Barcode Crosstalk in ONT Multiplex Sequencing: Quantification and Mitigation Strategies

This study quantifies barcode crosstalk in Oxford Nanopore multiplex sequencing, demonstrating that while updated protocols significantly reduce misassignment compared to older versions, pooling barcoded samples only after adapter ligation (an in-house protocol) offers the most effective mitigation for low-input DNA samples.

The Gist

The Big Idea: The "Mix-Up" at the DNA Party

Imagine you are hosting a massive party (a DNA sequencing run) where you invite 24 different guests (24 different bacterial samples). To make sure you know who is who when they leave, you give each guest a unique, colored name tag (a barcode) before they enter the dance floor.

In an ideal world, everyone keeps their own name tag, and when you look at the photos later, you know exactly which guest did which dance move.

However, the researchers in this paper discovered a problem: Barcode Crosstalk.

Sometimes, during the preparation, a few name tags get ripped off one guest and accidentally stuck onto another guest's shirt. When the photos are taken later, you might think the "wrong" guest did the dance, or you might think a guest was there when they weren't. This creates "ghost data" that messes up your results.

The Problem: Why It Happens

The paper explains that this mix-up happens because of how the "name tags" are attached.

1. The Old Way (Protocol A): You put all the guests in a big room, hand out the name tags, wash off the extra tags with a standard cleaner (ethanol), and then attach the final "admission tickets" (sequencing adapters) to the whole group at once.
   • The Flaw: The wash wasn't strong enough. Some loose name tags remained in the room. When the admission tickets were handed out, those loose tags got stuck onto the wrong people.
   • The Result: If you had a huge crowd of one type of guest (high DNA) and a tiny, lonely group of another (low DNA), the loose tags from the big crowd would jump onto the lonely guests. The lonely guests would suddenly look like they were part of the big crowd.

The Experiments: Testing Three Solutions

The scientists tested three different ways to run this party to see which one stopped the mix-ups.

1. The Old Standard (Protocol A)

• The Method: The standard Oxford Nanopore instructions used until recently.
• The Result: Disaster for small groups. When the DNA input was low (the "lonely guests"), up to 2.4% of the data was wrong. It was like looking at a photo of a quiet librarian and suddenly seeing a football player dancing in the background because a loose name tag got stuck on them.

2. The Manufacturer's Fix (Protocol B)

• The Method: Oxford Nanopore realized the problem and updated their instructions. Instead of using the standard cleaner (ethanol), they used a special "Short Fragment Buffer" (SFB) to wash the guests.
• The Result: Much better, but not perfect. The mix-ups dropped from 2.4% down to almost nothing (0.01%). It's like using a super-strong magnet to pull the loose tags off before they can stick to the wrong people. However, a tiny bit of "ghost data" still remained.

3. The "Super Safe" Home-Brew Method (Protocol C)

• The Method: The researchers changed the order of operations. Instead of mixing everyone together before the final step, they kept every guest in their own private room until the very last second. They attached the admission tickets individually, and only then did they mix everyone into one big group.
• The Result: Perfect. The mix-ups were virtually eliminated (0.0000%). Because the guests were never in the same room while the loose tags were floating around, the tags couldn't jump to the wrong person.

The Trade-Off: Cost vs. Perfection

The paper concludes with a very practical piece of advice:

• If you are rich and want 100% perfection: Use Protocol C. It is the "Gold Standard." However, it is expensive and slow because you have to process every sample individually, like paying a private chef for every single guest instead of a buffet.
• If you want a good balance: Use Protocol B (the new manufacturer update). It's cheap, fast, and reduces the error rate so much that it's usually fine for most jobs.
• When to be careful: If you are studying very rare samples (like finding a single virus in a huge drop of blood) or using the old "Protocol A," you must be very careful. The "ghost data" could trick you into thinking you found something that isn't there.

The Bottom Line

This paper is a warning label for scientists using DNA sequencing. It says: "Be careful with your name tags!"

If you are sequencing samples with very little DNA, the old methods might give you fake results. The new methods are better, but if you need absolute certainty, the only way to guarantee no mix-ups is to keep the samples separate until the very last step of the process.

Technical Summary

1. Problem Statement

Oxford Nanopore Technologies (ONT) multiplex sequencing allows for cost-effective high-throughput analysis by ligating unique native barcodes to individual DNA samples. However, the authors identified a significant, previously underestimated issue: barcode crosstalk (misassignment of reads).

• The Phenomenon: Reads from one sample are incorrectly assigned to another sample's barcode.
• The Cause: The authors hypothesize that residual native barcodes remain after the initial ligation and washing steps. During the subsequent pooling and adapter ligation steps, these free barcodes mis-ligate to DNA fragments from other samples in the pool.
• Impact: This leads to false positives, distorted quantitative results, and reduced reproducibility, particularly in samples with low DNA input concentrations (low biomass), which are common in clinical metagenomics (e.g., stool, wound swabs, cell-free DNA).

2. Methodology

To systematically quantify and mitigate this effect, the researchers designed an experiment using four bacterial type strains (E. coli, P. mirabilis, A. baumannii, S. epidermidis) across a wide range of DNA input concentrations (from 400 ng down to 0.4 ng).

They compared three distinct library preparation protocols on a PromethION device:

• Protocol A (Standard/Old ONT): The standard ONT Native Barcoding protocol (version valid until July 2025).
  • Workflow: Barcode ligation $\rightarrow$ Ethanol wash $\rightarrow$ Pooling of samples $\rightarrow$ Sequencing adapter ligation.
• Protocol B (Updated ONT): The new ONT protocol (released July 2025).
  • Modification: Replaces the Ethanol wash with Short Fragment Buffer (SFB) after barcode ligation to better remove residual barcodes.
  • Workflow: Barcode ligation $\rightarrow$ SFB wash $\rightarrow$ Pooling $\rightarrow$ Adapter ligation.
• Protocol C (In-house/Proposed): A modified workflow developed by the authors.
  • Modification: Samples are not pooled after barcode ligation. Instead, they are kept separate through the adapter ligation step and only pooled after the sequencing adapters are attached.
  • Workflow: Barcode ligation $\rightarrow$ Ethanol wash $\rightarrow$ Adapter ligation (individual) $\rightarrow$ Pooling $\rightarrow$ Sequencing.

Data Analysis:

• Reads were base-called (Dorado) and mapped to reference genomes using minimap2.
• Strict filtering criteria were applied (unique mapping, read length >2kb, >70% identity, mapping quality 60, and exact barcode matching) to ensure high-confidence quantification of misassigned reads.

3. Key Results

Quantification of Crosstalk (Protocol A)

• Severe Impact on Low Input: Protocol A showed a strong correlation between low DNA input and high crosstalk.
• Misassignment Rates:
  • High input (400 ng): ~0.0001% incorrect reads.
  • Low input (0.4 ng): Up to 2.4% of reads were misassigned.
• Directionality: Crosstalk primarily flowed from high-concentration samples to low-concentration samples. High-input samples "donated" residual barcodes that ligated to the low-input DNA.

Efficacy of Protocol B (Updated ONT)

• Replacing ethanol with SFB significantly reduced crosstalk.
• Misassignment rates dropped to <0.01% across all dilutions.
• Limitation: While a vast improvement, low-level crosstalk was still detectable, suggesting SFB does not completely eliminate residual free barcodes.

Efficacy of Protocol C (In-house)

• Near-Total Elimination: Protocol C virtually eliminated barcode crosstalk.
• Data: Across all experiments and dilution levels, only 7 incorrect reads were detected in total (average error rate < $3.9 \times 10^{-5}$).
• Trade-off: Protocol C resulted in lower total read counts per sample due to processing inefficiencies and required significantly more reagents and labor (processing samples individually rather than in a pool).

Read Loss Analysis

• Protocol A and B also caused "lost" reads (correct DNA assigned to the wrong barcode).
• Surprisingly, Protocol B showed a higher percentage of lost reads compared to Protocol C, suggesting that while SFB reduces gain of foreign reads, it may not fully prevent the loss of native reads to incorrect barcodes in the pool.

4. Key Contributions

1. Quantification: Provided the first rigorous quantification of ONT barcode crosstalk as a function of DNA input concentration, demonstrating that error rates scale inversely with input DNA.
2. Mechanism Identification: Confirmed that crosstalk occurs due to residual free barcodes ligating to non-barcode DNA during the pooling/adapter ligation phase.
3. Protocol Evaluation: Validated that the manufacturer's update (Protocol B) is effective but incomplete.
4. Gold Standard Solution: Proposed and validated Protocol C (pooling after adapter ligation) as the only method to virtually abrogate crosstalk, establishing a new benchmark for high-accuracy, low-biomass sequencing.

5. Significance and Recommendations

• Clinical Relevance: For low-biomass samples (e.g., microbiome from spinal fluid, urine, or cell-free DNA), the standard protocol (A) produces unacceptable error rates that could lead to false pathogen detection.
• Protocol Selection:
  • Standard Use: The updated ONT protocol (B) is a cost-effective improvement and suitable for most applications.
  • High-Accuracy/Low-Input Use: For critical applications requiring maximum accuracy with low DNA input, Protocol C is strongly recommended despite higher costs and labor.
• Data Review: The authors urge researchers to critically review historical ONT multiplex data generated with older protocols, especially those involving low-input samples, as results may be compromised by crosstalk.

Conclusion: Barcode crosstalk is a critical, concentration-dependent artifact in ONT sequencing. While manufacturer updates mitigate the issue, a workflow modification that delays sample pooling until after adapter ligation is the only method to ensure near-perfect sample demultiplexing.