Most practitioners of PCR prefer to carry out DNA amplification in an accurate manner, introducing as few base substitutions as possible. This is especially critical when one is studying clonal isolates and must distinguish natural variation from artifactual variation that is introduced by polymerase error. Fortunately, thermostable DNA polymerases are available that operate with high fidelity because of an intrinsic 3 ' --> 5 ' exonuclease activity (for review see ref. 1). Manipulation of PCR conditions can lead to further improvement of copying accuracy.

Here, we consider the other side of the fidelity issue - those instances where promiscuity is a virtue. Oftentimes, in probing the structure or function of a protein or nucleic acid, one wishes to generate a library of mutants and apply a screening method to isolate individuals that exhibit a particular property. For mutations over a short stretch of nucleotides within a cloned gene, it is appropriate to replace a portion of the gene with a synthetic DNA fragment that contains random or partially randomized nucleotides.^(2-5) For mutations over a longer segment, up to the size of an entire gene, it may be preferable to scatter random mutations over the entire sequence, typically at a frequency of one or a few mutations per molecule. In such cases, it is most convenient to introduce random mutations through inaccurate copying by a DNA polymerase, especially if the polymerase is a thermostable enzyme that can operate in the context of the PCR. Each pass of the ploymerase during the PCR allows for the possibility of mutations, so that the cumulative error rate can become substantial.

The error rate of Taq polymerase is the highest of the known thermostable DNA polymerases, in the rage of 0.1 x 10^-4 to 2 x 10^-4 per nucleotide per pass of the polymerase, depending on reaction conditions.^(6-9) Over the course of the PCR, in which the polymerase makes an average of 20-25 passes, the cumulative error rate is ~10^-3 per nucleotide. In most cases this is insufficient to generate a diverse library of variant sequences, especially over a region shorter than 1000 nucleotides. A further drawback is that the errors made by Taq polymerase under standard PCR conditions are heavily biased toward A.T --> G.C changes.⁽⁶⁾ We have devised a mutagenic PCR that has an overall error rate of ~7x10^-3 per nucleotide and does not exhibit substantial sequence bias.⁽¹⁰⁾