TL;DR
This paper introduces mGEMS, a novel computational pipeline that enables bacterial genomic epidemiology from mixed-culture samples. By leveraging a new read-binning algorithm and a scalable pseudoaligner, mGEMS significantly reduces the cost and effort of genomic analysis, making it more accessible for public health applications.
What Problem Does This Paper Address?
Genomic epidemiology is a powerful tool for tracking the spread of bacterial pathogens, but it traditionally relies on sequencing pure bacterial isolates. This involves laborious steps like culturing, picking individual colonies, and then performing DNA extraction, library preparation, and sequencing for each. This process is costly and time-consuming, especially when dealing with samples that might contain multiple closely related strains (mixed infections). This bottleneck limits the widespread adoption of genomic epidemiology, particularly in resource-limited settings. The paper aims to overcome this by enabling accurate genomic analysis directly from mixed samples obtained from plate sweeps, where multiple colonies are present.
Methods
The core of the mGEMS pipeline consists of two novel components:
- Themisto: A scalable and efficient pseudoaligner for short-read sequencing data. It uses a compressed de Bruijn graph approach, allowing for faster index construction and reduced memory usage compared to existing tools, making it suitable for large reference databases.
- mGEMS read binner: This is a new algorithm that probabilistically assigns sequencing reads to reference lineages. It leverages the output from pseudoalignment and abundance estimation (using the mSWEEP method) to assign reads to bins representing specific clonal groups. A key feature is its ability to assign a single read to multiple bins, which is crucial for handling strain-level variation.
The pipeline then uses these binned reads for downstream analyses like genome assembly (using Shovill), SNP calling, and phylogenetic reconstruction.
Datasets
The study utilized several datasets for benchmarking and validation:
- In vitro mixed samples: Newly generated datasets created by mixing known concentrations of three strains each of E. coli and E. faecalis. These serve as a benchmark with known ground truth for relative abundances.
- Synthetic mixture data: Sequencing reads from previously published genomic epidemiology studies were synthetically mixed to create datasets for E. coli ST131, E. faecalis, and S. aureus.
- Reference datasets: Collections of E. coli, E. faecalis, and S. aureus genome assemblies from national, global, and local sources were used as references for the mGEMS pipeline.
Key Insights and Ideas
- Democratizing genomic epidemiology: The primary insight is that by processing mixed samples, mGEMS significantly lowers the barrier to entry for genomic epidemiology, making it more affordable and accessible.
- Plate sweeps as a viable source: The paper demonstrates that sequencing DNA from a plate sweep (multiple colonies) can yield results comparable to traditional single-colony sequencing, opening up new possibilities for sample processing.
- Probabilistic binning for accuracy: The novel probabilistic read binning approach is key to handling the complexity of mixed samples and accurately assigning reads to their respective lineages, even for closely related strains.
- Pseudoalignment efficiency: Themisto's efficient pseudoalignment is crucial for scaling the pipeline to large reference datasets without prohibitive computational demands.
Key Results
- Comparable accuracy: The mGEMS pipeline achieved SNP calling and phylogenetic inference results comparable to those obtained from single-colony sequencing data, even for closely related strains.
- Accurate abundance estimation: The mSWEEP component of the pipeline provided accurate estimates of relative strain abundances in the mixed samples.
- Reduced cost and effort: By eliminating the need for multiple colony picks and individual library preparations, mGEMS significantly reduces the cost and labor involved in genomic epidemiology.
- Improved phylogenetic inference: In some cases, the mGEMS pipeline even showed improved bootstrap support values in phylogenetic trees compared to isolate sequencing data, potentially due to better handling of contamination.
- Computational efficiency: The mGEMS pipeline, including Themisto and the mGEMS binner, was shown to be computationally efficient, with a full pipeline run taking around 112 minutes on a standard laptop for a complex synthetic sample.
Limitations and Future Work
- Reference data dependency: The performance of mGEMS relies on the availability of comprehensive and well-defined reference datasets for the target species. In cases of highly unstable accessory genomes or poorly defined lineages (like the E. coli ST131 C2 strains), binning accuracy can be affected.
- Mobile elements and plasmids: The paper acknowledges that mobile elements and plasmids not tied to specific lineages might pose challenges, suggesting that their inclusion as separate entities in the reference collection could improve handling.
- Unculturable species: mGEMS is currently limited to species that can be cultured, as it relies on plate cultures and existing genomic reference data.
- Future directions: The authors suggest that their method could be further improved by better handling of mobile elements and plasmids. They also envision mGEMS supporting more complex models of transmission dynamics and enabling studies of high-diversity samples like those from the human gut. The potential for this approach to support diagnostics where multiple strains are relevant is also highlighted.
Summary updated