Variant Allele Fraction

Definition

Variant allele fraction (VAF), also called variant allele frequency, is the proportion of sequencing reads at a genomic locus that carry a specific variant allele, expressed as a percentage. In cancer genomics, VAF is the primary observable from which clone abundance is inferred.

Relationship to Clone Frequency

VAF is a function of:

• The cancer-cell-fraction (CCF) — the true proportion of cancer cells carrying the mutation
• Sample purity (ρ) — the proportion of cancer cells in the sampled tissue
• Local copy number (N_T) — the total number of copies at the locus in the cancer cells
• Multiplicity (m) — how many of those copies carry the mutation

The relationship is: VAF = m × ρ × CCF / (ρ × N_T + 2 × (1 − ρ)) (Tarabichi et al., 2021)

The 1/f² Distribution

Under neutral-evolution in a growing population, the number of mutations as a function of their VAF follows a 1/f² distribution (Turajlic et al., 2019). This is the null model against which selection is tested. The distribution arises because each new mutation arises in a single cell when the population is of size N, and as N doubles, the mutation’s frequency halves. Older (higher-frequency) mutations are rarer because they arose when the population was smaller.

The Detection Ceiling

Bulk sequencing imposes a fundamental time bias through the VAF. “Each doubling of the cancer cell population halves the frequency of new mutations arising in the population; hence, after just seven doublings, new mutations are undetectable with 100× sequencing, and after ten doublings, new mutations are undetectable at 1,000× sequencing depth” (Turajlic et al., 2019, p. 406). Consequently, bulk sequencing mostly recovers mutations that arose early in the tumor’s history. Late-arising mutations are invisible.

Copy Number Confounding

copy-number-alterations alter the VAF independently of clone abundance. An SNV on one of three copies will have a different VAF than the same SNV on one of four copies, even if both are present in 100% of cancer cells. As Turajlic et al. (2019) note, “In a tumour sample composed of 50% cancer cells, the difference in frequency of an SNV present on one of three copies versus one of four copies is only ~3%, which is a level of accuracy that is rarely achievable with moderate-depth sequencing (~100×)” (p. 408). Without accurate copy number correction, VAF-based clone inference can produce misleading results.

FFPE-Induced VAF Artifacts

Formalin fixation and paraffin embedding — the standard preservation method for clinical tumor specimens — introduces ex vivo mutational artifacts that produce spurious low-VAF SNVs indistinguishable from true subclonal mutations (Greytak et al., 2015; greytak2015-ffpe-biospecimen-accuracy).

The artifact mechanism. Formalin cross-links proteins and nucleic acids, fragments DNA, and promotes cytosine deamination (C→U) during ambient-temperature storage. During PCR and sequencing, deaminated cytosines are read as thymines, producing artifactual C>T transitions at low allele fractions. These are chemically identical to APOBEC-mediated C>T mutations but occur ex vivo — they reflect fixation chemistry, not tumor biology. However, the FFPE artifact spectrum extends well beyond C→T: oxidation produces C→A/G→T changes, and other base substitutions (T→A, T→C) are also prevalent — UDG treatment addresses only the deamination-derived C→T fraction (Steiert et al., 2023; steiert2023-ffpe-dna-ngs-paradigms). The non-C→T artifacts are largely driven by abasic sites: depurination (~10,000 events/genome/day at physiological conditions, accelerated by formalin-induced acid hydrolysis) produces abasic sites that follow the “A-rule” — DNA polymerases preferentially insert adenine opposite the lesion, generating characteristic A:T insertions during PCR amplification (Dahlmann et al., 2009; dahlmann2009-biochemical-dna-damage).

Quantitative impact on VAF-based inference:

Effect	Magnitude	Consequence for clonal analysis
False-positive SNV rate (FFPE vs. frozen)	1–15% for NGS depending on coverage (5–80×)	1–15% of called subclonal SNVs may be fixation artifacts
GC-content bias	Correlation strongest at 40% GC (r=0.97); drops sharply outside 35–55%	Regions with extreme GC-content systematically more prone to artifact
A:T > G:C mutation bias in FFPE	Elevated transversions and transitions at A:T pairs compared to frozen	Artifactual mutations cluster by sequence context — can mimic mutational signatures
Coverage mitigation	Increasing from 20× to 40× reduces discordant loci from 1% to 0.2%	Higher coverage partially compensates but does not eliminate artifacts
Artifact VAF ceiling	AAFs can exceed 10%, particularly in low-coverage regions; the highest AAF in a 13-year-old specimen was a C→A change, not C→T	Standard 5% VAF filtering fails to remove a substantial fraction of FFPE artifacts (Steiert et al., 2023)
Dominant damage type	Oxidative base modifications → abasic sites, not deamination, are the numerically dominant FFPE lesion type; UDG alone is insufficient	Full BER pathway reconstitution (glycosylases + AP endonuclease + polymerase + ligase) is the biochemically appropriate repair strategy (Buesco et al., 2020; buesco2020-ffpe-bacterial-dna-damage)

Clinical false negatives from FFPE. Beyond false-positive artifacts, FFPE degradation can cause false-negative variant calls: DNA fragmentation reduces the number of amplifiable templates, making real mutations undetectable at standard depth. This is particularly problematic for subclonal mutations already near the detection floor — the combined effects of low CCF, CNA dilution, AND FFPE degradation can push a real mutation below the calling threshold entirely. This is one reason clinical NGS protocols typically require ≥20% tumor content (TCA): below this threshold, the joint uncertainty from purity, CNA, and FFPE degradation makes reliable variant calling impossible (cancer-cell-fraction).

Implications for subclonal reconstruction. FFPE artifacts systematically inflate the subclonal mutation fraction (SMF): every false-positive low-VAF SNV from fixation is indistinguishable from a true subclonal mutation by frequency alone. In FFPE-derived data, the standard NGS false discovery rate of 1–15% sets a floor on measurable SMF — SMF values below this floor cannot be distinguished from fixation artifact without orthogonal validation (case-matched frozen controls or targeted resequencing at higher depth). This is a critical confounder for the compression-entrenchment ITH-outcome hypothesis: most large clinical cohorts with survival data are FFPE-derived, and the SMF values used to test the U-curve prediction carry an unknown proportion of fixation-induced phantom subclonal mutations (intratumor-heterogeneity).

Analyte-specific FFPE bias. The FFPE impact hierarchy is DNA > RNA >> protein. Formalin cross-linking degrades nucleic acids but preserves proteins from the proteolytic degradation that affects even fresh-frozen samples during long-term storage (Zhu et al., 2019; zhu2019-ffpe-proteomics-pct-swath). For DNA, FFPE is a degraded substrate requiring artifact mitigation. For proteins, FFPE is a stabilized substrate — proteome patterns from 1–15-year-old FFPE samples are highly concordant with fresh-frozen counterparts. This has practical implications for multi-omic clonal evolution studies: when only FFPE tissue is available, proteomic measurements of ITH are more trustworthy than genomic measurements.