A high-throughput assay for quantitative measurement of PCR errors

Abstract

The accuracy with which DNA polymerase can replicate a template DNA sequence is an extremely important property that can vary by an order of magnitude from one enzyme to another. The rate of nucleotide misincorporation is shaped by multiple factors, including PCR conditions and proofreading capabilities, and proper assessment of polymerase error rate is essential for a wide range of sensitive PCR-based assays. In this paper, we describe a method for studying polymerase errors with exceptional resolution, which combines unique molecular identifier tagging and high-throughput sequencing. Our protocol is less laborious than commonly-used methods, and is also scalable, robust and accurate. In a series of nine PCR assays, we have measured a range of polymerase accuracies that is in line with previous observations. However, we were also able to comprehensively describe individual errors introduced by each polymerase after either 20 PCR cycles or a linear amplification, revealing specific substitution preferences and the diversity of PCR error frequency profiles. We also demonstrate that the detected high-frequency PCR errors are highly recurrent and that the position in the template sequence and polymerase-specific substitution preferences are among the major factors influencing the observed PCR error rate.

smCounter2: an accurate low-frequency variant caller for targeted sequencing data with unique molecular identifiers

“…Background errors are sensitive to enrichment chemistry and DNA polymerase. The error pattern we observed in QIAseq DNA panels agrees with that in other PCR enrichment studies [25, 26] but differs from hybridization capture studies [6, 22] where A>C and G > T errors are dominant. Also, certain high-fidelity DNA polymerases have been shown to generate tens- or hundreds-fold lower error rates [25].…”

“…4 because it is unfair to compare MAGERI with smCounter2 using QIAseq data. MAGERI’s error model is based only on primer extension assays from a mix of DNA polymerases including several high-fidelity enzymes [26], while smCounter2’s error model is specific to the entire QIAseq targeted DNA panel workflow, including DNA fragmentation, end repair and PCR enrichment steps. Because the MAGERI error model does not include errors introduced at the typical DNA fragmentation and end repair process (their assays do not have those steps), MAGERI’s background error rates are lower than those in smCounter2.…”

A novel virtual barcode strategy for accurate panel-wide variant calling in circulating tumor DNA

“…The main substitution types of our stereotypical variants were C > T/G > A, C > A/G > T, and A > G/T > C (71.05%, Fig. 4c), which were consistent with the substitution types from Oncosmart3 RSDs (Table S2-4) and previously reported error profiles for 'Kapa HF' polymerase [43]. The percentage of these six substitutions further increased to 84.297% in 121 shared sites, which demonstrated that these substitutions introduced by PCR errors were likely to occur universally (Fig.…”

“…Additionally, after polishing based on Oncosmart2, no stereotypical noises were found among the 5 Oncosmart3 RSDs at the intersection region of the two panels (Table S2-2). Stereotypical noise is caused by many factors, such as DNA damage [42] and PCR errors [43], which have different substitution preferences. The main substitution types of our stereotypical variants were C > T/G > A, C > A/G > T, and A > G/T > C (71.05%, Fig.…”

An ultrasensitive T-cell receptor detection method for TCR-Seq and RNA-Seq data

“…For TCR-Seq datasets, TCR sequences were randomly selected from the pool and assigned with corresponding abundances to mimic the heavy-tailed distribution of TCR sequences in real situations, which satisfied the Zipf distribution with the parameter of α = 3(Bolkhovskaya et al 2014). To simulate the errors introduced by PCR amplification in the library preparation procedure of TCR-Seq, selected TCR sequences were amplified approximately 18 times, with a reaction rate of − 1 and a substitution error rate of 2 × 10 −5(Shagin et al 2017), where follows a normal distribution (E = 1.90, D = 0.1)(Karlen et al 2007). ART(Huang et al 2012) was then used to simulate paired-end reads of Illumina HiSeq 2500 (default error profiles provided by ART) on DNA libraries (mean fragment length 300 bp, standard deviation 100 bp) with different read lengths (75, 100, and 150 bp) and different library sizes (5, 10, and 15 M reads).…”