Sequence artefacts in a prospective series of formalin-fixed tumours tested for mutations in hotspot regions by massively parallel sequencing

Abstract

Background: Clinical specimens undergoing diagnostic molecular pathology testing are fixed in formalin due to the necessity for detailed morphological assessment. However, formalin fixation can cause major issues with molecular testing, as it causes DNA damage such as fragmentation and non-reproducible sequencing artefacts after PCR amplification. In the context of massively parallel sequencing (MPS), distinguishing true low frequency variants from sequencing artefacts remains challenging. The prevalence of formalin-induced DNA damage and its impact on molecular testing and clinical genomics remains poorly understood.

Methods: The Cancer 2015 study is a population-based cancer cohort used to assess the feasibility of mutational screening using MPS in cancer patients from Victoria, Australia. While blocks were formalin-fixed and paraffin-embedded in different anatomical pathology laboratories, they were centrally extracted for DNA utilising the same protocol, and run through the same MPS platform (Illumina TruSeq Amplicon Cancer Panel). The sequencing artefacts in the 1-10% and the 10-25% allele frequency ranges were assessed in 488 formalin-fixed tumours from the pilot phase of the Cancer 2015 cohort. All blocks were less than 2.5 years of age (mean 93 days).

Results: Consistent with the signature of DNA damage due to formalin fixation, many formalin-fixed samples displayed disproportionate levels of C>T/G>A changes in the 1-10% allele frequency range. Artefacts were less apparent in the 10-25% allele frequency range. Significantly, changes were inversely correlated with coverage indicating high levels of sequencing artefacts were associated with samples with low amounts of available amplifiable template due to fragmentation. The degree of fragmentation and sequencing artefacts differed between blocks sourced from different anatomical pathology laboratories. In a limited validation of potentially actionable low frequency mutations, a NRAS G12D mutation in a melanoma was shown to be a false positive.

Conclusions: These findings indicate that DNA damage following formalin fixation remains a major challenge in laboratories working with MPS. Methodologies that assess, minimise or remove formalin-induced DNA damaged templates as part of MPS protocols will aid in the interpretation of genomic results leading to better patient outcomes.

Figure 1

Association of fragmentation of DNA in FFPE samples with low sequencing coverage. Coverage of each sample (number of reads) versus the copies of the FTH1 gene as assessed by a Taqman PCR assay (copies per microlitre). n = 253. There was a positive correlation between the FTH1 result and coverage (Spearman correlation, r = −0.29, p < 0.0001). The 50 samples with the highest C>T/G>A levels in the 1-10% allele frequency range are shown in red.

Figure 2

Significant levels of C>T/G>A sequencing artefacts in FFPE samples. (A) Assessment of sequence artefacts in cell line DNA and FFPE samples. The prevalence of each type of nucleotide change in the 1-10% allele frequency range was computed. Likely true variants identified through the Varscan2 variant caller were operationally removed to enrich for sequencing artefact changes. The graph shows all FFPE samples sorted according to the counts of C>T/G>A changes. Zoomed view: HL-60 and H1975 cell lines were used as good quality DNA controls. (B) The prevalence of each type of nucleotide change in the 10-25% allele frequency range. The graph shows all FFPE samples sorted according to the counts of C>T/G>A changes.

Figure 3

Low coverage samples have higher rates of C>T/G>A sequencing artefacts. For all FFPE samples (x-axis), values for coverage (blue) and the counts of C>T/G>A sequencing artefacts (red) are plotted on the same y-axis. There was an inverse correlation between coverage and C>T/G>A sequence artefacts (Spearman correlation, r = −0.24, p < 0.0001).

Figure 4

Uracil-DNA glycosylase treatment of FFPE DNA samples distinguishes true and false positive clinical relevant mutations. Integrative Genomic Viewer (IGV) screenshots of two breast cancers and one melanoma sample pre- and post- uracil-DNA glycosylase (UDG) treatment samples. The two breast cancer samples have confirmed PIK3CA mutations (E545K for Ca309 and H1047Y for Ca285) as these mutations were still detected after UDG treatment. The NRAS G12D mutation identified in the pre-UDG sample (Ca97) was a false positive as it was not present after UDG treatment. The variant reads over the total reads and overall allele frequency (a.f.) are shown for each case.

Figure 5

Low template copies are associated with higher probability of sequencing artefacts post-PCR amplification. In good quality DNA from sources such as blood and fresh frozen tissue, fragmentation and uracil lesions are present at very low levels. In this circumstance, high amounts of amplifiable template increase the likelihood of accurately identifying mutations due to high sequencing coverage with little or no stochastic enrichment of sequencing artefacts. In FFPE DNA with moderate fragmentation, the number of amplifiable templates is reduced, with some formalin-induced uracil lesions being present in template DNA. Subsequently PCR amplification results in lower coverage due to less amplifiable template numbers. Uracil lesions are also amplified, and due to the lower copy numbers, can appear as non-reproducible sequencing artefacts (C>T/G>A changes). These artefacts will be low in frequency. In the case of FFPE with high amounts of fragmentation, the numbers of amplifiable template are severely limited. An artefact in one of these templates can then appear as a moderate to high frequency sequencing variant. These can subsequently be interpreted as real mutations.