Standalone nanopore sequencing for foodborne pathogen surveillance: a large-scale evaluation and quality control framework

This study demonstrates that standalone Oxford Nanopore sequencing of native DNA is sufficiently accurate for routine foodborne pathogen surveillance when supported by a new quality control framework, alpaqa, which identifies and mitigates systematic assembly errors caused by specific DNA modifications.

The Gist

Imagine you are a detective trying to solve a mystery: Who is the culprit behind a food poisoning outbreak?

In the past, detectives had to look at tiny, fragmented clues (short DNA reads) to guess the identity of the bacteria. It was like trying to identify a person by only seeing a few scattered hairs. You could tell it was a human, but you couldn't be sure which human it was, or if they were related to someone else.

Now, we have a new, super-powerful tool called Nanopore Sequencing (made by Oxford Nanopore Technologies). Think of this tool as a high-speed train that can read the entire DNA "book" of a bacteria in one go, from cover to cover. It's fast, cheap, and gives you the whole story, not just snippets.

However, there's a catch. Sometimes, the bacteria have "secret ink" on their pages (chemical modifications to their DNA) that confuses the train. The train might misread a word, thinking a "G" is a "T," which could make two identical bacteria look like strangers. This has made some health agencies hesitant to use this fast train alone, fearing they might miss the real culprit.

What did this study do?
The researchers acted like a massive quality control team. They took 294 different types of food-poisoning bacteria (like Salmonella, Listeria, and E. coli) and ran them through this "fast train" (Nanopore) to see if they could get a perfect reading without needing a second, slower "proofreader" (the old, expensive Illumina machines).

Here are the key takeaways, explained simply:

1. The Train is Mostly Perfect

The study found that 97 out of 100 times, the Nanopore train got the story exactly right. Even when they slowed the train down (using less data), it still got the job done. This means we can rely on this fast, standalone method for most food safety investigations. It's fast, cheap, and accurate enough to catch outbreaks before they spread.

2. The "Secret Ink" Problem

In about 3 out of 100 cases, the train got confused. Why? Because specific bacteria had special chemical "ink" (like phosphorothioation or methylation) on their DNA that the train's software didn't know how to read.

• The Analogy: Imagine reading a book where someone has written in invisible ink that looks like a different letter. The train sees a "G" but it's actually a modified "G," so it guesses wrong.
• The Result: In these rare cases, the bacteria looked like they had different DNA than they actually did. This could make the detectives think two related bacteria are unrelated, potentially breaking the chain of evidence in an outbreak.

3. The New "Spotter" Tool: Alpaqa

To fix this, the researchers invented a new digital tool called Alpaqa (pronounced like the animal, but it's a computer program).

• How it works: Alpaqa acts like a spell-checker that knows when the printer is jamming. It doesn't need a second opinion from another machine. It looks at the "confidence score" of every letter the train read.
• The Magic: When the train is confused by the "secret ink," it gets very nervous and assigns a very low confidence score to those letters. Alpaqa spots these "nervous" areas. If it sees too many nervous spots, it flags the report: "Hey, this DNA reading might be unreliable!"
• The Fix: Once flagged, the system can simply "black out" (mask) those confusing letters. It's better to have a few missing words than to have the wrong words. This ensures the final DNA profile is accurate enough to identify the bacteria correctly.

4. Why This Matters

This study is a green light for public health agencies.

• Before: You had to use the fast train and the slow, expensive proofreader to be safe.
• Now: You can just use the fast train. If Alpaqa says "all clear," you're good to go. If it says "warning," you know to double-check or ignore the confusing parts.

In a nutshell:
We can now use a fast, affordable, standalone machine to track foodborne diseases with near-perfect accuracy. We just need a smart "spotter" (Alpaqa) to tell us when the machine is confused by a bacteria's secret tricks, ensuring we never miss the real source of an outbreak. This means faster answers, safer food, and better protection for everyone.

Technical Summary

1. Problem Statement

Whole-genome sequencing (WGS) is the gold standard for foodborne pathogen surveillance and outbreak detection. While Oxford Nanopore Technologies (ONT) offers rapid, cost-effective, and complete genome assemblies (resolving repeats and plasmids that short-read Illumina sequencing misses), its adoption as a standalone platform for native DNA has been hindered by concerns regarding accuracy.

• The Core Issue: Epigenetic DNA modifications (e.g., methylation, phosphorothioation) can interfere with ONT basecalling, leading to systematic assembly errors.
• The Consequence: These errors can compromise core genome multi-locus sequence typing (cgMLST) profiles, potentially causing epidemiologically related isolates to be incorrectly excluded from outbreak clusters.
• The Gap: There was a lack of large-scale validation of ONT-only workflows across diverse foodborne pathogens and a missing framework to detect unreliable assemblies without requiring supplemental short-read (Illumina) data for cross-validation.

2. Methodology

The study employed a rigorous comparative approach involving 294 genetically diverse isolates representing ten major foodborne pathogens (Bacillus cereus group, Campylobacter, Cronobacter, Listeria, Salmonella, Shigella, Vibrio, and Yersinia).

• Sequencing Strategy:
  • ONT: Native DNA libraries prepared using the Rapid Barcoding Kit (SQK-RBK114) and sequenced on MinION/PromethION devices (R10.4.1 flow cells) to a minimum of 50x coverage. Basecalling was performed using the Dorado SUP@v5.2 model.
  • Illumina: Short-read data generated for the same isolates to serve as the reference standard.
• Assembly Pipeline (BOAP): A Nextflow pipeline was developed to process ONT data:
  1. Quality filtering and downsampling.
  2. De novo assembly using Flye.
  3. Consensus polishing using Dorado (or Medaka v2).
  4. Hybrid assemblies (ONT polished with Illumina reads) were generated to serve as the "truth" reference for accuracy comparison.
• Quality Control Tool (alpaqa): The authors developed alpaqa, a lightweight computational tool designed to identify unreliable assemblies using only ONT data.
  • Mechanism: It analyzes per-base Phred quality scores in the consensus assembly. It identifies "Low-Quality Bases" (LQBs) defined as scores between 1 and 5.
  • Algorithm: It calculates the density of LQBs per Megabase (Mbp) after masking repetitive/artifactual regions. It also performs k-mer enrichment analysis to detect sequence motifs associated with systematic errors.
• Validation:
  • cgMLST profiles were generated for both ONT-only and Hybrid assemblies using standardized schemes.
  • LC-MS/MS was used to chemically validate the presence of phosphorothioate modifications in error-prone Salmonella isolates.

3. Key Contributions

1. Large-Scale Validation: Demonstrated that standalone ONT sequencing of native DNA is highly accurate (>97%) for routine surveillance across a diverse panel of 294 isolates.
2. Identification of Error Sources: Pinpointed specific DNA modification systems responsible for systematic errors:
   • Phosphorothioation: A specific dndACDE system in Salmonella serovar Kentucky (targeting GGCC motifs).
   • Methylation: Specific Type II restriction-modification systems in Listeria monocytogenes (targeting GAAGAC and GCNGC motifs).
3. Development of alpaqa: Introduced a reference-free, short-read-free quality control tool that flags assemblies with systematic errors based on LQB density and motif enrichment.
4. Mitigation Strategy: Validated that masking low-quality bases (converting them to 'N') in flagged assemblies significantly improves cgMLST accuracy, albeit with a trade-off in genotyping resolution.

4. Key Results

• Overall Accuracy: At 50x coverage, 97.3% (286/294) of ONT-only assemblies produced identical or near-identical (≤3 allelic differences) cgMLST profiles compared to Illumina-polished references. 86.7% were fully identical.
• Depth Sensitivity: High accuracy was maintained even at reduced depths (30x), with 95.2% of assemblies remaining accurate. Increasing depth beyond 50x yielded marginal improvements.
• Error-Prone Isolates:
  • 4 Listeria monocytogenes: Showed 4–5 mismatches due to methylation. Increasing depth to 100x reduced errors to ≤3.
  • 4 Salmonella Kentucky: Showed severe errors (20–68 mismatches) due to phosphorothioation. Neither increased depth nor alternative polishing resolved these fully without masking.
  • 1 Salmonella Montevideo: Showed minor errors (2 mismatches) linked to 7-deazaguanine modifications.
• Performance of alpaqa:
  • A threshold of <5 LQBs/Mbp correctly identified 99.7% of accurate assemblies.
  • Error-prone Salmonella Kentucky isolates exhibited extremely high LQB densities (14–37 LQBs/Mbp).
  • The tool successfully identified systematic errors without needing a reference genome or short-read data.
• Impact of Masking: Masking bases with Q≤10 in error-prone assemblies reduced cgMLST mismatches significantly (e.g., reducing Salmonella mismatches from 20–68 down to 4–12), though it reduced the number of callable loci.

5. Significance and Conclusion

• Paradigm Shift: The study provides strong evidence that standalone ONT sequencing is sufficiently accurate for routine foodborne pathogen surveillance, removing the necessity for hybrid (short-read) polishing in the vast majority of cases.
• Operational Feasibility: By introducing alpaqa, the authors solved the "black box" problem of ONT-only workflows. Public health labs can now deploy ONT sequencing with an automated, built-in quality control step that flags problematic isolates for re-sequencing or targeted masking.
• Public Health Impact: This framework supports the transition of nanopore sequencing from a complementary technology to a standalone platform, enabling faster, cheaper, and more accessible genomic surveillance, particularly in resource-limited settings or decentralized outbreak scenarios.
• Future Outlook: While current chemistry handles most cases well, the study highlights that specific, rare epigenetic modifications still pose challenges. The integration of tools like alpaqa allows for the management of these edge cases without abandoning the native DNA workflow.