Performance of shotgun metagenomic sequencing for detection of fungi and parasites across clinical sample types: a multicenter retrospective study.

This multicenter retrospective study demonstrates that clinical shotgun metagenomic sequencing achieves high diagnostic accuracy for detecting fungi and parasites across various sample types, validating its performance against standard methods and establishing optimized, sample-specific read-based thresholds for standardized clinical implementation.

The Gist

Imagine you are a detective trying to solve a mystery in a crowded, noisy city. Your job is to find specific, tiny criminals (fungi and parasites) hiding among millions of innocent bystanders (human cells) and other harmless people.

This paper is about testing a new, super-powerful detective tool called Shotgun Metagenomic Sequencing (SMg). Instead of looking for clues one by one, this tool takes a "snapshot" of every single piece of DNA in a sample (like blood, stool, or lung fluid) and tries to identify the criminals based on their genetic fingerprints.

Here is the breakdown of their investigation, explained simply:

1. The Challenge: The "Hard Shell" Problem

The detectives (scientists) found that catching these specific criminals (fungi and parasites) is much harder than catching bacteria or viruses.

• The Analogy: Bacteria are like soft clay; you can easily break them open to see what's inside. Fungi and parasites, however, are like armored tanks with thick, tough shells.
• The Result: It's difficult to get the DNA out of these "tanks." Even when you do, there are very few of them hiding in the sample, making them easy to miss in the noise of human DNA.

2. The Experiment: The Great Comparison

The team gathered 198 samples from four different hospitals. They had two ways of solving the mystery:

• The Old Way (Standard of Care): Using microscopes, growing cultures in petri dishes, and specific PCR tests. This is like looking for the criminal with a magnifying glass.
• The New Way (SMg): Using the high-tech DNA scanner. This is like using a satellite to scan the whole city at once.

They compared the results of the New Way against the Old Way to see if the new tool was accurate.

3. The Results: How Good Was the New Tool?

The results were actually quite impressive, but with some caveats:

• The Score: If you imagine a test score out of 100, the new tool got an 84/100 when identifying the type of criminal (the genus). It was very good at saying, "Hey, there's a Candida fungus here!"
• The Misses: The tool missed about 20% of the criminals that the old method found. This usually happened when the "criminal" was hiding in a very tough spot (like a tissue biopsy) or was extremely rare.
• The False Alarms: The tool didn't raise many false alarms. If the old method said "nothing here," the new tool agreed.

4. The "Noise" Problem: Different Cities, Different Rules

The researchers realized that the "city" (the sample type) matters a lot.

• Blood: This is like a quiet, empty street. It's easy to spot a criminal here. The new tool worked almost perfectly.
• Stool (Faeces): This is like a massive, chaotic music festival. There are millions of harmless bacteria and food particles everywhere. Finding a specific parasite here is like finding one specific person in a crowd of 100,000. The tool struggled a bit more here because of the "noise."
• Lungs (Respiratory Fluids): This is like a busy subway station. It's tricky, but the tool did reasonably well.

5. The Solution: Setting the "Volume" Threshold

The biggest question was: "How many DNA 'snippets' do we need to see before we say, 'Yes, the criminal is definitely here'?"

If you set the volume too low, you might hear a whisper and think it's a criminal (False Positive). If you set it too high, you might miss a real criminal who is just whispering (False Negative).

• The Discovery: They found that the "volume" needed to be different for every type of sample.
  • For blood, you only need to hear a tiny whisper (very low threshold).
  • For stool, you need to hear a slightly louder shout because there is so much background noise.
  • For tissue, you need a very loud shout because the sample is so small and hard to process.

They created a "rulebook" with specific thresholds for each sample type to make the tool as accurate as possible.

6. The Bottom Line

This study proves that the high-tech DNA scanner is a fantastic new tool for finding fungi and parasites, especially for tricky cases where the old methods fail.

However, it's not a magic wand that replaces the old methods yet. It works best when:

1. You know how to adjust the "sensitivity" based on what kind of sample you have (blood vs. stool).
2. You use it as a second opinion to double-check difficult cases.

In short: The new tool is like a high-tech metal detector. It's amazing at finding hidden metal (pathogens), but you have to know how to tune it depending on whether you are scanning a quiet beach (blood) or a muddy construction site (stool). If you tune it right, it saves lives by catching infections that would otherwise go undetected.

Technical Summary

1. Problem Statement

Clinical shotgun metagenomic next-generation sequencing (SMg) has revolutionized pathogen detection but faces specific challenges when applied to eukaryotic pathogens (fungi and parasites) compared to bacteria and viruses. Key obstacles include:

• Biological Barriers: Complex, robust cell walls of eukaryotes require specialized lysis, often leading to low nucleic acid yields.
• Low Abundance: Eukaryotic pathogens often exist at very low loads in clinical samples, even during active infection.
• Bioinformatic Complexity: High genomic similarity between eukaryotic pathogens and the human host increases the risk of misclassification (false positives/negatives).
• Lack of Standards: Existing guidelines for SMg are primarily designed for prokaryotes and viruses. There is a critical gap in standardized positivity thresholds and read-count interpretation criteria for fungi and parasites across diverse sample types.

2. Methodology

The study was a multicenter, retrospective analysis conducted across four French university hospitals (2018–2022).

• Cohort: 198 clinical samples from 187 patients, categorized into five types:
  • Blood ($n=37$)
  • Faeces ($n=63$)
  • Respiratory fluids ($n=54$)
  • Other biological fluids ($n=24$)
  • Tissue biopsies ($n=20$)
• Standard of Care (SoC) Reference: SoC included microscopy, culture, MALDI-ToF, and qPCR. A composite gold standard was created using SoC results plus orthogonal testing (confirmatory qPCR or direct exam) for SMg-only positives.
• Sequencing Protocol:
  • Extraction: Combined mechanical (glass beads), enzymatic (Proteinase K), and chemical lysis to break eukaryotic cell walls.
  • Libraries: Both DNA and RNA libraries were prepared.
  • Platform: Illumina NovaSeq 6000 (300-cycle paired-end).
  • Depth: Minimum of 10 million reads per library (DNA and RNA pooled).
• Bioinformatics:
  • Software: MetaMIC v2.2.1 (ISO 15189-accredited).
  • Processing: Adapter removal, quality filtering, host sequence exclusion (human/mammalian/plant), and comparison against environmental controls.
  • Normalization: Reads were normalized to Reads Per Million (RPM).
• Statistical Analysis:
  • Metrics: F1-score (harmonic mean of precision and recall) calculated at genus and species levels.
  • Thresholding: Receiver Operating Characteristic (ROC) analysis used to derive optimal RPM thresholds for each sample type against the composite gold standard.
  • Correlation: Spearman's correlation between RPM and qPCR Quantification Cycle (Cq) values.

3. Key Contributions

• First Large-Scale Evaluation: This is the first study to systematically evaluate SMg performance specifically for fungi and parasites across a wide variety of clinical sample types.
• Definition of RPM Thresholds: The study establishes specific, data-driven RPM thresholds for interpreting eukaryotic detection, moving away from arbitrary read counts.
• Sample-Type Specificity: It demonstrates that a "one-size-fits-all" threshold is suboptimal; performance and optimal cutoffs vary significantly by sample matrix (e.g., blood vs. faeces).
• Correlation with Load: It validates the inverse relationship between qPCR Cq values and SMg read counts, providing a quantitative link between traditional molecular load and metagenomic signal.

4. Key Results

• Overall Performance:
  • SoC identified microorganisms in 152/198 samples (76.8%).
  • SMg detected all 46 SoC-negative samples as negative (no false negatives in the negative cohort).
  • Genus-level F1-score: 0.84 overall.
  • Agreement: 78.6% of taxa were detected by both methods. SMg detected 28 unique taxa missed by SoC, while SoC detected 40 unique taxa missed by SMg.
• Performance by Sample Type:
  • Highest Accuracy: Blood and other biological fluids (F1 = 0.98 and 0.97, respectively).
  • Moderate Accuracy: Faeces (F1 = 0.82) and Respiratory fluids (F1 = 0.79).
  • Variability: Tissue biopsies showed high variability (median F1 = 1, but wide IQR), likely due to sampling constraints.
• Optimal RPM Thresholds (derived via ROC):
  • Faeces: 0.06 RPM (AUC 0.89)
  • Respiratory Fluids: 0.07 RPM (AUC 0.93; Sens 88.9%, Spec 90.7%)
  • Blood: 0.09 RPM (AUC 0.99)
  • Other Fluids: 0.19 RPM (AUC 0.94)
  • Biopsies: 0.57 RPM (AUC 0.89)
  • Global Threshold: A pragmatic uniform threshold of 0.1 RPM maintained high performance (AUC 0.92) across all types.
• Taxon-Specific Findings:
  • High detection consistency for Plasmodium, Cryptococcus, and Taenia (F1 = 1).
  • Lower performance for Giardia (F1 = 0.53) and Blastocystis (F1 = 0.79).
  • Rare taxa (e.g., Strongyloides, Trichuris) were frequently missed, often associated with low microbial loads (high Cq values).
• Load Correlation:
  • Cq values correlated inversely with RPM ($\rho = -0.66, p < 0.001$).
  • Critical Cutoff: No organism with a Cq < 28.6 was missed by SMg.
  • Read Depth Implication: To reliably detect low-abundance pathogens (requiring 10 pathogen-specific reads), the authors suggest increasing sequencing depth to **100 million reads** per sample, rather than the standard 10–30 million used for bacteria/viruses.

5. Significance and Conclusion

This study provides a critical framework for the clinical implementation of SMg for eukaryotic pathogens.

• Diagnostic Utility: SMg acts as a powerful second-line tool, complementing standard methods and identifying rare, unexpected, or travel-related pathogens that conventional assays miss.
• Standardization: The proposed RPM thresholds (specifically the 0.1 RPM pragmatic cutoff) offer a standardized approach to reduce ambiguity in clinical reporting.
• Future Directions: The authors emphasize that while SMg is highly sensitive, detection does not equal pathogenicity (colonization vs. infection). Results must be interpreted in the context of clinical, radiological, and biological data.
• Cost-Benefit: The study argues that increasing sequencing depth to ~100 million reads is cost-effective given declining sequencing costs, as it significantly improves the reliability of detecting low-abundance eukaryotic pathogens.

In summary, the paper validates SMg as a robust diagnostic tool for fungi and parasites but highlights the necessity of sample-type specific thresholds and increased sequencing depth to overcome the challenges of low pathogen load and complex host backgrounds.