TL;DR
This paper investigates the reliability of pan-genome analysis, a key tool for understanding bacterial evolution, by highlighting how different computational choices can lead to wildly different results. They introduce panqc, a new software tool to correct for common errors like redundant gene annotations, improving the accuracy of pan-genome estimates, especially for closely related bacteria.
What Problem Does This Paper Address?
Understanding bacterial evolution and diversity relies heavily on analyzing the "pan-genome" – the complete set of genes found across all strains of a species. However, the way researchers define and compare these genes can significantly influence the results, leading to inconsistent estimates of a species' genetic repertoire. This is particularly problematic when trying to pinpoint subtle evolutionary changes, like gene gain or loss, especially in bacteria with limited genetic variation. The authors aim to quantify these sources of variability and develop a method to improve the accuracy of pan-genome analyses, focusing on Mycobacterium tuberculosis (Mtb) as a model organism due to its clonal nature and limited accessory genome.
Methods
The study employed a multi-faceted approach:
- Pan-genome Graph Construction: They used the
minigraphalgorithm to build a pan-genome graph representing structural variations (rearrangements, deletions, duplications) across a collection of Mtb genomes. This approach focuses on nucleotide-level differences. - Pan-genome Analysis Pipeline Comparison: They systematically evaluated three popular pan-genome analysis pipelines (Panaroo, Roary, and PPanGGolin) using different genome assembly types (hybrid vs. short-read) and annotation pipelines (Bakta vs. PGAP).
- Development of
panqc: A novel software tool,panqc, was developed. It addresses two main issues: - Annotation Discrepancy Detection: It identifies genes that are present at the nucleotide level but absent at the coding sequence (CDS) level due to annotation errors.
- Nucleotide Redundancy Collapse: It merges CDSs with highly similar nucleotide sequences, effectively collapsing redundant entries that arise from minor variations or annotation differences.
- Gene Gain/Loss Event Reconstruction: Ancestral state reconstruction was used to infer gene gain and loss events before and after applying
panqcto assess its impact on evolutionary interpretations.
Datasets
- Mycobacterium tuberculosis (Mtb): A curated dataset of 151 complete Mtb isolates was used. These isolates were sequenced using both short-read (Illumina) and long-read (Oxford Nanopore, PacBio) technologies, providing both hybrid and short-read de novo assemblies. This dataset spans global Mtb diversity across several lineages.
- Escherichia coli (E. coli): A dataset of 50 E. coli isolates was also analyzed to assess the applicability of
panqcto a more genetically diverse population. These were also assembled using a hybrid approach.
Key Insights and Ideas
- Structural Variation Dominates in Mtb: Contrary to expectations of novel gene acquisition, the majority of genetic variation in the Mtb pan-genome arises from rearrangements, deletions, and duplications of existing DNA sequences, rather than entirely new genes.
- Amino Acid-Based Comparisons Can Be Misleading: Pan-genome analyses relying solely on amino acid sequences can produce highly variable and inflated estimates of accessory genes, especially when assembly quality or annotation pipelines differ.
- Annotation Discrepancies are a Major Pitfall: Differences in how genes are annotated (e.g., identifying pseudogenes vs. functional CDSs) are a significant driver of inflated accessory genome sizes, particularly in closely related strains.
panqcCorrects for Redundancy: The developedpanqctool effectively reduces inflated pan-genome estimates by accounting for nucleotide redundancy caused by annotation discrepancies and highly similar sequences.- Genomic Diversity Matters: The impact of
panqcand the nature of pan-genome biases differ significantly between highly clonal populations like Mtb and more diverse populations like E. coli.
Key Results
- Mtb Structural Variation: The Mtb pan-genome graph revealed 536 core nodes (3.9 Mb) and 2602 SV nodes (1.3 Mb), with only 23% of SV nodes representing novel sequence content.
- Pipeline Variability: Pan-genome estimates for Mtb varied dramatically, from 220 to 2951 accessory genes, depending on the pipeline, assembly, and annotation choices.
panqcImpact on Mtb: Applyingpanqcreduced the Mtb accessory genome size by an average of 420 genes (44%). CDS annotation discrepancies accounted for an average of 49% of initially reported absent CDSs.panqcImpact on E. coli: For E. coli,panqchad a more modest effect, reducing the accessory genome by only 0.7% on average, as CDS annotation discrepancies were responsible for only 1.0% of absent genes.- Assembly Type Matters: Hybrid assemblies generally led to more consistent pan-genome estimates (larger core, smaller accessory) compared to short-read assemblies, which often suffer from incompleteness.
- Annotation Pipeline Impact: PGAP consistently produced smaller pan-genome estimates than Bakta, largely due to PGAP's higher frequency of annotating pseudogenes.
Limitations and Future Work
- Short-Read Assembly Limitations: The study highlights that short-read assemblies, while useful, can be incomplete and lead to inflated accessory genome predictions. Careful quality control of short-read assemblies is crucial.
- Mtb's Unique Nature: The findings regarding annotation discrepancies and redundancy are particularly pronounced in highly similar genomes like Mtb. Their impact may lessen in more diverse species, though
panqcstill offers value. - Future Directions: The authors envision
panqcbeing used in conjunction with other quality control tools for pan-genome analysis. They also suggest that future pan-genome analysis methods should aim to integrate both nucleotide and amino acid level information for a more comprehensive understanding of bacterial evolution.
Summary updated