Artifactual mutations resulting from DNA lesions limit detection levels in ultrasensitive sequencing applications

Abstract

The need in cancer research or evolutionary biology to detect rare mutations or variants present at very low frequencies (<10^-5) poses an increasing demand on lowering the detection limits of available methods. Here we demonstrated that amplifiable DNA lesions introduce important error sources in ultrasensitive technologies such as single molecule PCR (smPCR) applications (e.g. droplet-digital PCR), or next-generation sequencing (NGS) based methods. Using templates with known amplifiable lesions (8-oxoguanine, deaminated 5-methylcytosine, uracil, and DNA heteroduplexes), we assessed with smPCR and duplex sequencing that templates with these lesions were amplified very efficiently by proofreading polymerases (except uracil), leading to G->T, and to a lesser extent, to unreported G->C substitutions at 8-oxoguanine lesions, and C->T transitions in amplified uracil containing templates. Long heat incubations common in many DNA extraction protocols significantly increased the number of G->T substitutions. Moreover, in ∼50-80% smPCR reactions we observed the random amplification preference of only one of both DNA strands explaining the known 'PCR jackpot effect', with the result that a lesion became indistinguishable from a true mutation or variant. Finally, we showed that artifactual mutations derived from uracil and 8-oxoguanine could be significantly reduced by DNA repair enzymes.

Keywords: DNA lesions; PCR jackpot; artifactual mutations; sequencing errors; ultrasensitive detection.

Figure 1

(A) Inserts used in the analysis of different lesions. The inserts were designed with uracils (U) on one or both strands, different mismatches placed randomly in the sequence, an 8-oxoG with a deletion (-) at the opposite position, or methylated cytosines (5-me) within a CpG context. The underlined dinucleotides represent the sequence difference between the plasmid (HSI_vector) and the vector-insert construct (HSI_insert). (B) Strategy used for the analysis of amplifiable DNA lesions. Three different DNA sources were amplified with smPCR: the vector-insert construct (HSI_insert 1-6), a plasmid DNA (HSI_vector), and human genomic DNA, and then analyzed with Sanger sequencing (or in some cases with genotyping). Duplex sequencing was performed directly on the inserts 2, 3, 5, and 6.

Figure 2

Types of nucleotide substitutions in the amplification of the 8-oxoG lesion. The percentage of different substitutions observed at the 8-oxoG site is shown for SSCSs in duplex sequencing (57,092 total analyzed reads) and Sanger sequencing reads of smPCR products (25 total analyzed smPCR reactions), detailed numbers are shown in Supplementary Table S3. GC and GT in smPCR represent different nucleotides opposite the 8-oxoG in the heterogeneous peak of the sequencing chromatogram. In duplex sequencing, these are represented as an N, where N represents the presence of more than one nucleotide in the duplex sequencing reads, with the predominant nucleotide being found in less than 70% of the reads. Error bars represent Poisson 95% CIs.

Figure 3

Amplification of uracils with different Phusion polymerases with smPCR. The amplification efficiency of Phusion Hot Start II and Phusion U was compared for samples that contain uracil in both strands (forward and reverse; HSI_insert_1 construct). Efficiency was measured as the percentage of positive smPCR reactions. In total, 372 smPCR reactions were analyzed for each condition (without USER treatment, and USER treatment before amplification). Error bars represent Poisson 95% CIs.

Figure 4

Strand-amplification bias of insert_3 with and without USER treatment measured by duplex sequencing. Based on a total of 34,072 reads for insert_3 (one uracil in the forward strand), 6,096 SSCSs were formed with the strand containing uracil (forward strand) and 27,976 for the strand without the uracil (reverse strand). After USER treatment, a total of 34,636 reads were obtained, of which, 138 and 34,498 formed SSCSs for the forward and reverse strand, respectively. Detailed numbers can be found in Supplementary Table S6. The proportion of amplifiable forward strands could be significantly decreased with USER treatment (P < 2.2 × 10 ^{− 16}, Fisher’s exact test). Error bars represent Poisson 95% CIs.

Figure 5

Measured substitutions in duplex sequencing of insert 6 in differently treated samples. Duplex sequencing of insert 6 was performed without any treatment (control), after repeated freeze-thaw cycles (frozen), and after heating to 65 °C for 3 h (heated). (A) Comparison of substitution frequencies at 5-methylcytosines (C-met) versus unmethylated C; detailed numbers are shown in Supplementary Table S9. (B) Transition and transversion frequencies in SSCS reads of differently treated/stored samples (repeatedly frozen/thawed or heated to 65 °C). Due to the relatively high number of indels introduced during DNA Ultramer synthesis, this type of mutation was not considered in the analysis. Significant differences were only observed between the control and heated sample for G->T and C->G transversions. An analysis of the nucleotide context of the observed G->T mutations is shown in Supplementary Fig. S5. Error bars represent Poisson 95% CIs. Significance values were estimated using a Fisher’s exact test.