Low-fidelity Pyrococcus furiosus DNA polymerase mutants useful in error-prone PCR

Abstract

Random mutagenesis constitutes an important approach for identifying critical regions of proteins, studying structure-function relations and developing novel proteins with desired properties. Perhaps, the most popular method is the error-prone PCR, in which mistakes are introduced into a gene, and hence a protein, during DNA polymerase-catalysed amplification cycles. Unfortunately, the relatively high fidelities of the thermostable DNA polymerases commonly used for PCR result in too few mistakes in the amplified DNA for efficient mutagenesis. In this paper, we describe mutants of the family B DNA polymerase from Pyrococcus furiosus (Pfu-Pol), with superb performance in error-prone PCR. The key amino acid changes occur in a short loop linking two long α-helices that comprise the ‘fingers’ sub-domain of the protein. This region is responsible for binding the incoming dNTPs and ensuring that only correct bases are inserted opposite the complementary base in the template strand. Mutations in the short loop, when combined with an additional mutation that abolishes the 3′–5′ proof-reading exonuclease activity, convert the extremely accurate wild-type polymerase into a variant with low fidelity. The mutant Pfu-Pols can be applied in error-prone PCR, under exactly the same conditions used for standard, high-fidelity PCR with the wild-type enzyme. Large quantities of amplified product, with a high frequency of nearly indiscriminate mutations, are produced. It is anticipated that the Pfu-Pol variants will be extremely useful for the randomization of gene, and hence protein, sequences.

INTRODUCTION

DNA polymerases possess high fidelity, defined as the ability to add a dNTP to the extending strand, while faithfully maintaining Watson–Crick base pairing with the coding template. Although polymerase accuracy is important during PCR, enabling DNA to be accurately copied, it is a serious drawback in error-prone PCR, where the aim is to deliberately introduce mistakes into the amplified products (1–6). Two classes of thermostable DNA polymerase are commonly applied in PCR; bacterial enzymes, e.g. from Thermus aquaticus (Taq-Pol) and archaeal proteins, exemplified by Pyrococcus furiosus (Pfu-Pol). Taq-Pol lacks a 3′–5′ proof-reading exonuclease and so is intrinsically more error prone than Pfu-Pol, which possesses this activity (7,8). Nevertheless, wild-type Taq-Pol is still too accurate for use in error-prone PCR (3) and non-standard conditions, including the use of Mn²⁺ (rather than the normal divalent metal ion, Mg²⁺) and unbalanced dNTP levels (4,6) or unnatural mutagenic bases (5), are needed for its application. In addition to protocol variations, the use of Taq-Pol in error-prone PCR is often associated with poor yields, low levels of mutation and, critically, a biased mutation spectrum. Random mutagenesis aims at producing indiscriminate changes to a gene, and consequently a complete set of mutations in a protein, where there is a high probability of changing each amino acid into all possible alternatives. Biased protocols only give a subset of altered genes, and hence not all protein variants. The archaeal DNA polymerases are usually considered the gold standard for faithful PCR amplifications (7,8), although their high fidelity may be reduced by disabling the proof-reading exonuclease. As far as we are aware, these enzymes, due undoubtedly to their extreme accuracy, have never found application in the error-prone PCR.

Most DNA polymerases are multi-functional proteins; a pol domain (comprising three sub-domains termed the palm, fingers and thumb) being responsible for DNA replication, with the fingers playing an important role in dNTP recognition (9–11). The archaeal family B polymerases fit this pattern, and crystal structures of four proteins (12–15) demonstrated that their fingers sub-domain consists of two long anti-parallel α-helices, the N and O-helices, separated by a short loop (shown in Figure 1A for the enzyme from Thermococcus species 9°N-7). Folding of the helix–loop–helix motif results in the proximal location of a number of highly conserved amino acids derived from both α-helices. Comparison of the archaeal structures with a homologous viral polymerase, for which an enzyme-primer/template-dNTP structure has been solved (16), implicates the conserved amino acids in dNTP binding and polymerase fidelity. The fingers sub-domain is well conserved between archaeal family-B DNA polymerases (Figure 1B), suggesting that all enzymes (including that from P.furiosus investigated in this paper) have a similar helix–loop–helix motif and, therefore, use the same mechanisms for dNTP recognition and accurate DNA replication. The viral polymerase structure clearly shows a large movement in the fingers sub-domain on dNTP binding (16), and kinetic studies have indicated that DNA polymerase fidelity is strongly dependent on protein conformational changes (17–19). A minimum scheme involves the transient association of any dNTP with enzyme-bound primer-template; only when the incoming dNTP forms a Watson–Crick base pair with the template is the conformational change triggered, assembling the active site and allowing polymerization. In this manner, incorrect dNTPs are rejected and fidelity maintained.

Figure 1.

(A) Structure of the fingers sub-domain of polymerase 9°N-7. The sub-domain is made up of two long α-helices (the N and O-helices), separated by a short loop (amino acids in the loop shown in green). Folding of the sub-domain brings a number of highly conserved amino acids, shown in red, into close proximity on the same side of the motif. The red and green amino acids in this panel correspond to the identically coloured amino acid for the second entry (Therm. 9°N-7 sequence) in (B). (B) Amino acid alignment of the fingers sub-domain of a number of archaeal family-B DNA polymerases (accession numbers given in brackets). The highly conserved amino acids are shown in red, as in (A). Amino acids that form the loop between the two helices are shown in green. For clarity, only sequences that can be lined up without the introduction of gaps (many polymerases have a few extra amino acids in the loop region, necessitating gaps) have been included. Abbreviations: Pyr, Pyrococcus; Therm, Thermococcus; Methsar, Methanosarcina; Methcoc, Methanococcoides; Ferro, Ferroplasma; Meththerm thermauto, Methanothermobacter thermautotrophicus; Sulfur, Sulfurisphaera; and Desulfur, Desulforococcus.

Extrapolating structural and kinetic data, we reasoned that a conformational change involving the fingers sub-domain may play a role in the selection and incorporation of the correct dNTP, thereby contributing to polymerase accuracy. The short loop in the fingers sub-domain (Figure 1A and B) influences the relative orientation of the two long α-helices, and mutagenesis of these three amino acids (Figure 1B, T471, Q472 and D473 in Pfu-Pol) appeared a promising approach for the manipulation of any conformational changes controlling the relative disposition of the two helices. All three amino acids have been individually converted to both glycine and alanine, small amino acids that may endow the loop with increased flexibility; the central glutamine has also been changed to proline, an attempt at rigidifying the region. Pfu-Pol has a 3′–5′ proof-reading exonuclease activity, which removes incorrectly incorporated bases. In order to directly study the influence of the loop mutations on fidelity, and in the case of low-fidelity variants to prevent excision of aberrant bases, the exonuclease activity was disabled with the mutation D215A (20).

MATERIALS AND METHODS

Purification/mutagenesis/activity assay of Pfu-Pol

Wild-type Pfu-Pol (both exo⁺ and exo⁻) was purified from an Escherichia coli overexpressing strain containing the plasmid pET-17b[Pfu-Pol], which has the polymerase under control of the phage T7 promoter (20). Loop mutations of Pfu-Pol were prepared using ‘overlap extension PCR’ (21) on the wild-type exo⁻ (D215A) gene, excised from the plasmid pET-17b[Pfu-Pol]. Altered genes were inserted into pET-17b and the protein overexpressed in E.coli BL21 (DE3) pLysS. Purification involved a heat step and chromatography on DEAE–Sephacel and Heparin–Sepharose (20). All protein samples were >95% pure by Coomassie blue stained SDS–PAGE. Concentrations were determined by absorbance at 280 nm with an extinction coefficient of 1.2 × 10⁵ M⁻¹ cm⁻¹ (http://ca.expasy.org/cgi-bin/protparam). Determination of the specific activity of Pfu-Pol used an assay based on the incorporation of dNTPs into activated calf thymus DNA (22). Approximately 0.5–1.5 μg Pfu-Pol, in a volume of 15 μl, was added to 135 μl of pre-warmed reaction buffer [20 mM Tris–HCl (pH 8.0), 10 mM KCl, 2 mM MgSO₄, 10 mM (NH₄)₂SO₄, 0.1% Triton X-100, 100 μg/ml BSA, 66.67 μM each dNTP, 200 μg/ml activated calf thymus DNA and 1 μCi [α-³²P] dATP (3000 Ci/mmol) and incubated at 72°C. The samples were removed every minute, up to a time of 10 min, and the reaction terminated by the addition of EDTA (30 mM final concentration). The samples were diluted with 1 ml of carrier solution (2 mM EDTA, 50 μg/ml activated calf thymus DNA) and the DNA precipitated by the addition of 1 ml of 20% (w/v) trichloroacetic acid and 2% (w/v) sodium pyrophosphate, followed by incubation on ice for >15 min. DNA was colleted on GF/B glass fibre filters (Whatman) and washed extensively with ice-cold 1 M HCl containing 100 mM sodium pyrophosphate. The radioactivity incorporated into DNA was determined by scintillation counting and specific activity determined from plots of counts versus time. One unit of enzyme is defined as the amount of enzyme that incorporates 10 nmol of dATP into acid-precipitable material in 30 min at 72°C.

In vivo lacIOZα fidelity assay

The 1.9 kb lacIOZα sequence was amplified from pRIAZ (kindly supplied by Dr Holly Hogrefe; Stratagene) by PCR with the polymerases under investigation using the following conditions: 2.5 U polymerase, 2.5 ng pRIAZ (corresponding to 1.2 ng lacI target) (8), 100 ng each primer (forward primer: 5′-CATAGCGAATTCGCAAAACCTTTCGCGGTATGG-3′, reverse primer: 5′-ACTACGGAATTCCACGGAAAATGCCGCTCATCC-3′), 200 μM each dNTP, 20 mM Tris–HCl (pH 8.0), 10 mM KCl, 10 mM (NH₄)₂SO₄, 2 mM MgSO₄, 0.1% Triton X-100® and 0.1 mg/ml BSA. The amount of PCR product was determined using ethidium bromide-stained 1% agarose gels by comparison with known amounts of double-stranded DNA of the same length. Products were cleaved at the unique flanking EcoRI sites introduced during PCR by the primers and ligated into λgt10 arms (Stratagene). Ligated products were packaged into λ phage using Gigapack® III Gold packaging extract (Stratagene) and used to infect the α-complementing E.coli strain SCS-8 by incubation at room temperature for 20 min. To determine the phage titre (in plaque-forming units) and the percentage of phage with viable lacZα inserts, infected cells were serially diluted with SM buffer [50 mM Tris–HCl (pH 7.5), 10 mM NaCl, 8.1 mM MgSO₄.7H₂O, 0.01% gelatine], mixed with NZY top agar (Stratagene) supplemented with X-Gal (1 mg/ml) and isopropyl-β-d-thiogalactopyranoside (IPTG) (1.5 mM) and plated onto NZY bottom agar. After incubation at 37°C for 16 h, the total number of plaques and percentage of blue plaques (corresponding to phage with a viable lacZα) were scored. The mutation load of the lacI gene was determined by plating 8000–10000 plaque-forming units in Big Blue® top agar (Stratagene) containing X-Gal (1.5 mg/ml) onto 25 cm dishes containing Big Blue® bottom agar (Stratagene). Following incubation at 37°C for 16 h, the total number of plaques and the number of blue plaques were scored.

In vitro dNTP incorporation fidelity assay

Polymerase-catalysed incorporation of correct and incorrect dNTPs was measured under the following conditions: 20 mM Tris–HCl, pH 8, containing 10 mM KCl, 10 mM (NH₄)₂SO₄, 0.1% Triton X-100®, 12.5 mM MgCl₂, 10 nM primer-template duplex (produced by the hybridization of a 17mer template, 5′-GAAGCTCCGCGGGCGCC-3′ with an 11mer primer, 5′-GGCGCCCGCGG-3′; the primer contained a ³²P-5′-phosphate label), 1.5 μM polymerase and either 100 μM dATP or 100 μM dGTP. Reactions were initiated by the addition of MgCl₂ and run at 50°C. Samples were taken at 7, 14, 25, 35, 60 and 300 s and incorporation terminated by the addition of EDTA to a final concentration of 25 mM. The substrate and product oligodeoxynucleotides were analysed by separation on 20% denaturing polyacrylamide gels, followed by phosphorimaging.

Determination of the mutation spectrum of Pfu-Pol(exo⁻)(D473G)

The D473G variant was used to amplify the entire length pET17b[Pfu-Pol] (20), essentially using a QuickChange® PCR (Stratagene), under the following conditions: 20 mM Tris–HCl, (pH 8.0), containing 20 mM MgSO₄, 10 mM KCl, 10 mM (NH₄)₂SO₄, 0.1% Triton X-100®, 0.1 mg/ml BSA, 55 ng of pET17b[Pfu-Pol], 125 ng of each primer and 500 μM of the four dNTPs. Primer sequences were 5′-GTAAAACGACGGCCAGT-3′ and 5′-GAACTATGATATCGCTCC-3′, which hybridize around the codon corresponding amino acid isoleucine114 of the Pfu-Pol gene. The primers introduce a single base pair change into amplified products, enabling differentiation from the starting plasmid. The amount of D473G used was 2.5 U, and the PCR protocol comprised 1 cycle of 95°C for 30 s to denature double-stranded DNA, followed by 25 cycles of 95°C for 30 s to separate plasmid strands, 55°C for 30 s to anneal the primers and 68°C for 15 min to extend the primers. The amount of PCR product was determined as above. The resulting mixture was treated as described in the Stratagene QuickChange® PCR kit, i.e. incubated with ∼10 U of DpnI at 37°C for 2 h to degrade starting plasmid and then used to transform E.coli XL-10 Gold (Stratagene) to ampicillin resistance. Plasmids were isolated from 19 of the resultant colonies by miniprep (Qiagen) and the bases corresponding to the first 500 nt of the Pfu-Pol gene were sequenced (Lark Technologies Inc.).

RESULTS

Accuracy of polymerase variants assessed using an in vivo assay

The accuracy of DNA polymerases can be assessed using a lacIOZα fidelity assay (7,8), in which a 1.9 kb DNA fragment encoding the lacIOZα sequence is amplified. All the polymerase variants were able to carry out the PCR with comparable efficiency to the wild type (Figure 2A). Furthermore, the specific activities of all the mutants were, to within a factor of two, identical to the wild type (data not shown). Determination of the amount of PCR product (typically 4–7.5 μg) allowed evaluation of the number of template doublings (Table 1), a requirement for the determination of the polymerase error rate. Polymerases that make no mistakes copying the lacI gene result in an active lac repressor, which prevents the transcription of the lacZα gene. Consequently, appropriate E.coli strains infected with the amplified product cannot express an active β-galactosidase and plaques are unable to hydrolyze X-Gal, appearing white. Polymerases of less than perfect fidelity introduce errors into lacI, which may give rise to a repressor unable to bind DNA; in these cases transcription of lacZα results in the production of β-galactosidase and blue plaques. The number of ‘detectable sites’ (i.e. mutations giving rise to inactive repressor) in the lacI gene is known (7,8,23), enabling ratios of blue and white plaques to be converted to mutation load (number of mistakes in the amplified product) and, hence, to the error rate of the polymerase (Table 1). Wild-type Pfu-Pol(exo⁺), the enzyme routinely applied in high-fidelity PCR, makes about one mistake for every 10⁶ bases incorporated, Taq-Pol and Pfu-Pol(exo⁻) are about 10- and 60-fold less accurate, respectively; similar values have been reported previously (8). The T471G and T471A variants were slightly less accurate, by factors of 1.2 and 1.6, respectively, than the Pfu-Pol(exo⁻) from which they are derived (Table 1). Along with lacI, the lacIOZαassay results in the amplification of the lacZα gene, coding for the α-fragment of β-galactosidase; polymerases may also alter this gene, resulting, ultimately, in an inactive β-galactosidase. Providing the error rate of the polymerase is not too high, a correction can be made for mutations to lacZα (Table 1, see Materials and Methods), by noting the phenotype of plaques in the presence of IPTG, which abolishes lac repressor binding to DNA. Here, β-galactosidase should be produced, independent of the nature of the lac repressor, and the majority of plaques should appear blue, as was found for Taq-Pol, wild-type Pfu-Pol and T471G/A. However, with the Q472 and D473 variants, more than 97.5% of plaques appeared white in the presence of IPTG, indicating that very few possessed an active β-galactosidase. Such excessive mutation to lacZα makes scoring based on mutations to lacI impossible; however, these results clearly indicate that the Q472 and D473 have a lower fidelity than the parent exo⁻ polymerase.

Figure 2.

(A) PCR amplification of a 1.9 kb lacIOZα sequence by Pfu-Pol. The lanes containing four bands, separating the different Pfu-Pol mutants, are length markers of 3, 2, 1.5 and 1 kb. (B) Pfu-Pol catalysed extension of the primer strand of a primer-template in the presence of a single dNTP, either dATP (left of dotted line) or dGTP (right of dotted line). The primer-template possesses the structure: 5′-[³²P]pGGCGCCCGCGG/3′-CCGCGGGCGCCTCGAAG. The primer runs at the position indicated ‘N’; addition of a first and second dNTP gives the extended products running at ‘N+1’ and ‘N+2’, respectively. The time points are 0, 7, 14, 35, 60 and 300 s.

Table 1. Error rates of polymerases during PCR.

Polymerase	Template doublings (d)a	lacI− plaquesb (% ± SD)	Mutation loadc (per kilobase) (±SD)	Error rated (per base) (×10−6 ± SD)
Pfu-Pol (exo+)	12.3	0.61 ± 0.09	0.017 ± 0.002	1.4 ± 0.2
Pfu-Pol (exo−)	11.8	20 ± 1.7	0.58 ± 0.05	49 ± 4
Taq-Pol	11.6	3.9 ± 0.16	0.12 ± 0.006	10 ± 0.5
Pfu-Pol (exo−) T471G	11.8	27 ± 2.8	0.78 ± 0.08	66 ± 7
Pfu-Pol (exo−) T471A	12.6	40 ± 3.3	1.1 ± 0.1	91 ± 8
Pfu-Pol (exo−) Q472G	12.1	Undeterminable
Pfu-Pol (exo−) Q472A	11.2	Undeterminable
Pfu-Pol (exo−) Q472P	11.6	Undeterminable
Pfu-Pol (exo−) D473G	12.2	Undeterminable
Pfu-Pol (exo−) D473A	12.6	Undeterminable
Pfu-Pol (exo−) D473G*	10.3	—	7.2	700

^aTemplate doublings (d) (for all polymerases except D43G^*) were calculated using 2^d = (amount of PCR product)/(amount of starting target) (8). All values are for amplification of the 1.9 kb lacIOZα sequence from pRIAZ (lacIOZα fidelity assay).

^bThe % of lacI⁻ plaques (for all polymerases above the dotted line) were evaluated using the lacIOZα fidelity assay. The values are calculated as number of blue plaques (lacI mutants)/‘corrected’ total number of plaques (both counted on plates lacking IPTG). The total number of plaques is ‘corrected’ by multiplying by the fraction of plaques containing an active lacZα (number of blue plaques/total number of plaques on plates containing IPTG) (8). The average of three experiments is given. This method could not be applied to mutations to Q472 and D473 as >97.5% of plaques on the IPTG plates were white (shown as undeterminable).

^cMutation load corresponds to the actual number of mutations seen in the amplified DNA. For the first five entries, the mutation load was evaluated by dividing the fraction of lacI⁻ plaques by the number of detectable sites in lacI (= 349) (8,23). The mutation load depends on template doublings and cannot be rigorously compared due to slight variations in the value of d for each polymerase.

^dError rate is the frequency at which a polymerase incorporates an incorrect base during the replication of DNA. Error rate = mutation load/template doublings (8). The error rate is a fundamental property of the polymerase and is directly comparable irrespective of the method used for determination.

Owing to the inapplicability of the lacIOZα fidelity assay for mutations to Q472 and D473, a different assay was applied with D473G (results shown in row D473G^*). Here, the template doublings were obtained from the amplification of pET17b[Pfu-Pol] (direct sequencing assay); the mutation load was determined by directly sequencing the DNA produced by PCR and is the number of incorrect bases/total number of bases sequenced (68/9500); Error rate = mutation load/template doublings (8) was calculated using the values found in the direct sequencing assay.

Accuracy of polymerase variants assessed using an in vitro assay

To further investigate the fidelity of the polymerases, an assay based on dNTP incorporation into primer-templates has been used. An excess of polymerase was incubated with a short synthetic primer-template and only one of the four dNTPs, either dATP or dGTP. The first unpaired base in the template is T, and an ‘accurate’ polymerase would be expected to incorporate dATP, but not dGTP. As shown in Figure 2B, this is the behaviour observed for Pfu-Pol(exo⁻), T471G and T471A; a single dATP is rapidly incorporated to give the N+1 product, dGTP incorporation is minimal and only seen after extended incubation times. In contrast, the mutations to Q472 and D473 not only correctly incorporate dATP to give the N+1 product, but rapidly mis-incorporate a second dATP, opposite template dC, forming the N+2 derivative. Analogously, dGTP is rapidly mis-incorporated opposite T (N+1 product) and this incorrectly base paired product can be further extended, by correct incorporation of dG opposite dC, yielding N+2 (Figure 2B). These experiments are qualitative and designed to give an initial impression of polymerase fidelity rather than measure it accurately (these experiments are currently being carried out). However, it is very apparent that mutations to Q472 and D473 result in less accurate polymerases than Pfu-Pol(exo⁻) and the T471 variants (in agreement with the lacIOZα assay). Moreover, following incorporation of an incorrect base into the newly synthesized strand, the Q472/D473 variants are able to further extend the mismatched base pair at the primer-template junction, allowing DNA synthesis to continue.

Mutation spectrum of Pfu-Pol(exo⁻)(D473G)

As a representative of a low-fidelity mutant, D473G was selected for further investigation. This mutant was used in a PCR reaction to amplify the entire plasmid pET17b[Pfu-Pol] (this comprises the standard pET17b plasmid with a gene coding for wild-type Pfu-Pol inserted at the multiple cloning site) (20). An aliquot of 55 ng of pET17b[Pfu-Pol] was subject to 25 rounds of PCR giving 68 μg of amplified product, which was used to transform E.coli to ampicillin resistance. Following plating, 19 plasmids were isolated from single colonies and sequenced over an identical stretch of 500 nt, corresponding to the bases that code for the first 167 amino acids in the Pfu-Pol gene. A total of 68 mistakes in the 9500 bases sequenced were observed, giving a mutation load of 7.2 mistakes per kilobase in the amplified product (a value eminently suitable for error-prone PCR) and an error rate for D473G of 700 × 10⁻⁶ (Table 1). Although D473G is evidently less accurate than the wild type, the 14-fold difference in fidelity, implied by the data in Table 1, calls for a degree of caution. Separate methods were used for the error-rate evaluation; direct sequencing for D473G, the lacIOZα fidelity assay for the wild type and confirmation of the factor of 14 will require investigation of both polymerases using the same assay. Thus, a slightly better fidelity for wild-type Pfu-Pol was observed with an assay based on constant denaturant capillary electrophoresis (24) as compared with lacIOZα (8). The raw data, obtained from sequencing the PCR products, are shown in Figure 3, which illustrates the alterations found in the sense strand of the Pfu-Pol gene and, hence, give an indication of the changes that can be expected in the protein. Although deletions and insertions are observed, they are few in number relative to substitutions and, furthermore, the spectrum of the substitution mutations is predominantly random (for a completely random mutagenic polymerase, the 12 possible combinations should occur with an equal frequency of 8.3%). Additionally, no hot-spots were observed and the mutations were evenly distributed along the stretch of sequenced DNA.

Figure 3.

Alterations in DNA found after error-prone PCR using Pfu-Pol(exo⁻)D473G. The nomenclature A-G indicates a change in the sequenced strand from adenine to guanine. The actual numbers observed for each alteration are given in brackets; the arrow indicates the frequency expected (8.3%) for completely random changes (excluding insertions and deletions). As explained in the text, certain substitutions, e.g. T-G and A-C, are equivalent and their frequencies need to be pooled for analysis of bias (Table 2).

Some of the changes produced during error-prone PCR, and shown in Figure 3, are equivalent. Thus, mutation of an A:T base pair, during DNA replication, by incorporating either G, rather than T, opposite A (T→G) or C, rather than A, opposite T (A→C) are indistinguishable; the initially formed A:G or C:T mismatches are both converted to an identical C:G transition mutation in the next PCR cycle. Therefore, in order to assess the bias of a mutagenic polymerase, it is necessary to combine equivalent mutations, as shown in Table 2 for both the D473G variant of Pfu-Pol and a number of published investigations using Taq-Pol. An ideal error-prone polymerase, with no bias, should produce each of the six paired mutations at a frequency of 16.7% and, preferably, no insertions and deletions. While the mutation spectrum of D473G is not completely random, all the expected changes are observed at reasonable frequency, and deviations from the results expected for a bias-free enzyme are minimal. The results shown in Table 2 for Taq-Pol with Mn²⁺/unbalanced dNTPs (4,6) (the differences in these two spectra probably arise from variation in dNTP balance) and mutagenic bases (5) clearly indicate that these systems show greater bias than observed with the Pfu-Pol mutant, which, we believe, is probably the most random mutagenic polymerase yet reported.

Table 2. Mutation spectra observed using various DNA polymerases/different conditions in error-prone PCR.

Mutation	Pfu-Pol(exo−) D473Ga	Taq-Pol (Mn2+/unbalanced dNTPs)b	Taq-Pol (Mn2+/unbalanced dNTPs)c	Taq-Pol (unnatural mutagenic bases)d
A→T/T→A	28	40.9	11.4	0.2
A→C/T→G	7.4	7.3	3.3	8.4
A→G/T→C	19.2	27.6	60.9	78.3
G→A/C→T	22	13.6	18.1	13.2
G→C/C→G	7.3	1.4	4.3	0.7
G→T/C→A	10.3	4.5	1.8	0.0
Insertion	2.9	0.3	Not given	∼0
Deletion	2.9	4.2	Not given	∼0

^aThis study; data compiled from Figure 3.

^bTaq-Pol in the presence of Mn²⁺ and [dATP] = [dGTP] = 0.2 mM, [dCTP] = [TTP] = 1 mM (6).

^cTaq-Pol in the presence of Mn²⁺ and [dCTP] = [dATP] = 0.03 mM, [dGTP] = [TTP] = 1 mM (4).

^dTaq-Pol in the presence of Mg²⁺, the four standard dNTPs (0.5 mM) and the two unnatural mutagenic bases 8-oxodGTP and dPTP (5).

All figures are percentages.

DISCUSSION

Rationalization of the low fidelity shown by loop mutants

Structures of family-B DNA polymerases (12–15) demonstrate that all three amino acids in the loop of the fingers domain (Figure 1A and B) are too far from the dNTP binding region to have a direct influence on incorporation/rejection of the incoming base and hence fidelity; any effects must, therefore, arise by indirect perturbations to the active site. Polymerase variants with the most compromised fidelity are produced by changing the highly conserved (Figure 1B) aspartic acid at position 473 to glycine or alanine. As shown for 9°N-7-Pol (14) (Figure 4), this amino acid uses both the carboxyl side chain and the amide carbonyl-oxygen to produce an extensive hydrogen bonding network to the amide NH of three contiguous amino acids (LEK) at the beginning of the O-α-helix. Near identical hydrogen bonding schemes are observed for the three other structurally characterized archaeal polymerases from Thermococcus gornonarius (Tgo-Pol) (12), Desulfurococcus Tok (D.Tok-Pol) (13) and Pyrococcus kodakaraensis (KOD-Pol) (15); in each case, the conserved aspartic acid makes hydrogen bonds to the amide NH of the same three amino acids in the O-helix. Changes to the aspartic acid are likely to remove (in the case of hydrogen bonds made by the side chain) or weaken (for hydrogen bonds made via the backbone carbonyl oxygen) these interactions, leading to a more flexible fingers sub-domain with increased conformational freedom. Following the binding of a dNTP, to produce the initial enzyme/primer-template/dNTP ternary complex, the conformational change that creates the active site may be facilitated for incorrectly base-paired dNTPs, leading to the lack of fidelity observed. The middle amino acid of the loop, Q472 in the case of Pfu-Pol, often comprises a large hydrophilic or hydrophobic R-group (Figure 1B) and with 9°N-7-Pol the corresponding amino acid is valine. The side chain points towards the solvent, making no contacts to the rest of the enzyme; additionally, the amide backbone is not involved in hydrogen bonding (Figure 4). Identical patterns are observed with Tgo-Pol, D.Tok-Pol and KOD-Pol (all three have a hydrophobic amino acid at the central position of the loop) and it is likely that hydrophilic amino acids at this location, e.g. Q472 for Pfu-Pol, similarly make no direct interactions with the protein. Although the central loop amino acid appears unimportant for the overall structure of the fingers sub-domain, mutations strongly decrease fidelity; perhaps by causing perturbations to the orientation of the critical neighbouring amino acid, D473. The amino acid at position 471 is often threonine (as in Pfu-Pol, 9°N-7-Pol and the three other structurally investigated polymerases) or serine and, in the case of 9°N-7-Pol, a single hydrogen bond is seen between the threonine side chain and the carboxylate of a glutamic acid in the O-helix (Figure 4). A similar paucity of interactions is observed with Tgo-Pol, D.Tok-Pol and KOD-Pol, where the threonine makes a minimal number of hydrogen bonds and there is no consistent interaction pattern as observed with the key aspartic acid. Thus, T471 appears to be relatively unimportant in structural terms, mutations to this amino acid are minimally perturbing and so have little influence on polymerase fidelity.

Figure 4.

Close up of the loop region of polymerase 9°N-7, comprising the three loop amino acids TVD (green) and the four flanking amino acids from both the N α-helix (cyan) and the O α-helix (magenta) (see Figure 1B, second entry for the identification of these amino acids). The loop aspartic acid (side chain and backbone carbonyl) makes a network of hydrogen bonds (black dashed lines) to the backbone NH group of three amino acids (LEK) in the O-helix. The middle valine of the loop does not make any interactions with the rest of the protein and the side chain points towards the solvent. The loop threonine makes a single hydrogen bond to a glutamic acid in the O-helix.

Potential uses of loop mutants

Mutations in the loop region of the fingers sub-domain of Pfu-Pol result in proteins with the same specific activity, robustness and thermostability as the wild-type. However, as exemplified by Pfu-Pol(exo⁻)D473G (D473A and the mutations to Q472 are likely to have similar properties), certain changes to the loop amino acids result in variants able to mutagenise DNA sequences. Alterations are of a random nature, with little bias in the mutation spectrum, and produced at a high frequency. Critically ‘normal’ amplification conditions, i.e. exactly as for a typical high-fidelity PCR, can be used and unusual metal ions, abnormal dNTP concentrations or mutagenic base analogues are not required. These favourable properties should allow facile exploitation of the loop mutants for random mutagenesis by error-prone PCR.