Sequencing Library Artifacts

Summary

Sequencing library artifacts are technical errors introduced during library preparation, amplification, and sequencing that produce false variant calls indistinguishable from true somatic mutations by frequency alone. The two dominant classes are index swapping (misassignment of reads between samples via jumping PCR during library amplification) and chimeric molecules (hybrid insert sequences formed by incomplete extension products annealing across templates). Under unoptimized conditions, cross-contamination from index swapping can reach 7–58% of assigned sequences per library (Zavala et al., 2022) — sufficient to destroy subclonal resolution entirely. Even under standard singleplex protocols, sporadic cross-contamination at 0.0008–0.021% persists, and chimeric insert sequences can constitute up to ~15% of reads in some libraries. Optimized protocols — individual per-well PCR cycle adjustment to avoid plateau, two-step PCR with elevated annealing temperature (68°C), and computational cross-contamination detection via corner library symmetry analysis — suppress cross-contamination to ≤0.0004% (≤1 sequence per event). These artifacts, together with FFPE chemical damage, define the technical noise floor below which true subclonal mutations cannot be reliably distinguished from artifacts without orthogonal validation.

Index Swapping and Cross-Contamination

Mechanism

Index swapping occurs via jumping PCR during library amplification (Zavala et al., 2022; Figure 1):

1. PCR amplification of a DNA library molecule normally produces a complete
   double-stranded product with intact index sequences on both ends.

2. When amplification reaches plateau — primer or dNTP depletion, polymerase
   saturation — incomplete extension products accumulate. These are
   single-stranded molecules truncated before the distal index.

3. In subsequent cycles, an incomplete extension product from library A can
   anneal to a complete or incomplete product from library B through
   complementary adapter sequences flanking the index.

4. The polymerase extends the hybrid, copying library B's index onto
   library A's insert (or vice versa). The result is a chimeric molecule
   carrying library A's insert with one index from library B.

5. If this happens a SECOND time — the singly-swapped molecule participates
   in another jumping PCR event — both indices are replaced, and the read
   is misassigned to library B while carrying library A's sequence.

Two successive index-swapping events are required for a complete misassignment in double-indexed libraries. The intermediate state — where only one index has been swapped — produces a “corner” combination in the index-pair count matrix. Symmetry of corner counts (both directions of exchange) is the diagnostic signature that distinguishes true jumping PCR from random errors.

Magnitude

Condition	Unexpected index pairs	Cross-contamination (per library)
Multiplex + PCR to plateau (30 cycles)	35.6%	7.0–57.9%
Multiplex + limited cycles (12–15)	—	0.01–2.13%
Multiplex + two-step PCR (68°C), even at plateau	8.85%	<0.25%
Optimized (limited cycles + two-step PCR + individual adjustment)	<1%	≤0.0004%
Singleplex + PCR to plateau (re-amplified libraries)	2.1%	0.0008–0.021%

Key insight: multiplexing amplifies but does not create the problem. Even singleplex capture — when libraries are amplified to plateau — produces detectable cross-contamination, likely from aerosol or droplet exchange during bead-based clean-up of libraries processed in parallel (Zavala et al., 2022).

Computational Detection: The Corner Library Method

Zavala et al. (2022) provide a simple method that requires only the observed counts of index pairs in a sequencing pool:

1. Tabulate all observed P5/P7 index pair combinations and their sequence counts
2. Identify corner pairs: For any two expected index pairs (P5_A, P7_A) and (P5_B, P7_B), the combinations (P5_A, P7_B) and (P5_B, P7_A) are “corner libraries” — they have one index from each expected pair
3. Check for symmetry: True index swapping produces approximately equal counts of both corner combinations. Asymmetric counts suggest sequencing errors or primer contamination rather than jumping PCR
4. Estimate single-exchange frequency: The number of corner-library sequences divided by the expected-pair sequences approximates the probability that one index was replaced
5. Square to get cross-contamination: The probability that BOTH indices were replaced — producing a misassigned read — is the square of the single-exchange frequency (two independent events)
6. Multiply by expected-pair count: (single-exchange frequency)² × expected-pair sequences = estimated contaminant sequences assigned to that library

The search space extends beyond the current library pool to all “known” index combinations — those physically available in the laboratory or used in previous experiments. This enables detection of cross-contamination across experiments and library pools.

Mitigation

Strategy	Mechanism	Effectiveness
Limit PCR cycles to avoid plateau	Prevents incomplete extension product accumulation	Reduces cross-contamination 27–58×
Two-step PCR (68°C annealing)	Extended primers (IS105/IS109) increase annealing stringency; fewer incomplete products cross-hybridize	Reduces unexpected index pairs from 33.6% → 8.85% even at plateau
Individual per-well cycle adjustment	Each reaction stops before its own plateau, determined by qPCR	Eliminates the need to limit all reactions to the richest library’s cycle number
Computational QC (corner library analysis)	Detects contamination events post hoc from index-pair counts	Identifies contamination across experiments; requires only index-pair count data already generated
Unique dual indices (UDI) + index registry	Ensures each library has a unique index pair; tracks indices across experiments	Enables cross-experiment contamination detection; blind spots where libraries share indices
Avoid reamplification of indexed libraries	Reamplification to plateau is the primary vector for sporadic singleplex contamination	Eliminates the most common source of pre-capture index swapping

Implications for Clonal Evolution Research

Cross-contamination sets a technical floor on measurable subclonal mutation fraction that is distinct from — and additive to — the FFPE chemical artifact floor. A true subclonal mutation at 1% VAF in a sample sequenced with unoptimized multiplex capture cannot be distinguished from a contaminant read from another sample’s clonal mutation at 40% VAF entering at 2.5% cross-contamination rate. The two would produce the same number of variant-supporting reads.

The practical consequence for subclonal reconstruction (Tarabichi et al., 2021): the minimum detectable CCF is limited not by sequencing depth alone, but by the combined artifact rate from FFPE damage + cross-contamination + sequencing error. In a sample with 0.1% cross-contamination and 1% FFPE artifact rate, subclonal mutations below ~1.1% VAF are indistinguishable from technical noise — regardless of sequencing depth.

Chimeric Molecules

Mechanism

Chimeric insert sequences are hybrid molecules formed when a jumping PCR event anneals two incomplete extension products that share the same index pair but have different insert positions (Zavala et al., 2022; Figure 1e). Unlike index swapping — which requires two events to cause misassignment — a single jumping PCR event can create a chimeric insert.

Conditions That Promote Chimera Formation

Chimeras form predominantly when sequence complexity is low — i.e., when libraries are already enriched for sequences similar to each other. This occurs during:

• Second and subsequent rounds of hybridization capture (libraries are already enriched for target molecules)
• Deep amplicon sequencing of small genomic regions
• Any situation where the same genomic region is represented by many highly similar molecules

Detection

Chimeric molecules cluster at identical alignment start or end coordinates — far more than expected by chance. Zavala et al. (2022) detected chimeras by comparing the proportion of shared alignment start/end coordinates against a null distribution from downsampled shotgun data. In some second-round singleplex capture libraries, up to ~15% of sequences showed excess sharing — indicating chimeric origin.

Why Multiplex Capture Suppresses Detectable Chimeras

Chimeras formed between molecules with different index pairs carry unexpected index combinations and are removed by standard index-filtering in downstream analysis. This is why chimeras were detected in singleplex capture (same index on all molecules in a reaction) but not in multiplex capture (inter-library chimeras are filtered out) — the chimeras still form, but they’re excluded from analysis because their index combinations don’t match expected pairs.

Mitigation

The same PCR optimization strategies that suppress index swapping also suppress chimera formation:

• Avoid PCR plateau (particularly for second-round capture products)
• Two-step PCR with elevated annealing temperature
• Individual per-well cycle adjustment

Relationship to FFPE Artifacts

Sequencing library artifacts and FFPE artifacts are the two major technical confounders in clonal evolution studies, but they operate through fundamentally different mechanisms:

Dimension	FFPE Artifacts	Library Artifacts
Origin	Chemical damage during fixation and storage	Physical misassignment during library prep and sequencing
Mechanism	Deamination, oxidation, depurination of DNA bases	Jumping PCR, index swapping, aerosol/droplet cross-contamination
Timing	Pre-analytical (tissue → DNA extract)	Analytical (DNA extract → sequencing reads)
Signature	C→T at CpG (deamination); C→A, G→T (oxidation); sequence-context-dependent	Sample-to-sample contamination detectable via index analysis; not sequence-context-dependent
Detection	Excess low-VAF C→T variants; highest AAF often at C→A (oxidation)	Corner library symmetry in index-pair count matrix; unexpected index combinations
Mitigation	UDG + BER repair; molecular tagging; high coverage	Limited PCR cycles; two-step PCR; per-well cycle adjustment; computational QC
False positive type	Single-base substitutions at damaged sites	Entire reads misassigned to wrong sample
VAF range of artifacts	0.1–15% depending on specimen age and coverage	0.0004–58% depending on PCR conditions and multiplexing

These artifact classes are additive and independent. A sample processed from FFPE tissue with unoptimized multiplex capture carries both chemical damage artifacts (C→T at low VAF from deamination) and cross-contamination artifacts (reads from other samples appearing as false variants). Disentangling them requires:

• Index-level QC (corner library analysis) to quantify cross-contamination
• UDG/BER repair + sequence-context analysis to quantify FFPE artifacts
• Orthogonal validation (matched FF, targeted resequencing) to establish the true biological signal

Together, FFPE artifacts and library artifacts set the combined noise floor for subclonal variant detection — the VAF below which a variant cannot be confidently called as biological rather than technical.

Cross-Contamination QC Protocol

Based on Zavala et al. (2022), a routine cross-contamination QC check for any sequencing run consists of:

1. Before sequencing: Ensure all libraries carry unique double-indices. Maintain a laboratory index registry to enable cross-experiment contamination detection. Do not reamplify indexed libraries to PCR plateau before capture.

2. During library preparation: Use two-step PCR with the highest possible annealing temperature (68°C if using extended IS105/IS109-type primers). Determine PCR cycle number by qPCR for each library or pool — stop before the richest library reaches plateau.

3. After sequencing: Extract index-pair counts from the sequencing output. Run the corner library analysis to estimate per-library cross-contamination. Flag any library with >0.1% estimated cross-contamination for investigation.

4. Report: Include cross-contamination estimates in the sequencing QC report alongside standard metrics (duplication rate, coverage, on-target rate). A sample’s variant calls are only as trustworthy as its index purity.

For clonal evolution studies that depend on detecting rare subclonal variants (VAF <5%), this QC protocol is not optional. A sample with 1% cross-contamination will have 1% of reads originating from other samples — and those contaminant reads will carry variants at whatever frequency they occur in the source samples, producing false-positive subclonal calls at exactly the VAF range of greatest interest.

Limitations

1. The corner library method requires unique double-indices for every library in the search space. Libraries that share one or both index sequences create blind spots where contamination is invisible. This is a library design constraint, not a method limitation — the fix is to use unique dual indices (UDIs).

2. The symmetry assumption is statistical, not deterministic. Individual corner pairs may be asymmetric due to stochastic sampling, PCR biases, or sequencing errors. The method works at the level of pooled estimates across all library pairs, not for individual pairs with low counts.

3. Cross-contamination estimation has a detection floor. At very low contamination rates (≤0.0004%), the corner library signal is too sparse to detect reliably — but at that level, the practical impact on variant calling is negligible for most applications.

4. The methods described were validated on mtDNA capture — a small (~16.6 kb) target. Large-panel and exome capture involve orders of magnitude more unique molecules, which may change the dynamics of jumping PCR and chimera formation. The principles (avoid plateau, increase annealing temperature) are expected to generalize, but the specific contamination rates at each condition may differ.

5. Chimera detection requires alignment data and is insensitive at low coverage. The shared coordinate method only detects chimeras when sequencing depth is sufficient to produce multiple reads with identical start/end positions by chance under the null model. At low coverage, chimeras are present but undetectable.

Revision history

• 2026-07-18 — Page created from Zavala et al. (2022) synthesis. (zavala2022-cross-contamination)