Identification, quantification, and elimination of barcode crosstalk in multiplexed Oxford Nanopore sequencing

This paper identifies and quantifies barcode crosstalk as a critical source of error in multiplexed Oxford Nanopore sequencing, particularly for low-biomass samples, and introduces post-ligation pooling (PLP) as an effective library-preparation modification to prevent rather than merely mitigate these false-positive signals.

The Gist

Imagine you are hosting a massive dinner party where 96 different guests (DNA samples) are all eating at the same long table. To keep track of who ate what, you give every guest a unique, color-coded name tag (a barcode).

In a perfect world, you would hand out the name tags, let everyone eat, and then ask, "Who ate the pizza?" The person with the "Red Tag" would say, "I did!" and the person with the "Blue Tag" would say, "I didn't."

But in the world of DNA sequencing (specifically with Oxford Nanopore technology), something weird was happening. It was as if the name tags were slipping off some guests and accidentally sticking onto others while they were eating.

The Problem: "Name Tag Hopping" (Barcode Crosstalk)

The researchers discovered a major glitch in how they were preparing these DNA samples.

The Old Way (The Standard Protocol):
Imagine the host gathers all 96 guests together in the kitchen before they sit down to eat. The host tries to hand out the name tags to the whole crowd at once. Because everyone is crowded together, some name tags get stuck to the wrong people, or extra loose tags float around and stick to the wrong person's shirt.

When the guests finally sit down and eat, the host looks at the table and sees a "Red Tag" person eating a "Blue Tag" meal. The host thinks, "Oh no! The Blue Tag person is contaminated! They ate the Red Tag food!"

In reality, the Blue Tag person didn't eat the food; their name tag just got swapped. In scientific terms, this is called barcode crosstalk. It creates "false positives," making it look like a sample has DNA it doesn't actually have. This is a disaster for experiments with very low amounts of DNA (like water samples or single cells), because a tiny bit of "hopping" can look like a huge amount of contamination.

The Solution: "Post-Ligation Pooling" (PLP)

The researchers, led by Qing Dai and Joshua Granek, came up with a simple fix called Post-Ligation Pooling (PLP).

The New Way (PLP):
Instead of gathering everyone in the kitchen to hand out name tags, the host changes the order of operations:

1. First: Each guest sits at their own private table and gets their name tag attached securely.
2. Second: Only after everyone has their own tag securely attached do they all move to the big shared table to eat together.

By keeping the guests separate until the name tags are permanently glued on, there is no chance for a tag to slip off and stick to the wrong person.

What They Found

The team tested this in two ways:

1. The Real-World Test: They looked at water samples from a building's plumbing. Using the "Old Way," the water samples looked like they were full of bacteria from the positive control samples (like a ghost in the machine). Using the "New Way" (PLP), those ghost bacteria disappeared. The water samples were clean, just as they should be.
2. The Lab Test: They used pure, known bacteria to see exactly how often the "name tag hopping" happened.
   • Old Way: About 0.88% of the time, the wrong name tag was on the wrong DNA. That sounds small, but in science, that's a huge error rate.
   • New Way (PLP): The error rate dropped to 0.019%.
   • Super New Way (PLP + SFB): By adding a special "wash" step (SFB) to the new method, they got it down to 0.015%.

Why This Matters

Think of it like a security checkpoint at an airport.

• The Old Way: If 1% of people get the wrong boarding pass, the airport might think a terrorist is in the wrong terminal, or miss a real threat because the data is messy.
• The New Way: The security is tight. You know exactly who is where.

This is crucial for:

• Low-Biomass Samples: If you are testing a drop of water for bacteria, you don't want to be fooled by a "ghost" bacteria that just jumped from another sample.
• Medical Diagnostics: If a doctor is looking for a rare mutation in a patient's DNA, they need to be 100% sure that signal is real and not just a "hopped" barcode from another patient.

The Takeaway

The researchers didn't just find a way to fix the data after the fact (like deleting the wrong answers); they found a way to prevent the mistake from happening in the first place.

They showed that by simply changing the order of steps—waiting to mix the samples until after the barcodes are attached—you can stop the "name tag hopping." This makes Oxford Nanopore sequencing much more reliable, cheaper (because you don't waste money sequencing fake data), and trustworthy for scientists studying everything from environmental water to human disease.

Technical Summary

1. Problem Statement

The paper addresses barcode crosstalk (also known as barcode hopping), a critical source of error in multiplexed sequencing where reads are incorrectly assigned to the wrong sample.

• Mechanism: Unlike demultiplexing errors (caused by sequencing base-calling mistakes), barcode crosstalk is an upstream physical failure where an incorrect barcode becomes attached to a nucleic acid fragment during library preparation or on-flowcell amplification.
• Consequences: This phenomenon mimics cross-contamination, generating false-positive signals. It is particularly detrimental in:
  • Low-biomass samples: Where trace contamination can dominate the signal.
  • Imbalanced designs: Where high-abundance samples "leak" into low-abundance samples.
  • Quantitative analyses: Leading to inflated diversity estimates and distorted abundance data.
• Context: While Unique Dual Indexes (UDIs) are standard for mitigating crosstalk on Illumina platforms (by allowing bioinformatic filtering), they do not prevent the physical occurrence of crosstalk and reduce usable data. Furthermore, existing solutions for Oxford Nanopore Technologies (ONT) were insufficient.

2. Methodology

The authors employed a multi-phase approach involving real-world metagenomics, controlled defined-genome experiments, and a novel workflow modification.

A. Workflow Modification: Post-Ligation Pooling (PLP)

The core innovation is Post-Ligation Pooling (PLP).

• Standard ONT Workflow: Samples are pooled immediately after index barcoding but before adapter ligation. This creates a shared environment where excess free barcodes can ligate to DNA from other samples.
• PLP Workflow: Samples remain separate until after the final adapter ligation step. Pooling occurs only after ligation is complete. This physically prevents free barcodes from one sample from ligating to the DNA of another.

B. Experimental Designs

1. Real-World Metagenomics:

   • Samples: Built-environment water samples (p-trap), positive controls (DCS lambda, $\Phi$X174), and negative controls (water blank, empty blank).
   • Comparison: Standard protocol vs. PLP protocol.
   • Analysis: Taxonomic profiling (MEGAN) and microbial source tracking (FEAST) to quantify the origin of reads in negative controls.

2. Controlled Defined-Genome Experiment:

   • Design: A Latin-square-inspired design to eliminate batch effects. Four distinct DNA sources (DCS lambda, Mycoplasma hominis, Phocaeicola vulgatus, Corynebacterium striatum) were used.
   • Treatments Compared:
     1. Standard ONT protocol.
     2. PLP (Post-Ligation Pooling).
     3. SFB (Short Fragment Buffer) cleanup (a recent ONT protocol update).
     4. Combined SFB + PLP.
   • Metric: Reads assigned to a barcode but mapping to a different genome were counted as crosstalk-induced misassignments.

3. Key Contributions

• Identification: First comprehensive characterization of substantial barcode crosstalk in ONT Native Barcoding Kit workflows.
• Quantification Framework: Development of metrics to measure crosstalk, including "misassigned reads per million" and standard classification metrics (Precision, Recall, F1, Misassignment Rate).
• Solution Development: Introduction of PLP, a simple, cost-free "drop-in" modification to library preparation that prevents crosstalk rather than just mitigating it.
• Comparative Analysis: Demonstration that PLP is superior to or complementary with the recently released SFB cleanup method.

4. Key Results

A. Metagenomic Validation

• Standard Protocol: Negative controls (blanks) contained sequences matching the positive controls and environmental samples, indicating severe apparent cross-contamination. Source tracking (FEAST) attributed ~90% of these reads to barcode crosstalk.
• PLP Protocol:
  • Reduced reads in negative controls by 1–2 orders of magnitude (e.g., 1043 reads $\to$ 77 reads in water blanks).
  • Residual reads in PLP blanks did not match microbial taxonomy, suggesting they were non-microbial artifacts.
  • Nearly eliminated the appearance of positive control reads in other samples.

B. Quantitative Crosstalk Reduction

In the controlled defined-genome experiment:

• Standard Protocol: Exhibited a misassignment rate of 0.882% (thousands of misassigned reads per million).
• SFB Cleanup: Reduced misassignment to 0.067%.
• PLP: Reduced misassignment to 0.019%.
• SFB + PLP: Achieved the lowest misassignment rate of 0.015%.
• Conclusion: PLP alone reduces crosstalk by approximately one order of magnitude compared to the standard protocol, and combining it with SFB yields the highest fidelity.

5. Significance and Implications

• Data Integrity: The study proves that standard ONT native barcoding workflows generate significant false positives, which can invalidate conclusions in low-biomass studies (e.g., single-cell sequencing, environmental DNA, clinical samples).
• Superiority over UDIs: Unlike UDIs (which filter data post-sequencing), PLP prevents the physical occurrence of crosstalk. This preserves sequencing capacity and data volume that would otherwise be discarded.
• Immediate Adoption: The PLP workflow is a simple modification requiring no new reagents or expensive equipment. The authors strongly urge the community to adopt PLP (or SFB+PLP) immediately.
• Retrospective Caution: Researchers analyzing historical ONT data generated with the standard Native Barcoding Kit must consider that observed "contamination" or low-abundance signals may actually be barcode crosstalk artifacts.

Summary: This paper identifies a critical flaw in standard ONT multiplexing, quantifies its impact, and provides a simple, highly effective protocol modification (PLP) that reduces barcode misassignment from ~0.9% to ~0.015%, significantly enhancing the reliability of long-read sequencing for sensitive applications.