Zavala, E. I., Aximu-Petri, A., Richter, J., Nickel, B., Vernot, B., & Meyer, M. (2022). Quantifying and reducing cross-contamination in single- and multiplex hybridization capture of ancient DNA. Molecular Ecology Resources, 22, 2196–2207.

Zavala et al. (2022) — Quantifying and Reducing Cross-Contamination in Single- and Multiplex Hybridization Capture of Ancient DNA

Bibliographic Reference

Zavala, E. I., Aximu-Petri, A., Richter, J., Nickel, B., Vernot, B., & Meyer, M. (2022). Quantifying and reducing cross-contamination in single- and multiplex hybridization capture of ancient DNA. Molecular Ecology Resources, 22, 2196–2207. DOI: 10.1111/1755-0998.13607.

Core Argument

Hybridization capture — the enrichment of specific genomic targets from complex DNA libraries — has become the dominant strategy for data acquisition in population-scale studies, but its throughput is limited by reagent consumption and hands-on time. Multiplex capture (pooling multiple sample libraries into a single reaction) could dramatically increase throughput, but has been avoided in degraded-DNA applications because of the elevated risk of cross-contamination via index swapping. The authors systematically characterize this cross-contamination — demonstrating that under unoptimized conditions, 7–58% of sequences assigned to a sample may originate from a different sample — and show that the mechanism is jumping PCR: incomplete extension products from one library anneal to molecules from another library during post-capture amplification, copying foreign indices. They then develop and validate a combination of laboratory protocols (increased annealing temperature via two-step PCR, individual per-well PCR cycle adjustment to avoid plateau) and a computational detection method (corner library symmetry analysis) that together suppress cross-contamination to ≤0.0004% of sequences — below the detection threshold. Crucially, they also show that even singleplex capture produces sporadic cross-contamination at 0.0008–0.021% when libraries are reamplified to PCR plateau, demonstrating that index swapping is not exclusively a multiplex concern.

Methods

• Experimental design: Four successive capture experiments, each comparing singleplex (individual) and multiplex (pooled) hybridization capture of double-indexed ancient DNA libraries using human mtDNA probes
• Sample material: Ancient sedimentary DNA from Palaeolithic sites, converted to single-stranded double-indexed libraries
• Capture protocol: Two successive rounds of hybridization capture using single-stranded biotinylated DNA probes targeting human mtDNA (Fu et al., 2013; Slon et al., 2017)
• PCR conditions compared: (1) 30-cycle amplification to PCR plateau with 60°C annealing (IS5/IS6 primers), (2) qPCR-determined cycle limit with 60°C annealing, (3) two-step PCR with 68°C annealing (IS105/IS109 primers, extended length for higher Tm)
• Sequencing: Illumina MiSeq, 2×76 bp paired-end with dual 8-bp index reads
• Cross-contamination detection method: Computational analysis of observed index pair counts — corner library identification via rectangular index combinations, symmetry expectation for true index swapping events, per-library contamination estimation by squaring the single-index exchange frequency
• Chimera detection: Comparison of shared alignment start/end coordinate proportions against a null expectation from downsampled shotgun Neandertal genome data
• Automation: Bravo NGS Workstation B with individual per-well PCR cycle adjustment in 384-well format using pre-sorted pipette tip boxes and mineral oil overlay

Key Findings

• Unoptimized multiplex capture produces >35% unexpected index pairs. After two rounds of capture with PCR to plateau (30 cycles), 35.6% of sequence molecules in the multiplex pool showed unexpected index combinations, compared to 2.1% in singleplex. Even a pool of 92 libraries subjected to 35 cycles of PCR without capture showed 33.6% unexpected index pairs — confirming that PCR, not capture itself, is the source.

• Cross-contamination in unoptimized multiplex capture reaches 7–58% per library. The computational corner-library method estimated that between 7.0 and 57.9% of sequences assigned to each expected index pair were products of cross-contamination (137,589–269,805 sequences per library), compared to 0.0008–0.021% in singleplex (9–246 sequences). Two successive index-swapping events are required for a misassignment in double-indexed libraries.

• Limiting PCR cycle number to avoid plateau dramatically reduces index swapping. When qPCR was used to stop amplification before the richest library reached plateau (12–15 cycles instead of 30), cross-contamination dropped to 0.01–2.13% in multiplex capture — a 27–58× reduction. The tradeoff is uneven sequence representation across libraries.

• Two-step PCR with higher annealing temperature (68°C) further suppresses index swapping. Extending primer length (IS105/IS109 vs. IS5/IS6) raised the annealing temperature from 60°C to 68°C and reduced unexpected index pairs from 33.6% to 8.85% in pooled amplification without capture, with cross-contamination below 0.25% per library — even at PCR plateau.

• The optimized protocol — individual cycle adjustment + two-step PCR — reduces cross-contamination to ≤0.0004%. In the final experiment with automated per-well cycle adjustment, both singleplex and multiplex capture produced <1% unexpected index pairs, and cross-contamination estimates did not exceed 0.0004% of sequences (≤25 sequences total, ≤1 per cross-contamination event). No chimeric molecules were detected.

• Even singleplex capture is not immune. Sporadic cross-contamination was detected in singleplex libraries that had been reamplified to PCR plateau pre-capture — likely from aerosol or droplet exchange during bead-based clean-up of indexed libraries. This demonstrates that cross-contamination risk exists whenever indexed libraries are amplified in parallel, regardless of whether capture is multiplexed.

• Chimeric insert sequences form when sequence complexity is low. During second-round singleplex capture (where libraries are already enriched for mtDNA), up to ~15% of sequences in some libraries were derived from chimeras — molecules formed by jumping PCR between two template molecules with the same index but different insert positions. This was absent in multiplex capture because chimeric products carry unexpected index combinations and are filtered out.

Concepts Introduced or Used

• Index swapping (via jumping PCR): Incomplete extension products from PCR amplification of one library anneal to complete or incomplete products from another library during subsequent cycles; the polymerase extends the hybrid, copying the foreign index onto the insert from the other library
• Corner libraries: Transient states where only one of the two indices has been swapped — these form the corners of a rectangle in the index-pair count matrix whose other corners are the two expected index pairs. Symmetry of corner counts is the diagnostic signal for true index swapping vs. random errors
• Cross-contamination estimation: The probability of a full index swap (both indices replaced) is the square of the single-index exchange frequency; multiplying this by the expected-pair sequence count estimates the number of contaminant sequences assigned to a library
• Chimeric molecules: Insert sequences formed by jumping PCR between molecules with different insert positions but the same index pair — inflate apparent coverage and produce false structural variant signals
• Plateau PCR: Amplification beyond the exponential phase where primer and dNTP depletion cause incomplete extension — the primary driver of jumping PCR
• Two-step PCR: Combining annealing and extension at a single elevated temperature (68°C vs. 60°C), enabled by extended primer length, to increase stringency and reduce cross-hybridization of incomplete products

Entities Referenced

• Illumina MiSeq platform
• Bravo NGS Workstation B (Agilent Technologies) — automated liquid handling
• Herculase II Fusion DNA polymerase (Agilent Technologies)
• IS5/IS6 primers (Meyer & Kircher, 2010) — standard library amplification primers with 60°C annealing
• IS105/IS109 primers — extended primers enabling 68°C two-step PCR
• leeHom (Renaud et al., 2014) — paired-end read merging
• bam-rmdup (mpieva/biohazard-tools) — PCR duplicate removal
• MEGAN (Huson et al., 2007) — taxonomic assignment via lowest common ancestor algorithm
• Denisova Cave — sediment DNA source site
• Max Planck Institute for Evolutionary Anthropology (MPI-EVA), Leipzig

Limitations (as stated by authors)

• Multiplex capture was only tested for a small genomic target (human mtDNA, ~16.6 kb). It is unclear whether multiplex capture with larger genomic targets (e.g., whole-exome, large gene panels) can be efficiently performed with degraded DNA. The authors currently recommend multiplex capture primarily for small targets or as a screening tool.
• Library-to-library variability in target content produces uneven representation. Even when libraries are pooled in equal mass, differences in the number of target molecules and copies per molecule lead to heterogeneous sequence yield across samples — multiplex capture does not produce uniform coverage across pooled libraries.
• The cross-contamination detection method requires unique double-indices for every library. Shared indices between libraries produce blind spots where contamination cannot be detected. The method cannot detect contamination between libraries that share one or both index sequences.
• The detection method depends on symmetry of corner library counts. Non-symmetric processes (sequencing errors, primer contamination, index misincorporation during synthesis) can produce corner-like patterns without true cross-contamination. Symmetry is the diagnostic, but it is a statistical expectation, not a guarantee for individual events.
• Cross-contamination across experiments is only detectable if index sequences are tracked across runs. The method evaluates all known index combinations — including those from previous experiments — but this requires maintaining a laboratory-wide index registry.

Relevance to Clonal Evolution

This paper is directly relevant to the measurement pipeline that underlies all clonal evolution inference from sequencing data. The conceptual spine of the wiki traces the flow from mutational source through VAF measurement to phylogenetic inference — and every step in that flow depends on the assumption that the reads assigned to a sample actually originate from that sample. Zavala et al. demonstrate that this assumption can be violated at rates (7–58%) that would make subclonal reconstruction impossible under unoptimized conditions, and that even under standard singleplex protocols, cross-contamination at 0.0008–0.021% persists — a level that matters when detecting rare subclonal variants at 0.1–1% VAF.

Three specific implications for clonal evolution research:

1. Cross-contamination produces phantom subclonal variants. A true somatic mutation from sample A, present at clonal frequency (VAF ~40%), will appear in sample B at the cross-contamination rate. With unoptimized multiplex capture, a 50%-CCF mutation from a contaminant library would appear as a false 3.5–29% VAF variant in the target library — indistinguishable from a genuine subclonal mutation. Even optimized singleplex protocols produce 9–246 contaminant sequences, which could manifest as false variants at 0.001–0.02% VAF in low-input samples.

2. Double-indexing alone is insufficient — PCR conditions matter more. The standard cancer genomics assumption that dual-indexed libraries prevent sample cross-talk is incorrect. The paper demonstrates that PCR conditions (cycle number, annealing temperature) are the dominant determinants of cross-contamination, not the indexing strategy itself. This has immediate practical implications for clinical sequencing pipelines: post-capture amplification must avoid PCR plateau, and two-step PCR with elevated annealing temperature should be standard.

3. Routine cross-contamination QC is feasible and should be standard. The corner-library detection method requires only index-pair counts — data that is already generated during sequencing and can be analyzed without additional experiments. The authors recommend including cross-contamination estimates in routine quality control and publishing the results alongside sequencing data. For clonal evolution studies where rare subclonal variants are the primary signal, this QC step is not optional — it distinguishes biological signal from technical artifact at the most fundamental level.

This paper joins the FFPE pre-analytical variables literature as part of a broader measurement integrity framework: FFPE artifacts (chemical damage during fixation and storage) and cross-contamination artifacts (physical misassignment during library preparation and sequencing) are the two major technical confounders that set the floor on measurable subclonal mutation fraction. Together, they define the pipeline from tissue to variant call — and the limits of what clonal evolution inference can extract from sequencing data.

Revision history

• 2026-07-18 — Source summary created. (zavala2022-cross-contamination)