Fidelity and Processivity of DNA Synthesis by DNA Polymerase κ, the Product of the Human DINB1 Gene*

Mammalian DNA polymerase κ (pol κ), a member of the UmuC/DinB nucleotidyl transferase superfamily, has been implicated in spontaneous mutagenesis. Here we show that human pol κ copies undamaged DNA with average single-base substitution and deletion error rates of 7 × 10−3 and 2 × 10−3, respectively. These error rates are high when compared to those of most other DNA polymerases. pol κ also has unusual error specificity, producing a high proportion of T·CMP mispairs and deleting and adding non-reiterated nucleotides at extraordinary rates. Unlike other members of the UmuC/DinB family, pol κ can processively synthesize chains of 25 or more nucleotides. This moderate processivity may reflect a contribution of C-terminal residues, which include two zinc clusters. The very low fidelity and moderate processivity of pol κ is novel in comparison to any previously studied DNA polymerase, and is consistent with a role in spontaneous mutagenesis.

The recently discovered UmuC/DinB nucleotidyl transferase superfamily of DNA polymerases (1-3) includes a subfamily whose members share extensive amino acid sequence homology with the Escherichia coli dinB gene product. The dinB gene is required for untargeted mutagenesis of phage (4), and overexpression of dinB in E. coli increases the spontaneous mutation rate in plasmids, especially for single-base deletions in a run of guanine residues (5). The dinB gene encodes DNA polymerase (pol) IV, a distributive enzyme that lacks detectable 3Ј35Ј exonuclease activity (6). pol IV has limited ability to bypass UV radiation-induced photoproducts, but misinserts nucleotides at undamaged template sites at rates that are higher than those observed for DNA pol III, the major replicative enzyme in E. coli (7). Moreover, when incubated in the presence of a template-primer with a terminal mismatch with a potential for misalignment, pol IV generates DNA products that are one nucleotide shorter than expected (6). This is con-sistent with the Ϫ1 frameshift mutations seen when dinB is overexpressed (5).
Orthologs of E. coli dinB have been identified in eukaryotes (2,8). The human DINB1 gene is localized at chromosome 5q13 and encodes an 870-amino acid DNA polymerase (9 -11), which we refer to here as DNA polymerase (pol ). 1 The product of the hDINB1 gene has also been designated DNA polymerase (10), a designation used earlier (12) for the human homolog of the Drosophila melanogaster mus308 gene. pol has several properties in common with E. coli pol IV. When purified from insect cells expressing the full-length polymerase fused to glutathione S-transferase (GST) (11) or purified as a catalytically active fragment of amino acids 1-560 (9), pol lacks detectable 3Ј35Ј exonuclease activity. The purified full-length GST fusion protein has optimal activity at 37°C over the pH range 6.5-7.5, it is insensitive to inhibition by aphidicolin or dideoxynucleotides, and Mg 2ϩ is preferred over Mn 2ϩ as the essential divalent cation (11). Neither the full-length GST-enzyme purified from yeast cells (10) nor truncated pol (9) bypass UV radiationinduced photoproducts. However, as has been reported (10), the full-length enzyme misinserts nucleotides at undamaged template sites at rates between 1.5 and 30 ϫ 10Ϫ 4 . Moreover, when extending from template-primers containing terminal mismatches that have the potential to misalign, the full-length polymerase generates DNA products that are one or two nucleotides shorter than expected (10). Finally, transient expression of the homologous mouse Dinb1 gene in cultured mouse cells enhances spontaneous frameshift and base substitution mutagenesis at the Hprt locus (8).
These observations are consistent with the possibility that DNA synthesis errors by pol contribute to spontaneous mutagenesis in vivo and prompted the present study of the fidelity of human pol during copying of an undamaged DNA template. We demonstrate here that pol has relatively low fidelity and moderate processivity and discuss these properties in relation to other DNA polymerases and to observations in vivo.

EXPERIMENTAL PROCEDURES
Materials-All materials for the fidelity assay were from previously described sources (13). Human pol was expressed and purified as a full-length 870-amino acid polymerase fused to GST on the N terminus and to hexahistidine on the C terminus (11). We refer to this protein as full-length pol . pol was also purified as a C-terminal hexahistidinetagged, catalytically active fragment comprised of amino acids 1-560 (9). We refer to this protein as pol  . Neither pol preparation excised a nucleotide from a mismatched primer terminus (14). The amount of 3Ј35Ј exonuclease activity was calculated to be Յ2% of the intrinsic exonuclease activity of the Klenow fragment of E. coli DNA polymerase I.
Forward Mutation Assay-DNA products of the reactions were examined for the frequency of lacZ mutants as described previously (13). DNA from independent mutant phage was sequenced to identify the errors made during gap-filling synthesis. Error rates were calculated in three different ways. The standard approach expresses the error rate as "errors per detectable nucleotide synthesized," by considering only changes in the 275-nucleotide LacZ ␣-complementation sequence that yield detectable light blue or colorless M13 plaque phenotypes (13). However, because most of the lacZ mutants generated by pol contain multiple sequence changes that include both silent and phenotypically detectable changes, the error rate can also be described simply as the number of observed mutations divided by the total number of copied nucleotides that were sequenced. A third calculation was performed using all base substitutions found in lacZ mutants containing known detectable frameshift mutations, or all frameshifts found in lacZ mutants containing known detectable base substitution mutations. Error rates generally differed by less than 2-fold when these three calculations were compared. The error rates shown in the tables use the second, simplest method.
Processivity Analysis-Measurements were performed with M13mp2 single-stranded DNA primed at a 1.2:1 molar ratio with a 5Ј-32 P-labeled 15-mer complementary to nucleotides 106 through 120 of the LacZ gene (where ϩ1 is the first transcribed nucleotide). Reactions with HIV-1 RT and exonuclease-deficient Klenow fragment pol were previously described (15,16). pol reactions (30 l) were performed as described above but contained 5 nM template-primer and the enzyme concentrations described in Fig. 3. Reactions were incubated at 37°C. Aliquots (10 l) were removed at 5, 15, or 30 min and mixed with 10 l of 99% formamide, 5 mM EDTA, 0.1% xylene cyanole, and 0.1% bromphenol blue. DNA products were analyzed by electrophoresis in a 16% polyacrylamide gel, in parallel with products of DNA-sequencing reactions using the same template. Product bands were quantified by phosphorimaging, and the probability of terminating processive synthesis was calculated as described previously (15).

RESULTS
Average Fidelity of Human pol -The fidelity of pol was determined using a forward mutation assay that scores a variety of substitution, addition, and deletion errors during synthesis to copy a 407-nucleotide template present as a singlestranded gap in M13mp2 DNA (13). Correct polymerization to fill the gap produces DNA that yields blue M13 plaques, whereas errors are scored as light blue or colorless plaques. DNA synthesis by both pol 1-560 and full-length pol filled the 407-nucleotide gap (Fig. 1). When the DNA products were introduced into E. coli, the lacZ mutant frequency among the resulting M13 plaques was 25% for pol 1-560 and 34% for full-length pol . These lacZ mutant frequencies are 10-to 100-fold higher than those generated by eukaryotic pol ␤ (17), pol ␣, pol ␦, or pol ⑀ (18) but similar to that observed with human pol (14). The data indicate that both full-length pol and pol 1-560 have very low fidelity overall.
To determine the nature and number of polymerase errors, we isolated DNA from independent lacZ mutants and sequenced all 407 nucleotides in the gap. The lacZ mutants generated by both pol 1-560 and full-length pol contained an average of 4.2 and 3.7 mutations per mutant clone, respectively (Table I). A variety of different sequence changes was observed (Table I), and these changes were distributed throughout the target sequence (Fig. 2). The majority of sequence changes were single-base substitutions ( Fig. 2A). Given the total number of template nucleotides analyzed, the single-base substitution error rates of pol 1-560 and full-length pol are 7.4 ϫ 10 Ϫ3 and 5.8 ϫ 10 Ϫ3 , respectively (Table II). The second most frequent errors were single-base deletions, which were generated at average rates of 1.6 ϫ 10 Ϫ3 and 1.8 ϫ 10 Ϫ3 by pol 1-560 and full-length pol , respectively. When compared with the error rates of other mammalian DNA polymerases determined in this assay (Table II), the pol error rates for base substitutions and single-base deletions are intermediate between those of pol and pol ␤. The pol error rates are much higher than those of the polymerases that replicate the nuclear and mitochondrial genomes.
Error Specificity-The error rates mentioned above are average rates for all 407 template nucleotides copied. We next considered error rates for individual subsets of errors. pol 1-560 and full-length pol both generated all 12 possible base substitutions ( Fig. 2A). Base substitution error rates (Table III) were similar for the two forms of pol , and these rates varied between 0.2 ϫ 10 Ϫ3 (C⅐dCMP) and 8.2 ϫ 10 Ϫ3 (T⅐dCMP). Although mismatch-dependent variations are typical of all polymerases studied to date (recently reviewed in Ref. 19), the pol base substitution specificity is unusual in that the highest error rate is observed for the T⅐dCMP mispair (Table III). In contrast, other DNA polymerases generate the T⅐dGMP mismatch at the highest rate (14,18). From this bias and less apparent differences in the proportions of other substitutions, the ratio of misinsertion of pyrimidine dNTPs versus purine dNTPs (from Table III) is 60:40 for pol . This misinsertion bias is different from that of other polymerases, whose general preference is to misinsert purine dNTPs (14,18,20).
Analysis of the distribution of the single-base deletions within the 407-nucleotide target sequence (Fig. 2B) also revealed sequence-dependent variations in deletion error rates. The deletion rate per template nucleotide copied is highest for loss of nucleotides within homopolymeric runs, and the highest rate is observed in the longest runs (Table IV). This suggests the formation of misaligned intermediates stabilized by correct base pairing. However, the deletion rate is high even at non-   17 8 a Other changes include tandem double-base substitutions, substitution-addition and substitution-deletion errors, deletions of larger numbers of nucleotides, and complex errors.  Ϫ87, Ϫ86, and Ϫ85). Also underlined is the sequence of the palindromic Lac operator that can form a hairpin structure in the template strand. Single-base substitutions generated by pol 1-560 are shown above the template sequence, and iterated nucleotides (Table IV), such that there is only a 2-to 3-fold difference in error rate for loss of non-iterated nucleotides as compared with loss of nucleotides in homopolymeric runs of four and five bases. This difference is much smaller than that observed with several other DNA polymerases (reviewed in Ref. 19). For example, the pol ␤ deletion error rate in runs of four and five nucleotides is 35-fold higher than that for loss of non-iterated nucleotides (Table IV). As discussed below, this error specificity suggests possible mechanisms of deletion formation by pol and has implications for spontaneous frameshift mutagenesis in human cells.
Both pol 1-560 and full-length pol also frequently generate single-nucleotide additions (Table I). Unexpectedly, many of these errors involve adding a nucleotide that differs from both of its neighbors (Fig. 2B). These include 10 of 14 additions of guanine between template nucleotides 5Ј-T and C-3Ј and four of 20 additions of thymine between template nucleotides 5Ј-(C/ G/A) and C-3Ј. This addition specificity is different from that of most other DNA polymerases investigated to date, which usually add nucleotides to homopolymeric runs (reviewed in Ref. 19). This suggests that pol generates some addition errors by a mechanism other than classical strand slippage. pol also produces two base deletions (Table I), and these too are nonrandomly distributed (Fig. 2B). Seven of 13 two-base deletions generated by pol 1-560 and four of five cases by full-length pol occurred at template 5Ј-GCT-3Ј sites, where the template nucleotides C and T were deleted and the 5Ј-neighboring template base was a G. Among these, seven were at one location, nucleotides Ϫ58 and Ϫ59 (Fig. 2B), which can therefore be considered a hot spot for this deletion by pol . Finally, "other" sequence changes were also observed (Table I), including tandem double-base substitutions, substitution-addition and substitution-deletion errors, deletions of larger numbers of nucleotides, and complex errors.
Processivity Analysis-We examined the processivity of DNA synthesis by pol , i.e. the number of nucleotides polymerized per cycle of polymerase-DNA association-dissociation. Primer extension reactions were performed using a large excess of template-primer over polymerase, such that once the polymerase completes a cycle of processive synthesis, the probability that the extended product is used again is negligible. Analysis of the products of the reaction catalyzed by pol   (Fig. 3A,  lanes 1-3) shows incorporation of one to five nucleotides. Quantification of band intensities reveals that 20 primers were extended per molecule of input pol 1-560 , indicating that after terminating processive synthesis the polymerase dissociates and rebinds to a previously unused primer. The probability of termination following each incorporation event was calculated at between 65 and 80%. Low processivity and high termination probabilities were also observed with two other template-primers (data not shown).
In contrast to these results, analysis of the products of the reaction catalyzed by full-length pol revealed incorporation of one to more than 25 nucleotides per cycle of pol associationdissociation (Fig. 3A, lanes 4 -9). Thus, full-length pol is more processive than is pol  . The probability of termination of processive synthesis by the full-length enzyme varied by template position, from 46% at nucleotide 102 to 2.8% at nucleotide 81 (Fig. 3B). A relatively intense band was observed corresponding to incorporation of the 76th nucleotide (position 29, Fig. 3A). This is the beginning of the palindromic operator sequence in the LacZ gene (underlined in Fig. 2A), suggesting that full-length pol has difficulty extending through a hairpin structure in the template.
When copying this same template sequence, Klenow fragment pol and HIV-1 RT have higher processivity (15,16). For comparison, the termination probabilities of all three polymerases at 25 template positions are shown in Fig. 3B. Fulllength pol terminates processive synthesis more frequently than does Klenow fragment pol during incorporation of the first seven nucleotides, after which these two enzymes have somewhat similar termination patterns. However, the termination probability and overall termination pattern of full-length pol is distinct from that of HIV-1 RT across the region scanned (Fig. 3B). Thus, the processivity of these three polymerases is different and is variably responsive to template sequence. DISCUSSION The biochemical properties reported here are informative with respect to possible roles for pol in vivo. Human pol has moderate processivity (Fig. 3) and low fidelity (Table II-IV), suggesting a specialized role in vivo. At least two general and not mutually exclusive possibilities can be considered. pol may be recruited to replace another polymerase when that polymerase is stalled, perhaps during replication, repair, or recombination. By analogy to the role of pol in bypassing UV radiation-induced photoproducts, the relaxed discrimination of pol may allow it to copy DNA molecules containing particular lesions or undamaged templates with unusual helical dimensions or abberrant structures. A second possibility is that pol is not a "replacement" enzyme but is the normal polymerase used in some special process, such as somatic mutation of immunoglobulin genes, where relaxed selectivity offers a specific advantage. In either case, the moderate processivity of the full-length enzyme (Fig. 3) implies a DNA transaction involving synthesis of a substantial number of nucleotides.
The concept of extensive synthesis by low fidelity pol is distinct from that proposed for human pol , another polymerase in the UmuC/DinB nucleotidyl transferase superfamily. pol is encoded by the XPV gene (21-23), which is required to reduce UV radiation-induced mutations and hence suppress susceptibility to sunlight-induced skin cancer. pol has low fidelity (14,24) and low processivity (25), suggesting a model in which efficient bypass of template-distorting lesions is accomplished via relaxed geometric selectivity during incorporation of only a very few nucleotides. The intrinsically low processivity of pol may limit its opportunity to generate synthesis errors those generated by full-length pol are shown below the sequence. B, distribution of deletions and additions. Errors generated by pol 1-560 are shown above the template sequence and those generated by full-length pol are shown below the sequence. Single-base deletions are depicted by open triangles, and two-base deletions are depicted by adjacent open triangles with a slash. Single-base additions are shown with a letter to indicate the added base and a slanted line indicating where that base was added. When deletions or additions occur within repetitive sequences, the actual base that is deleted or added is not known. and perhaps also allow a separate exonuclease to proofread any misinsertions that do occur. In this way, pol promotes efficient lesion bypass, and UV radiation-induced mutations are suppressed. The situation appears to be different for pol . Because it is not only inaccurate but is also moderately processive, it could have a greater probability of contributing to   Fig. 2A) divided by the total number of A, T, G, or C template nucleotides in the 407-base target (shown in parentheses in the first column) among the 108 (pol 1-560 ) or 51 (full-length pol ) lacZ clones sequenced.

TABLE IV Sequence dependent variations in single-base deletion rates of human pol
Error rates are the number of observed single-base deletions (from Fig. 2B) divided by the total number of template nucleotides present in runs of the lengths listed (shown in parentheses in the first column) among 108 (pol 1-560 ) or 51 (full-length pol ) lacZ clones sequenced. Error rates for pol ␤ were calculated by considering only the phenotypically detectable changes, using the data in Fig. 1  mutagenesis in vivo. Indeed, earlier studies implicate the E. coli pol homolog, pol IV, in untargeted mutagenesis of phage (4) and overexpression of pol IV, strongly enhances spontaneous mutagenesis in E. coli cells transfected with plasmids (5). When mouse pol was transiently expressed in cultured mouse cells, the spontaneous mutation rate was elevated about 10-fold (8). Thus, pol might contribute to spontaneous mutagenesis in human cells by virtue of low fidelity, processive DNA synthesis. pol might also contribute to DNA damagedinduced mutagenesis in humans, through its ability to bypass guanine-AAF (N-2-acetyl-aminofluorene) adducts and abasic sites under certain in vitro conditions (9). In either case, errors generated by pol could contribute to the initiation of progression of cancer in mammals.
For spontaneous mutagenesis at the Hprt locus in cultured mouse cells, which is associated with transient overexpression of mouse pol , 30% of the 6-thioguanine-resistant mutants contain single-base frameshift mutations (8). This induction of spontaneous frameshifts is consistent with the present study showing that pol generates single-base frameshifts at very high rates (Tables II and IV). Deletion error rates are highest for loss of reiterated nucleotides (Table IV), which is consistent with the location of the frameshifts seen in vivo (8) and consistent with a model involving slippage of the template and primer strands (26). However, even non-iterated nucleotides are deleted in vitro at a rate that is extraordinarily high when compared with most other polymerases. pol also generates single-nucleotide additions (Table I), including many wherein the added nucleotide is different from both of its neighbors (Fig.  2B). In these instances, strand slippage is unlikely to initiate frameshift errors, because the non-iterated nature of the bases would not result in a misaligned intermediate stabilized by correct base pairing. Therefore, the frameshift error specificity observed here clearly suggests that pol initiates some deletion and addition errors by a mechanism other than classical strand slippage. One possibility is that frameshifts are initiated by nucleotide misinsertion (27,28). If a misinserted base is complementary to the preceding or next template base, primer relocation before continuation of synthesis could create the misaligned intermediate for addition or deletion, respectively. The base substitution error rate of pol is substantially higher than the frameshift error rate (Table II), demonstrating that the rate of misinsertion is more than sufficient to account for the observed frameshift error rates. A third model that merits consideration involves frameshifts initiated by pairing of the incoming dNTP with an adjacent template base (29 -31). It is also worth noting that pol produced errors involving more than a single nucleotide, including two-base deletions, tandem double-base substitutions, substitution-addition and substitution-deletion errors, deletions of larger numbers of nucleotides, and more complex errors. Although the number of these events is small in the collection of sequenced lacZ mutants, the rate at which pol generates these errors may be significant, because the overall lacZ mutant frequency is high.
Pol generates all 12 possible base⅐base mismatches at high rates during copying of the undamaged M13mp2 DNA template ( Fig. 2A, Table III). Human pol therefore joins pol as the second eukaryotic member of the UmuC/DinB nucleotidyl transferase superfamily to exhibit very low base substitution fidelity when copying undamaged DNA. Although it lacks an intrinsic proofreading exonuclease activity (9,11), the low fidelity of human pol is not merely due to the inability to proofread errors. On average, pol is about 10-fold less accurate than human pol ␤ and 30-to 50-fold less accurate than human pol ␣ (Table II), both of which also lack intrinsic proofreading exonuclease. The 50-fold difference in base substitu-tion error rates of pol and pol ␣ we observe (Table II) contrasts with a recent report concluding that pol and pol ␣ have about the same rate of misinsertion as determined by steadystate kinetic analysis (10). This apparent difference in the two studies may partly reflect sequence context effects on fidelity. The values in Table II are average error rates for all 12 mispairs scored at numerous template positions having variable neighboring sequences, whereas the kinetic analysis (10) measured misinsertion rates for each mismatch at a single template position.
The low fidelity of pol suggests relaxed geometric selection by the pol active site. Discrimination against the various mismatches varies over a 40-fold range (Table III). Differences in base substitution (32,33) and single-base deletion error rates (31) have been suggested to reflect differences in active site geometry, including amino acid side-chain interactions with substrate atoms in the DNA minor groove. It may eventually be possible to interpret the unusual error specificity of pol based on the crystal structure of the active site. It will also be of interest to examine the extent to which relaxed discrimination with undamaged DNA correlates with lesion bypass efficiency. Presently we know that pol is less accurate than pol when copying undamaged DNA (Table II) and that human pol efficiently bypasses cis-syn thymine-thymine dimers (21,22,24,25) and cisplatin adducts (34), whereas pol does not (9). These data are consistent with a simple working hypothesis that greatly relaxed polymerase selectivity correlates with efficient bypass of certain lesions. Clearly, more quantitative and direct comparative studies are required to determine how well this correlation holds with various lesions and DNA polymerases. Studying exceptions to this simple idea could be highly informative regarding determinants of spontaneous and damage-induced mutational specificity in humans.
The error rates for synthesis by pol 1-560 and full-length pol are remarkably similar (Tables I-IV), although we cannot exclude small site-to-site differences. This similarity suggests that the major determinants of base substitution and frameshift fidelity are encoded by the N-terminal two-thirds of the human DINB1 gene. In contrast, full-length pol is more processive than is the truncated enzyme (Fig. 3A). It remains to be determined whether the processivity of the full-length enzyme is influenced by the presence of the N-terminal GST tag or the C-terminal histidine tag. It is also possible that fidelity and/or processivity will vary depending on reaction conditions, including pH and mono-and divalent cation concentrations (11). However, because these experiments with pol 1-560 and full-length pol were performed using the same conditions, the higher processivity of full-length pol suggests that the 310 C-terminal amino acids may contribute to processive synthesis.
The C terminus of mammalian pol contains a duplicated C2HC Zn-cluster domain (2) similar to a C2HC Zn-cluster domain in the Rad18 protein. Because the Rad18-Rad6 heterodimer binds to DNA (35), the duplicated C2HC Zn-cluster domain in human pol may also bind DNA and thereby contribute to the processivity of polymerization. This possibility is particularly interesting, because the Rad18-Rad6 heterodimer has a higher affinity for single-stranded DNA than it does for double-stranded DNA (35). If a similar situation holds true for pol , it suggests that processivity might be enhanced via single-stranded DNA binding. This would be distinct from three other types of protein-DNA interactions known to enhance polymerase processivity. These include polymerase interactions with sliding clamps that topologically encircle the DNA (36), electrostatic and hydrophobic interactions between polymerases and the template-primer (37-40), and interactions of the N-terminal domain of pol ␤ with the 5Ј-phospho-rylated end of short gaps that promotes processive gap filling (41). However, higher processivity need not necessarily depend on enhanced DNA binding. For example, the C-terminal residues of pol could enhance processivity by increasing the polymerization rate, or the amino acids of the Zn-cluster might indirectly effect processivity and/or DNA binding by determining protein structure rather than by directly contacting the DNA.
The moderate processivity of full-length pol does not preclude further enhancement of its processivity or other properties via interactions with polymerase accessory proteins. PCNA did not stimulate copying of an oligonucleotide template by full-length pol (11). However, the properties of the pol homolog E. coli DNA polymerase IV are strongly influenced by the presence of the ␤⅐␥ processivity complex (7). Full-length pol is clearly sensitive to the sequence of the template-primer, because the probability of termination of processive synthesis varies by template position over more than a 15-fold range (Fig.  3B). The processivity of full-length pol is higher than that of pol 1-560 but lower than that of Klenow fragment pol and HIV-1 RT (Fig. 3B). Structure-function studies indicate that processive synthesis by HIV-1 RT depends heavily on interactions between hydrophobic side chains and the DNA minor groove 2-6 base pairs upstream of the active site (42,43). In contrast, pol A family enzymes such as Klenow fragment polymerase contact DNA via electrostatic interactions with the sugar-phosphate backbone and hydrogen bonds to atoms in the DNA minor groove 4 -5 base pairs upstream of the active site (37)(38)(39)(40)44). The fact that pol , Klenow fragment pol and HIV-1 RT generate different termination patterns when processively copying the same template (Fig. 3B) suggests that pol interactions with the DNA may differ from those observed for polymerases in the other two families. This is also implied by the fact that the fidelity of pol is much lower than that of Klenow fragment pol and HIV-1 RT. In fact, the very low fidelity, unusual error specificity, and moderate processivity of pol is unique among any eukaryotic or prokaryotic DNA polymerase described to date.