henhaobio
Sign in

Molecular Biology and EvolutionMicrobiologyPeer-reviewed journal article

Revealing reassortment in influenza A viruses with TreeSort

Alexey MarkiniD, Catherine A. Macken, Amy L. Vincent BakeriD, Tavis K. AndersoniD

PDFDOIPreprintCode

In-depth summary

Claude Sonnet 4.6 · xhigh effort · 8 Nov 2026 · from the full text of version 1 (15 Nov 2024) · CC0 1.0

TreeSort detects influenza A reassortment in thousands of genomes via clock-based tests; avian H7 reassorts most, and HA, NA, MP rarely reassort in human and H5N1.

Public influenza A whole-genome collections have grown large, but the authors state that no existing approach accurately infers reassortment on datasets with thousands or tens of thousands of genomes. TreeSort addresses this with a statistical test on a reference-segment phylogeny and was more accurate than the Bayesian CoalRe on simulated data.

Key results

  1. SuggestedOn simulated data, TreeSort's detection of reassortment events improved with sampling density: precision reached about 95% or more at 2,000+ strains, and recall rose with dataset size.

    Precision near or above 95% from 2,000 strains upward (median branch length below 2 substitutions per branch); recall approximately 80% at 500 strains vs 95% at 5,000 strains; 25 replicates per sampling regime. · 2.1 TreeSort demonstrated high accuracy in…; 4.5 Simulating influenza A evolution with…

  2. SuggestedGiven the true tree, TreeSort outperformed the Bayesian network method CoalRe on every simulated dataset. CoalRe degraded sharply at the largest size because its MCMC did not converge.

    CoalRe accuracy dropped to only 40% at 5,000 strains (F1 comparison, Supplemental Fig. S2); CoalRe v1.0.4 was run at n = 200, 400, 600, 800, 1000 and 5000 with the reassortment-rate prior mean set to the true value. · 2.1 TreeSort demonstrated high accuracy in…; 4.5 Simulating influenza A evolution with…

  3. SuggestedTreeSort estimates the reassortment rate (events per ancestral lineage per year) accurately in simulations once sampling is dense enough. Averaging over replicates is recommended.

    True rate recovered within 10% error at 1,500 or more strains and within 5% error at 3,500 or more strains; replicate-averaged estimate within 20% of the true rate under all sampling densities. · 2.1 TreeSort demonstrated high accuracy in…; 4.4 Estimating the reassortment rate

  4. ShownReassortment rates differed by host and subtype in 2010-2023 whole-genome data: avian highest (H7 top, then H5), swine above human. Human H3 vs H1 did not differ significantly.

    Avian H7 about 0.6 events per ancestral lineage per year, almost 2-fold above any other subtype; avian H5 0.3; swine H1/H3 average 0.1; human H1/H3 average 0.07; human H3 0.08 vs H1 0.07 (not significant). · 2.2. Avian IAV had a higher reassortment rate…; 3 Discussion

  5. ShownIn H5 clade 2.3.4.4b, the reassortment rate was 0.72 events per ancestral lineage per year after 2020 versus 0.44 before. In the 2020-2023 H5N1 clade, HA-PB2 was the most frequent pairing, with almost no HA-NA or HA-MP reassortment.

    0.72 events per ancestral lineage per year after 2020 vs 0.44 before 2020; high HA-NP reassortment in 2020-2023 H5N1 unlike human H1/H3; replaced PB2, PB1, PA and NP segments were more divergent post-2020. · 2.4 Changes in reassortment patterns of the H5…; 3 Discussion

Methods, data and reproducibility

For each query segment, a clock-based binomial test on Fitch parsimony distances flags reassortment on branches of a rooted reference-segment tree. The reassortment rate is then estimated under a Poisson point process on that tree.

  • Simulated two-segment influenza A datasets from the Coalescent with Reassortment model in BEAST 2 (1500 nt sequences, GTR+Gamma, parameters from swine IAV) · n = 25 replicates per sampling regime; 200 to 5000 strains
  • NCBI Virus influenza A gene sequences (accessed November 12, 2023) before filtering to whole genomes collected after 2010 · Influenza A virus (avian, swine, human) · n = n=971,784 sequences
  • Whole-genome strains of avian H3, H5, H6, H7, H9, swine H1, H3 and human H1, H3, capped per subtype; 80% subsampling · Influenza A virus (avian, swine, human) · n = up to 10,000 strains per subtype; 10 replicates
  • H5Nx clade 2.3.4.4b strains (Nextclade classification), split into background and the 2020-2023 H5N1 clade · Influenza A virus H5Nx (avian)
  • Published MCC trees of avian H3N8 and H7N7 segments (Worobey et al.) used to calibrate the clock-deviation parameter · Avian influenza A virus · n = 10 gene segments
Methods & tools
TreeSort, TreeTime v.0.11.3, DendroPy 4 (Fitch parsimony), MAFFT v7.475, FastTree v2.1.11, CoalRe v1.0.4, BEAST 2, SciPy SLSQP optimizer, Nextclade H5Nx classifier, NCBI Virus database
Software & environment
TreeTime v.0.11.3; MAFFT v7.475; FastTree v2.1.11; CoalRe v1.0.4; DendroPy 4; SciPy SLSQP optimization.
Reproducibility gaps
Custom Java code for projecting simulated reassortment events onto gene trees is described without a URL; no accessions or data links for the filtered strain sets are given.
Limitations (4)
  • Robustness to variation in substitution rates across segments, and to co-evolution among segments during reassortment, has not been assessed; relaxed-clock models are planned for future versions. · stated by the authors
  • TreeSort cannot identify the donor of reassorting segments. Multiple annotations on one branch cannot be resolved as one event or consecutive events, and consecutive events become more likely at lower sampling density. · stated by the authors
  • Accuracy was benchmarked only on simulated two-segment data with settings (p-value 0.001, deviation 1.1) that differ from the empirical analyses (p-value 0.0001, deviation 2.5); empirical rates have no ground truth. · noted in this summary
  • The fitness interpretation of low HA-NA-MP reassortment is a hypothesis without phenotypic testing. Host rate differences may also reflect uneven surveillance sampling. · noted in this summary

Read this if you need a scalable reassortment detector for large influenza surveillance datasets, or comparative reassortment rates across avian, swine and human hosts and H5N1 clade 2.3.4.4b.

Citations
20
FWCI
12.23
Field percentile
top 1%

Abstract

Reassortment among influenza A viruses (IAV) facilitates genomic evolution and has been associated with interspecies transmission and pandemics. We introduce a novel tool called TreeSort that accurately identifies recent and ancestral reassortment events on datasets with thousands of IAV whole genomes. The algorithm has immediate relevance to modern large-scale surveillance studies that generate thousands of whole genomes. TreeSort uses the phylogeny of a selected IAV segment as a reference and finds the branches on the phylogeny where reassortment has occurred with high probability by incorporating information from other gene segments. The tool reports the particular gene segments that were involved in reassortment and how different they are from prior gene pairings. Using TreeSort, we studied reassortment patterns of different IAV subtypes isolated in avian, swine, and human hosts. Avian IAV demonstrated more reassortment than human and swine IAV, with the avian H7 subtype displaying the most frequent reassortment. Reassortment in the swine and human H3 subtypes was more frequent than in the swine and human H1 subtypes, respectively. The highly pathogenic avian influenza H5N1 clade 2.3.4.4b had elevated reassortment rates in the 2020-2023 period; however, the surface protein-encoding genes (HA, NA, and MP) co-evolved together with almost no reassortment among these genes. We observed similar co-evolutionary patterns with very low rates of reassortment among the surface proteins for the human H1 and H3 lineages, suggesting that strong co-evolution and preferential pairings among surface proteins are a consequence of high viral fitness. Our algorithm enables real-time tracking of IAV reassortment within and across different hosts, can be applied to determine how specific genes evolve in response to reassortment events, and can identify novel viruses for pandemic risk assessment. TreeSort is available at https://github.com/flu-crew/TreeSort.

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISGong '21Taylor '22Müller '19Castelán-Sá '25Ganti '21Ganti '22
79% semantic matchEvolutionary Biology
X. Gong, M. Hu, W. Chen+6 more·bioRxiv·Sep 9, 2021