Long-read sequencing of Mycobacterial tuberculosis is comparable to short-read sequencing for antimicrobial resistance prediction and epidemiological studies.

This study demonstrates that Oxford Nanopore long-read sequencing achieves comparable accuracy to Illumina short-read sequencing for *Mycobacterium tuberculosis* lineage assignment, antimicrobial resistance prediction, and outbreak detection, thereby validating the aggregation of data from both platforms for large-scale public health surveillance.

The Gist

Imagine you are trying to solve a massive, complex puzzle to stop a dangerous thief (in this case, the Mycobacterium tuberculosis bacteria). For the last decade, the gold standard for solving this puzzle has been using a high-speed, ultra-precise scanner called Illumina (short-read sequencing). It's like using a super-powered microscope that takes thousands of tiny, perfect snapshots of the bacteria's DNA. It's incredibly accurate, but it's also slow, bulky, and requires a fancy laboratory.

Recently, a new tool has arrived on the scene: Oxford Nanopore Technologies (ONT) (long-read sequencing). Think of this as a portable, handheld scanner that can read long, continuous strips of the DNA puzzle in real-time. It's fast, it can be carried in a backpack, and it's much simpler to use. But the big question was: Is this new, portable scanner good enough to replace the gold standard, or will it miss crucial clues?

The Great Race: 508 Samples

To find out, the researchers organized a massive race. They took 508 samples of the bacteria from South Africa and Vietnam (including some tricky, drug-resistant strains and local outbreaks) and ran them through both scanners simultaneously.

They treated the old Illumina scanner as the "referee" and checked if the new ONT scanner could tell the same story.

The Results: A Tie for the Win

Here is what they discovered, broken down into simple terms:

1. Identifying the "Family Tree" (Lineages)
Just like humans have different family backgrounds, bacteria have different lineages. The two scanners agreed on the family tree 96% of the time. The few times they disagreed, it was because the sample was a "mixed bag" (containing two different types of bacteria), and one scanner spotted the mix while the other didn't. This is a minor glitch, not a failure.

2. Checking for Drug Resistance (The "Villain" Check)
The most important job is to see if the bacteria is resistant to medicine. If the scanner says "Safe" but the bacteria is actually "Resistant," that's a dangerous mistake.

• The new ONT scanner made very few dangerous errors.
• Out of 15 different drugs, the error rate was so low (around 1%) that it passed the strict safety tests set by international regulators.
• Analogy: Imagine a security guard checking 15 different types of weapons. The new guard missed only 1 or 2 out of every 100 checks, which is good enough for the job.

3. Tracking the Outbreaks (The "Connect the Dots")
To stop an outbreak, scientists need to see how closely related different bacteria samples are by counting tiny differences (SNPs) in their DNA.

• When comparing the two scanners on the same sample, the difference was almost zero.
• For nearly 98% of the samples, the two scanners found the exact same number of differences (or just one tiny difference).
• Analogy: If you and a friend both try to count the steps between two houses, you would both get the exact same number. This means the new scanner can track outbreaks just as accurately as the old one.

The Big Takeaway

The researchers concluded that the new, portable scanner is now good enough to be trusted alongside the old, expensive one.

Why does this matter?

• Before: You had to send samples to a big, central lab with the expensive machine, waiting weeks for results.
• Now: You can use the portable scanner in a remote clinic, get results in hours, and still be confident the data is accurate.

The Final Verdict:
Because the two technologies now agree so perfectly, public health agencies can finally mix their data. They can combine the massive historical data from the old scanners with the new, fast data from the portable scanners. This allows the world to fight tuberculosis faster, smarter, and more effectively, using the speed of the new tool without sacrificing the accuracy we rely on.

Technical Summary

Technical Summary: Long-read vs. Short-read Sequencing for Mycobacterium tuberculosis

1. Problem Statement

For the past decade, short-read sequencing technologies (primarily Illumina) have served as the de facto gold standard for Mycobacterium tuberculosis complex (MTBC) genomic analysis, specifically for outbreak investigation and genomic drug susceptibility testing (gDST). While effective, these methods lack portability and speed. Long-read sequencing, particularly Oxford Nanopore Technologies (ONT), offers significant advantages in speed, simplicity, and portability, which are critical for real-time public health responses. However, there has been a lack of large-scale, rigorous validation comparing the clinical and epidemiological accuracy of modern long-read platforms against the established short-read standard for MTBC. This study addresses the question of whether ONT can reliably replace or complement Illumina for critical outputs like lineage assignment, drug resistance prediction, and cluster detection.

2. Methodology

The study conducted the largest comparison to date between Illumina and ONT sequencing technologies for MTBC samples.

• Dataset: A total of 508 samples were analyzed, originating from South Africa and Vietnam. The cohort was intentionally over-selected for drug resistance and included various local outbreaks and diverse lineages.
• Sequencing Protocol:
  • All samples had prior Illumina sequencing data.
  • Samples with Illumina read depth $\ge$50 were selected for ONT sequencing using GridION or PromethION platforms.
  • ONT sequencing utilized R10.4 flowcells, updated basecalling models, and deep-learning variant calling to maximize accuracy.
• Bioinformatics: Processing was performed on a modified cloud platform featuring:
  • Reference-based variant calling.
  • Catalogue-based gDST for 15 drugs.
  • SNP counting for outbreak detection and lineage assignment.
  • Illumina data served as the reference standard for comparison.
• Analysis Metrics: The study compared lineage assignments, gDST predictions (calculating Very Major Errors [VME], Major Errors [ME], and unclassified rates), and SNP distances between platforms.

3. Key Contributions

• Scale: This represents the largest head-to-head comparison of Illumina and ONT for MTBC to date (508 samples).
• Technological Advancement: It validates the use of the latest ONT hardware (R10.4 flowcells) and bioinformatics pipelines (deep-learning variant calling) specifically for TB, moving beyond previous iterations of the technology.
• Regulatory Benchmarking: The study explicitly evaluates performance against Clinical and Laboratory Standards Institute (CLSI) thresholds for antimicrobial resistance testing.
• Data Aggregation Feasibility: It provides the empirical evidence necessary to justify the aggregation of data from different sequencing platforms for large-scale public health surveillance.

4. Results

• Sample Retention: Of the 508 samples, 425 (83.7%) met the sufficient read depth criteria for direct comparison.
• Lineage Assignment:
  • Lineage assignments were identical for 95.8% (407/425) of samples.
  • Discordances were primarily attributed to mixed lineage identification by one technology.
  • Evidence of non-tuberculous mycobacterium (NTM) subpopulations was detected in nine samples.
• Drug Susceptibility Testing (gDST):
  • Using Illumina as the reference, the ONT Very Major Error (VME) rate (false susceptibility) across 15 drugs was 1.0% (95% CI: 0.6–1.5%).
  • The Major Error (ME) rate (false resistance) was 1.7% (95% CI: 1.3–2.2%).
  • The unclassified rate was 6.9%.
  • These error rates fall below the thresholds specified by the CLSI.
  • For individual drugs, VME and ME point estimates were $\le$3% in 29/30 cases and $\le$1.5% in 25/30 cases.
• Epidemiological Concordance (SNP Analysis):
  • After filtering for mixed samples (leaving 382 isolates) and masking the reference genome, the mean SNP distance between platforms was 0.13 (median of zero).
  • 98.4% of samples showed a difference of $\le$1 SNP between the two platforms.
  • This high concordance resulted in negligible differences in the identification of the 43 putative clusters among 172 isolates.

5. Significance

The study concludes that the differences between long-read (ONT) and short-read (Illumina) sequencing for key clinical outputs in MTBC are now within the tolerances set by regulatory agencies.

• Clinical Impact: The technology has reached a maturity threshold where long-read sequencing can be reliably used for genomic drug susceptibility testing and outbreak analysis.
• Public Health Utility: The findings enable the aggregation of sequencing data from both long- and short-read platforms. This allows national and international public health agencies to conduct large-scale, unified analyses without being restricted to a single technology.
• Operational Advantage: Public health systems can now leverage the portability and speed of ONT for rapid, on-site sequencing while maintaining data compatibility with historical Illumina datasets, facilitating more responsive and scalable tuberculosis control strategies.