Do, H., & Dobrovic, A. (2015). Sequence artifacts in DNA from formalin-fixed tissues: Causes and strategies for minimization. Clinical Chemistry, 61(1), 64–71.

Do & Dobrovic (2015) — Sequence Artifacts in DNA from FFPE Tissues

Bibliographic Reference

Do, H., & Dobrovic, A. (2015). Sequence artifacts in DNA from formalin-fixed tissues: Causes and strategies for minimization. Clinical Chemistry, 61(1), 64–71. DOI: 10.1373/clinchem.2014.223040.

Core Argument

DNA extracted from FFPE tissues contains specific types of molecular damage — crosslinks, fragmentation, abasic sites, and deaminated bases — that are the direct sources of sequence artifacts in downstream molecular analysis. These artifacts predominantly manifest as C:G→T:A transitions (60–80% of artifactual SNVs), are chemically indistinguishable from true somatic mutations, and can mimic clinically actionable variants (e.g., KRAS codon 12/13, EGFR T790M). The authors provide a mechanistic taxonomy of DNA lesions and a practical toolkit of strategies — UDG pretreatment, short amplicons, high-fidelity polymerases, molecular tagging, and capture-based sequencing — to minimize artifacts and distinguish true mutations from fixation-induced false positives.

Methods

This is a mini-review focused on the mechanistic basis of FFPE-induced sequence artifacts. It synthesizes evidence from targeted amplicon sequencing studies, Sanger sequencing comparisons, and emerging methods (Safe-SeqS, duplex sequencing, capture-based MPS) to characterize DNA damage types and evaluate mitigation strategies. No new primary data are presented; the paper is a conceptual synthesis of published experimental evidence organized around a mechanistic framework.

Key Findings

• Cytosine deamination is the dominant artifact source. Hydrolytic deamination of cytosine to uracil occurs at ~70–200 events/day per cell in vivo (repaired by UNG). Ex vivo, in FFPE tissues, uracil lesions accumulate unrepaired. During PCR, DNA polymerase incorporates adenine opposite uracil → artifactual C:G→T:A transitions. This single mechanism accounts for 60–80% of sequence artifacts in FFPE DNA. An additional contribution comes from deamination of 5-methylcytosine (5-mC) to thymine at CpG dinucleotides — 5-mC deaminates ~2× faster than unmethylated cytosine and is the most deamination-susceptible base in DNA. This produces C:G→T:A artifacts concentrated at CpG sites.

• The artifact spectrum is consistent across studies. Across four independent investigations: C:G→T:A transitions dominated (42–100%), followed by C:G→A:T (0–35%) and A:T→G:C (0–35%). The specific proportions depend on the gene, GC-content, and whether UDG pretreatment was used. Critically, artifactual C:G→T:A SNVs at CpG sites and TpC sites are chemically identical to spontaneous age-related (SBS1) and APOBEC-mediated (SBS2/SBS13) mutations, respectively — meaning FFPE artifacts can masquerade as endogenous mutational signatures.

• Clinically actionable mutations are directly affected. Artifactual KRAS codon 12/13 mutations were found in 53/993 (5%) of FFPE colorectal cancers (Lamy et al.). EGFR T790M — the gatekeeper resistance mutation predicting response to third-generation EGFR TKIs — was detected in 41.7% of FFPE lung tumors vs. only 1 matched frozen tumor, indicating the vast majority of FFPE-detected T790M calls in that study were artifacts. A systematic review of 3,381 EGFR mutations in 12,244 NSCLC patients found that 71% were singletons — seen in only a single case — suggesting many reported EGFR mutations may be fixation artifacts.

• UDG pretreatment removes uracil-derived artifacts. Treatment of FFPE DNA with uracil-DNA glycosylase before PCR selectively removes uracil bases from U:G mismatches, generating abasic sites that block extension by most DNA polymerases. This reduces artifactual C:G→T:A SNVs by 60–80%. However, UDG does not remove thymine lesions from 5-mC deamination, so residual C→T artifacts at CpG sites remain — these would require MBD4 or thymine-DNA glycosylase treatment.

• Template quantity is the critical pre-analytic variable. Fragmentation directly reduces the number of amplifiable templates. Both spectrophotometry and fluorometry seriously overestimate amplifiable template quantity in FFPE DNA; qPCR or digital PCR with amplicon sizes matching the sequencing protocol is recommended. When amplifiable template numbers are low, stochastic variation in allelic representation causes false negatives (true mutations lost to sampling), while DNA lesions become stochastically enriched, increasing false positives. The fewer the templates, the higher both the false-negative and false-positive rates — a dual threat to accurate variant calling.

• Molecular tagging distinguishes artifacts from true mutations. Safe-SeqS (unique identifier tagging of individual templates) reduces error rates ~20-fold by requiring that a variant be present in ≥95% of reads from the same UID family. Duplex sequencing — tagging both strands of each template — achieves sensitivities of 1:10,000 by requiring that a variant be present in both strands of the same molecule. Because DNA lesions affect only one strand, artifacts are strand-specific; true mutations are present in both strands. This is the most principled solution to the artifact problem.

• Capture-based sequencing provides inherent template tracking. Unlike amplicon-based methods where all reads from the same amplicon are identical, capture-based approaches retain variable insert sequences that serve as natural template identifiers. A variant seen in multiple independent templates with different insert sequences is a true mutation; a variant seen in only one template (even with high read depth) should be interpreted with caution.

Concepts Introduced or Used

FFPE DNA damage, cytosine deamination, uracil lesion, 5-methylcytosine deamination, abasic site, DNA crosslinking, DNA fragmentation, UDG pretreatment, high-fidelity polymerase, Safe-SeqS, duplex sequencing, unique molecular identifier (UID), capture-based sequencing, template quantity estimation, stochastic enrichment

Entities Referenced

• EGFR — T790M resistance mutation; frequently artifactual in FFPE
• KRAS — codon 12/13; 5% artifactual rate in FFPE colorectal cancer
• BRAF, KIT — additional clinically actionable genes vulnerable to FFPE artifacts
• UNG/UDG — uracil-DNA glycosylase; the key enzymatic countermeasure
• MBD4, thymine-DNA glycosylase — potential countermeasures for 5-mC deamination artifacts
• Pfu, KAPA — family B DNA polymerases with uracil read-ahead function
• Safe-SeqS, duplex sequencing — molecular tagging methods for artifact detection

Limitations (as stated by authors)

• UDG treatment does not remove thymine lesions from 5-mC deamination — residual artifacts at CpG sites remain
• Bioinformatic filtering can reduce artifacts but risks false negatives for clinically important low-level resistance mutations
• Capture-based approaches require more setup time and may require DNA shearing
• The review focuses on preexisting template damage; other artifact sources (oxidative damage during preparation, polymerase errors, pseudogene amplification, adaptor artifacts) are noted but not reviewed in detail

Relevance to Clonal Evolution

This paper provides the molecular mechanism underlying the FFPE accuracy problem surveyed by Greytak et al. (2015). Its relevance to clonal evolution is threefold:

1. FFPE C→T artifacts mimic APOBEC mutagenesis. Both APOBEC3A/B-mediated mutagenesis and FFPE-induced deamination produce C:G→T:A transitions via cytosine → uracil → adenine read-through. APOBEC prefers TpC contexts; FFPE deamination is concentrated at CpG sites (via 5-mC). But the chemical pathway is identical. In FFPE-derived mutational catalogs, a fraction of SBS2/SBS13-attributed mutations may be fixation artifacts — particularly those at low VAF, where both APOBEC subclonal mutations and FFPE artifacts cluster. This is a direct confounder of mutational signature analysis in FFPE cohorts (APOBEC-mutagenesis, mutational-signature).

2. The template-quantity problem compounds the CNA-discarding problem. The wiki already documents that SNVs in CNA regions are discarded because multiplicity (m) is unknown (copy-number-alteration §5.4). Do & Dobrovic add an orthogonal mechanism: even in diploid regions, FFPE DNA fragmentation reduces amplifiable template numbers, causing stochastic dropout of true low-VAF mutations (false negatives) AND stochastic enrichment of DNA lesions into apparent low-VAF variants (false positives). The joint effect is that the SMF measured from FFPE-derived data reflects an unknown mixture of true subclonal mutations, CNA-induced miscalibration, and FFPE-induced artifacts — three confounders with identical phenotypic signatures (low VAF) but distinct origins.

3. The 20%-tumor-content rule has a mechanistic basis. The clinical practice of rejecting samples with <20% tumor content (TCA) — mentioned in the wiki’s discussion of purity estimation — is rooted in the template-quantity dynamics described here. Below 20% tumor content, the number of amplifiable tumor-derived templates is so low that (a) true clonal mutations approach the stochastic dropout threshold, producing false negatives, and (b) FFPE DNA lesions in normal-cell templates, combined with stochastic enrichment, produce false positives at frequencies approaching those of true subclonal mutations. The 20% threshold is the point below which the true signal and the artifact noise become statistically inseparable at standard sequencing depths. For clonal evolution studies using FFPE archives, tumor content is not merely a confounder to be adjusted for — it is a hard filter below which the data are unreliable (cancer-cell-fraction, variant-allele-fraction).

Molecular tagging as a partial solution. The Safe-SeqS and duplex sequencing methods described here provide a principled way to distinguish FFPE artifacts from true low-frequency mutations: artifacts are strand-specific (present in one strand only), while true mutations are present in both. For future clonal evolution studies using FFPE tissue, molecular tagging offers the prospect of artifact-free low-frequency variant detection — enabling reliable SMF measurement from archival specimens that currently produce confounded data.