UltraStrain: An NGS-Based Ultra Sensitive Strain Typing Method for Salmonella enterica

Wenxian Yang, Lihong Huang, Chong Shi, Liansheng Wang, Rongshan YuFrontiers in Genetics 2019 · DOI

Abstract

In the last few years, advances in next-generation sequencing (NGS) technology for whole genome sequencing (WGS) of foodborne pathogens have provided drastic improvements in food pathogen outbreak surveillance. WGS of foodborne pathogen enables identification of pathogens from food or environmental samples, including difficult-to-detect pathogens in culture-negative infections. Compared to traditional low-resolution methods such as the pulsed-field gel electrophoresis (PFGE), WGS provides advantages to differentiate even closely related strains of the same species, thus enables rapid identification of food-source associated with pathogen outbreak events for a fast mitigation plan. In this paper, we present UltraStrain, which is a fast and ultra sensitive pathogen detection and strain typing method for Salmonella enterica (S. enterica) based on WGS data analysis. In the proposed method, a noise filtering step is first performed where the raw sequencing data are mapped to a synthetic species-specific reference genome generated from S. enterica specific marker sequences to avoid potential interference from closely related species for low spike samples. After that, a statistical learning based method is used to identify candidate strains, from a database of known S. enterica strains, that best explain the retained S. enterica specific reads. Finally, a refinement step is further performed by mapping all the reads before filtering onto the identified top candidate strains, and recalculating the probability of presence for each candidate strain. Experiment results using both synthetic and real sequencing data show that the proposed method is able to identify the correct S. enterica strains from low-spike samples, and outperforms several existing strain-typing methods in terms of sensitivity and accuracy.

Introduction

Identifying the exact strain of a pathogen during a foodborne illness outbreak is a "needle in a haystack" problem. While Next-Generation Sequencing (NGS) allows researchers to observe every piece of DNA in a sample, the target pathogen—such as Salmonella—is often obscured by DNA from the host (e.g., human or food tissue) and harmless background microbes.

Existing methods often struggle with "low-spike" samples where pathogen abundance is minimal. Traditional assembly-based methods require high coverage to function, while standard mapping methods can be misled by DNA sequences shared across different bacterial species. UltraStrain was developed to address these limitations, providing a high-resolution, sensitive pipeline for strain-level identification.

The UltraStrain Pipeline

UltraStrain utilizes a three-stage computational pipeline designed to maximize sensitivity and specificity:

  1. Denoising (Reads Filtering): To avoid the "ambiguous mapping" problem, UltraStrain first maps raw reads to a "synthetic reference genome." This reference is constructed from 403 unique, 1,000-bp marker sequences found exclusively in S. enterica. This step effectively filters out genetic noise from other species.
  2. Statistical Strain Identification: Filtered reads are compared against a database of known S. enterica strains using a statistical learning approach. The model calculates the probability of a strain's presence by considering "edit distance" (match quality) and the "uniqueness" of the read across different potential strains.
  3. Refinement: For extremely low-abundance samples, the tool performs a final validation by re-mapping all original reads against the top candidate strains to recalculate probabilities, ensuring the highest possible specificity.
Overview of the UltraStrain workflow, from unassembled reads to strain identification.

Experimental Evaluation

The performance of UltraStrain was evaluated using a combination of simulated and real-world datasets:

  • Synthetic Testing Sets: 65 datasets featuring 13 different S. enterica strains spiked into a background of 10 other bacterial species at varying abundance levels (from 10% down to 0.001%).
  • Coverage Benchmarks: 102 datasets designed to test how sequencing depth (from 10x to 500x) affects sensitivity at a fixed low abundance (0.01%).
  • Real-World Data: 24 test samples from the PrecisionFDA CFSAN Pathogen Detection Challenge, including both in silico contaminated and naturally contaminated cilantro samples.

Key Results and Insights

UltraStrain demonstrated superior sensitivity compared to existing tools like PathoScope2, Sigma, and Kraken2. In synthetic tests, UltraStrain correctly identified all 13 target strains at 0.1% abundance, whereas competitors identified fewer than half. Notably, UltraStrain was the only tool capable of identifying strains at 0.01% abundance.

Detail of the strain identification process and the resulting conditional distribution at the strain level.

Analysis revealed that UltraStrain’s sensitivity scales with sequencing depth. Unlike other tools where more data can introduce more noise, UltraStrain benefits from higher coverage as it increases the likelihood of capturing rare, species-specific markers. In the 2018 PrecisionFDA challenge, UltraStrain was a top performer, correctly identifying target strains in 8 out of 13 positive samples.

Conclusion

UltraStrain provides a robust and efficient solution for pathogen surveillance, particularly in cases where the target organism is present at very low levels. While the tool is highly sensitive, users are encouraged to apply high probability thresholds to manage the risk of false positives. Future work will focus on improving the identification of multiple strains within a single sample and expanding the reference database to include emerging pathogen variants.

Summary updated

Abstract

In the last few years, advances in next-generation sequencing (NGS) technology for whole genome sequencing (WGS) of foodborne pathogens have provided drastic improvements in food pathogen outbreak surveillance. WGS of foodborne pathogen enables identification of pathogens from food or environmental samples, including difficult-to-detect pathogens in culture-negative infections. Compared to traditional low-resolution methods such as the pulsed-field gel electrophoresis (PFGE), WGS provides advantages to differentiate even closely related strains of the same species, thus enables rapid identification of food-source associated with pathogen outbreak events for a fast mitigation plan. In this paper, we present UltraStrain, which is a fast and ultra sensitive pathogen detection and strain typing method for <i>Salmonella enterica</i> (<i>S. enterica</i>) based on WGS data analysis. In the proposed method, a noise filtering step is first performed where the raw sequencing data are mapped to a synthetic species-specific reference genome generated from <i>S. enterica</i> specific marker sequences to avoid potential interference from closely related species for low spike samples. After that, a statistical learning based method is used to identify candidate strains, from a database of known <i>S. enterica</i> strains, that best explain the retained <i>S. enterica</i> specific reads.Finally, a refinement step is further performed by mapping all the reads before filtering onto the identified top candidate strains, and recalculating the probability of presence for each candidate strain. Experiment results using both synthetic and real sequencing data show that the proposed method is able to identify the correct <i>S. enterica</i> strains from low-spike samples, and outperforms several existing strain-typing methods in terms of sensitivity and accuracy.
BibTeX
@inproceedings{yang2019ultrastrain,
  author = {Wenxian Yang and Lihong Huang and Chong Shi and Liansheng Wang and Rongshan Yu},
  title = {{UltraStrain: An NGS-Based Ultra Sensitive Strain Typing Method for Salmonella enterica}},
  booktitle = {Frontiers in Genetics},
  year = {2019},
  doi = {10.3389/fgene.2019.00276},
}
Figure