bases (Fig. 6A). Cleavage of substrates with even longer 3Ј flaps (4 and 7 nucleotides) was at least 10-fold slower than cleavage of the 0-or 3-nucleotide 3Ј flap substrates and similar to that of a substrate without a primer upstream of the 5Ј nuclease cleavage site. The sizes of the 5Ј nuclease cleavage products indicated that some of the substrates, when bound to the 5Ј nuclease active site, may have adopted base pairing configurations that were different from the simple arrangements originally assumed (listed on the left of Fig.  6B). Assuming that the 5Ј nuclease always cuts between the two paired bases immediately adjacent to the 5Ј tail, the substrate structures shown to the right of the bar graphs in Fig. 6B could be deduced, as follows (note that not all of the base pairs shown are complementary in the Watson-Crick sense). In the absence of an upstream primer, the wild-type 5Ј nuclease gave approximately equal parts of the expected 15-mer, and the 16-mer corresponding to melting of an additional base pair of the duplex. With the nicked substrate, the products consisted of approximately equal amounts of 15-and 14-mer, the latter attributed to rearrangement to give a one-nucleotide 3Ј flap. The substrates with 1-and 2-nucleotide 3Ј flaps behaved as would be expected if no rearrangement took place, yielding almost exclusively the 15-mer on cleavage with the wild-type nuclease. By contrast, the substrate with the 3-nucleotide flap gave mainly 16-mer, implying that 5Ј nuclease cleavage took place preferentially on molecules that had rearranged to give a 2-nucleotide 3Ј flap. The 4-nucleotide flap substrate showed some rearrangement to a 3-nucleotide flap so that cleavage gave a mixture of 15-and 16-mer, whereas the 7-nucleotide flap showed less tendency to rearrange, yielding predominantly the 15-mer.
The R20A and D188A mutations influenced the pattern of cleavage products in different ways (Fig. 6B). 3 On all four substrates tested, D188A gave longer products than wild-type, regardless of the presence of an upstream primer. The behavior of R20A was more complicated; this 5Ј nuclease mutant gave shorter products than wild-type in the absence of an upstream primer, but the cleavage pattern was only subtly different from that of the wild-type enzyme on substrates having an upstream primer.
Modifications to the 5Ј Flap-Since previous studies had shown eukaryotic FEN-1 to be tolerant of a wide range of modifications within the single-stranded 5Ј tail, we tested the FIG. 7. Effect of 5 flap modifications. A, DNA substrates. The oligonucleotides containing the 5Ј flap were synthesized without the 3Ј-terminal C, and were labeled, after annealing to the complementary template, using Klenow fragment and [␣-32 P]dCTP. The labeled strand was purified on a denaturing gel, and then annealed sequentially to excess of the template strand, followed by the upstream primer. The relative amounts of the three DNA strands were determined empirically so as to give a high yield of the desired substrate, as judged by electrophoresis on a non-denaturing gel. "d" indicates a stable abasic site. B, gel electrophoretic analysis of 5Ј nuclease cleavage of oligonucleotides with 5Ј flap modifications. The reactions contained Ϸ12 nM DNA (expressed as the concentration of the labeled strand) and 2.5 M, 250 nM, or 25 nM wild-type 5Ј nuclease in 50 mM Tris-HCl, pH 7.5, 5 mM MgCl 2 , 0.1 M NaCl. In each panel, the leftmost lane, marked "0," shows unreacted substrate. The remaining six lanes show cleavage of the indicated substrate at the three nuclease concentrations used. Each pair of lanes under a triangular wedge corresponds to samples which were removed and quenched after 15 s and 1 min, respectively. Using appropriate markers, we have demonstrated that the first 5Ј nuclease cut is as indicated by the arrow in A. effect on the E. coli Pol I 5Ј nuclease of some fairly substantial changes to the 5Ј tail (Fig. 7). Cutting was not significantly affected if the entire single-stranded tail had the wrong polarity, with a 5Ј-to-5Ј linkage immediately adjacent to the cleavage site. Abasic deoxyribose spacers at either end of the 5Ј tail also had very little effect, except that, when the abasic sites were immediately 5Ј to the dinucleotide that defines the cut site, the cleavage rate was more sensitive to decreasing enzyme concentration (Fig. 7B), implying that the modification interfered with binding to the 5Ј nuclease. If the abasic residues were three nucleotides further away from the cleavage site, their effect on the 5Ј nuclease reaction was negligible (data not shown). By contrast, cleavage was completely blocked by duplex DNA within the 5Ј flap strand. The substrate in which the 5Ј strand had a terminal hairpin (Fig. 7A) was refractory to 5Ј nuclease cutting unless the hairpin had been pre-cleaved with EcoRI (data not shown).

DISCUSSION
Contacts on the DNA-Both the phosphate ethylation interference and methylphosphonate substitution approaches demonstrated that the most significant contacts to the phosphodiester backbone of the DNA substrate are the positions immediately adjacent to the 5Ј nuclease cleavage site (Ϫ1 and ϩ1). In the ethylation experiment, the inhibitory positions on the opposite DNA strand indicated that the enzyme contacts one face only of the DNA helix downstream of the cleavage site (Fig. 3E), and the position of the proposed contacts is entirely consistent with the conceptual model of T5 exonuclease bound to DNA (23) (Fig. 2B). Because ethylation increases the steric bulk of the modified phosphate, we believe that the inhibitory positions on the opposite DNA strand merely indicate a region of close approach between enzyme and substrate. By contrast, ethylation at the positions flanking the cleavage site may have compromised a binding interaction by removing a negative charge. Single methylphosphonate substitutions at the ϩ1 and Ϫ1 positions provided further information on phosphate contacts without complications due to steric effects. The biphasic kinetics of cleavage of the MP Ϫ1 substrate shows that substitution of one of the non-bridging oxygens is mildly deleterious (Ϸ4-fold decrease in k c ) whereas substitution of the other causes a much larger decrease in reaction rate.
The rather small effects of removing the 5Ј tail nucleotides are consistent with the data for methylphosphonate substitution at the Ϫ2 and Ϫ3 positions, and make biological sense since the 5Ј nuclease must be able to process DNA molecules having a variety of 5Ј flap lengths. The insensitivity to the presence of a 5Ј-terminal phosphate at a nick might seem surprising, given the effect of a methylphosphonate linkage immediately 5Ј to the cleavage site. However, a terminal phosphate differs from a phosphodiester linkage in overall negative charge, and may be bound somewhat differently to the active site when not attached to a single-stranded DNA tail. The 5Ј nuclease of Taq DNA polymerase is similarly insensitive to the phosphorylation status of a fully annealed (i.e. nicked) substrate (38), although eukaryotic FEN-1 has a strong preference for a 5Ј phosphate (27,39).
Comparison of DNA molecules with various lengths of 3Ј flaps indicates that the 5Ј nuclease does indeed have a binding site for unpaired nucleotides at the 3Ј end of a primer upstream of the cleavage site. The interpretation of these experiments hinges on the assumption that the DNA substrate is bound at the 5Ј nuclease active site in a consistent manner such that cleavage always takes place between two bases presented in a paired configuration adjacent to the 5Ј single-stranded tail. The observed product(s) can then be used to infer the base pairing arrangement(s) of each substrate. We and others have found this approach to provide a convincing interpretation of the products observed in previous studies (14,15). Our current data indicate that the 3Ј flap-binding site is optimized for one or two nucleotides. The oligonucleotides designed to have a 1or 2-nucleotide 3Ј flap were the preferred substrates, and the predominant product in either case did not require rearrangement from the most stably base paired structure. Other substrates tended to rearrange toward these favored structures; the nicked substrate gave a substantial amount of product implying rearrangement to a substrate having a single nucleotide 3Ј flap, and the substrate with a 3-nucleotide 3Ј flap gave primarily the product derived after rearrangement to a 2-nucleotide 3Ј flap. The formation of products derived from longer 3Ј flaps indicated that these substrates can be accommodated within the 5Ј nuclease active site; the slower reaction rate implied that this arrangement is not optimal, while rearrangement to give a shorter flap would require too many base pairs to be sacrificed.
Protein Contacts to the DNA Phosphates-The kinetic data (Table II) indicated that side chains at either end of the putative helical arch (Arg 20 , Arg 70 , Tyr 77 , Lys 78 , and Arg 81 ) play an important role in the 5Ј nuclease reaction, as has been found in other homologs (19, 40 -43). Additionally, three of our experiments provided data linking individual protein side chains with particular positions on the bound DNA substrate. In each case the evidence came from a comparison of the reaction of wild-type and mutant 5Ј nuclease(s) with two different substrates. In general terms, an observation that does not conform to expectations can indicate proximity, and possibly a functionally significant interaction, of the mutated side chain with the modified position on the substrate.
Our data with the MP Ϫ1 methylphosphonate substrate provided evidence for an interaction of Lys 78 and the surrounding region of the protein with the phosphodiester immediately 5Ј to the cleavage site. The Y77A, K78A, and R81A mutant proteins cleaved the MP Ϫ1 substrate more rapidly than the normal substrate, reversing the trend seen with the wild type 5Ј nuclease (Fig. 4). Preference of a mutant protein for a non-cognate substrate may seem hard to rationalize; the naïve expectation is that a protein mutation that removes a residue interacting with a particular phosphate oxygen should make the mutant protein insensitive to the replacement of that oxygen with a methyl group. However, if, following Fersht and colleagues (44), we take account of interactions with solvent (H 2 O) and consider the entire inventory of interactions made and broken when the enzyme interacts with its substrate, it can be seen that the combination of mutant protein and non-cognate substrate may be favored because the energetic cost of desolvating the interacting groups will be least. The effect on cleavage of the methylphosphonate substrate was greatest for the K78A mutant, suggesting that Lys 78 makes the most important interaction with the Ϫ1 phosphodiester position, while Tyr 77 and Arg 81 have lesser roles, perhaps in orienting the substrate or the Lys 78 side chain.
The comparison of the 5Ј-phosphorylated and 5Ј-hydroxyl nicked substrates provides further evidence placing Arg 81 close to the 5Ј side of the cleavage position. The R81A mutant 5Ј nuclease cleaved the phosphorylated substrate much more slowly than the non-phosphorylated substrate, whereas the wild-type 5Ј nuclease and other mutants, including K78A, cleaved both substrates at similar rates. A plausible explanation is that Arg 81 would normally neutralize the charge on the terminal 5Ј phosphate. As noted above, the 5Ј phosphate on a nicked substrate carries a greater negative charge than a backbone phosphodiester and may be bound differently at the 5Ј nuclease active site. This could explain why R81A had the greatest effect with the 5Ј phosphate, but K78A had the greatest effect in the MP Ϫ1 comparison.
Data from the MP ϩ1 methylphosphonate substitution suggested that Arg 20 may be close to the phosphodiester on the 3Ј side of the cleavage site. With the wild-type 5Ј nuclease, the MP ϩ1 modification decreased the cutting at the expected position to such an extent that a normally minor cleavage pathway became the dominant pathway. As shown in Fig. 5B, this second cleavage pathway can be explained by rearrangement of the DNA junction to an alternative (and less favored) base pairing arrangement. Cleavage of the MP ϩ1 substrate by the R20A mutant protein occurred in the same position as cleavage of the normal substrate by the wild-type 5Ј nuclease, suggesting two possible roles for Arg 20 . One is that Arg 20 interacts with the ϩ1 phosphodiester, providing the major discrimination against the MP ϩ1 substitution. In the absence of the Arg 20 side chain this discrimination is lost and the perfectly base paired configuration of the substrate (left side of Fig. 5B) dominates the reaction, giving the 11-mer. Alternatively, Arg 20 may be part of the binding site for an unpaired 3Ј flap; removal of this side chain would then disfavor the alternative base pairing arrangement shown on the right of Fig. 5B and tip the balance in favor of the 11-mer product. We prefer the former explanation because the R20A mutation did not have any unusual effect on the rate of cleavage of substrates with a 3Ј flap; rather, the rates of cleavage by both the wild-type and R20A nucleases were stimulated to a similar degree by the presence of a 1-or 2-nucleotide 3Ј flap (data not shown).
The Active Site Carboxylates-Our previous study showed that the nine invariant carboxylates are important for the 5Ј nuclease reaction but did not establish their individual roles (3). The results described here argue against any of the carboxylates being involved in DNA binding. Three mutations, D63A, D138N, and D185A, caused an increase in DNA binding affinity; as in the polymerase active site of Klenow fragment (32), this may indicate positions where removal of a negatively charged side chain close to the path of the phosphodiester backbone enhances binding of the DNA substrate, although not necessarily in a catalytically appropriate conformation. The side chains of Asp 63 , Asp 138 , and Asp 185 (shown in red in Fig.  2A) may therefore indicate regions of close approach of the phosphodiester backbone.
The original question then remains: why do nine carboxylates appear to be crucially important in the 5Ј nuclease reaction, even though the precedents of the polymerase and 3Ј-5Ј exonuclease active sites suggest that a smaller number of acidic side chains should suffice to coordinate the metal ions necessary for a two-metal-ion phosphoryl transfer mechanism (17,(45)(46)(47)? Our data do not completely rule out the possibility that a subset of these residues might interact with DNA, presumably via coordinated metal ions. A DNA binding role could have gone undetected in our experiments either because removal of a single carboxylate ligand was not sufficient to prevent binding of a metal ion under our assay conditions, or because the putative protein-DNA interaction is felt in the transition state and does not contribute to ground-state binding. An alternative scenario, that a subset of the carboxylates may appear to be important in the 5Ј nuclease reaction because they serve to maintain the appropriate geometry in the active site region, is consistent with the rather variable T m values for some of the carboxylate mutants and for the increased stability of the 5Ј nuclease domain in the presence of Mg 2ϩ (Table II). It might also explain why the phenotypes of the carboxylate mutants, while broadly consistent in the different nucleases that have been studied, are not always in complete agreement. 4 Moreover, if the conformation of the active site region is rather delicately balanced and therefore easily influenced by external factors, then this might account for the inconsistencies in coordination geometry and metal-metal distances seen in the 5Ј nuclease crystal structures (3,6,(22)(23)(24)26).
Does the 5Ј Nuclease Play an Active Role in Strand Displacement?-When the R20A and D188A mutant 5Ј nucleases cleaved a 5Ј flap substrate, the distribution of products differed from that of the wild-type enzyme (Fig. 6). These differences were particularly evident on a DNA substrate that lacked an upstream primer, ruling out any explanations based on interaction with the 3Ј end of the primer. The data therefore suggest that the amount of melting at the junction of the 5Ј flap and the DNA duplex is not simply a property of the DNA substrate, but that the 5Ј nuclease may play an active role in opening the downstream duplex. The R20A mutant tended to give shorter products than the wild-type nuclease, implying a decrease in strand displacement activity, which may be related to the proposed interaction between Arg 20 and the DNA 3Ј to the cleavage site. Conversely, the D188A mutant gave longer products, suggesting enhanced strand displacement activity. The D188A protein also gave longer cleavage products with the MP ϩ1 substrate, and we therefore infer that the MP ϩ1 result can be attributed to the strand displacement activity of D188A rather than to a specific interaction of Asp 188 with the ϩ1 phosphate position.
Interactions with the 5Ј Flap and Mechanism of DNA Binding-The Pol I 5Ј nuclease requires a free 5Ј end on the unpaired single-stranded tail or flap, as illustrated by the inhibition of 5Ј nuclease cleavage by a terminal hairpin in this work, and by the loop substrates described in our previous work (3). Based on similar observations with other structure-specific nucleases, it has been proposed that the 5Ј tail must be threaded through the enzyme (10,27), and the loop or arch structures seen crystallographically fit well with this idea, being large enough to bind single-stranded, but not duplex, DNA (24 -26). In several publications (26 -28), the proposed mechanism of DNA binding has the 5Ј nuclease entering via the free 5Ј end of the single-stranded overhang or flap, and sliding to the cleavage point at the junction between single-and double-stranded regions (like threading a bead onto a string). However, the data do not exclude other mechanisms of forming a catalytically competent enyzme-DNA complex, such as binding to the junction and then inserting the single-stranded tail through the loop or arch of the enzyme (like threading a needle).
We have shown that cleavage of a substrate with a 30nucleotide 5Ј tail is much less efficient than cleavage of shorter tails (Table III), and we believe this indicates a situation in which formation of the catalytically competent complex is ratelimiting. Although the less efficient cleavage of the 30-nucleotide tail could also be attributed to weaker (rather than slower) binding of the 5Ј nuclease, we do not favor this explanation because it is hard to understand why the binding affinity of a cleavage site adjacent to a 30-nucleotide tail should be substantially different from that with a 10-nucleotide tail. If formation of the complex is indeed rate-limiting for the substrate with the 30-nucleotide tail, then this would favor the second model for enzyme-DNA binding (threading the 5Ј strand through the enzyme after binding to the junction). Our reasoning is that sliding along a 30-nucleotide tail should not be dramatically slower than sliding along a 10-nucleotide tail, whereas, with the alternative threading model, one might anticipate a cut-off between short-tailed substrates, whose binding would be trivially simple, and longer substrates, whose binding could present considerable entropic problems.
Regardless of the mechanism by which the 5Ј single-stranded tail might be threaded through the nuclease, this process cannot involve close monitoring of the structure of the tail, since our data indicate that cleavage is remarkably insensitive to structural modifications. Similarly, mammalian FEN-1 tolerates a variety of modifications (27,28), and is inhibited only by those modifications which cause a substantial increase in size (bound protein, or duplex DNA), or which decrease flexibility (some cis-platinum adducts). The requirement for flexibility also supports the "needle-threading" model described above. More puzzling is the observation that FEN-1 can cleave a substrate with a branch within the 5Ј flap strand (28); this raises doubts about threading mechanisms in general, although it is difficult to envisage an alternative model that accounts for the requirement for a free 5Ј end. In vivo the question of threading may be moot, however, since a substantial fraction of the intermediates encountered by the structure-specific nucleases, particularly during lagging strand replication, are likely to have very short 5Ј overhangs of only one or two nucleotides (9, 48 -50).