---
title: "Phylogenetic reconstruction error inflates reassortment counts in influenza A virus analyses"
id: "plos-one-21-assessing-the-impact-of-phylogenetic-reconstruction-error-on-genetic"
canonical_url: "https://medichelpline.com/clinical-feed/plos-one-21-assessing-the-impact-of-phylogenetic-reconstruction-error-on-genetic"
content_type: "clinical_feed_article"
specialty: "Infectious Disease"
source_name: "PLOS ONE (Medicine)"
source_url: "https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0360852"
published_at: "2026-10-22T14:00:00.000Z"
evidence_level: "Journal Feed"
license: "CC-BY-NC-4.0 / Informational Use"
---
# Phylogenetic reconstruction error inflates reassortment counts in influenza A virus analyses
## Provenance & Clinical Metadata
- **Canonical URL:** https://medichelpline.com/clinical-feed/plos-one-21-assessing-the-impact-of-phylogenetic-reconstruction-error-on-genetic
- **Specialty:** [Infectious Disease](https://medichelpline.com/clinical-feed/infectious-disease.md)
- **Primary Source:** PLOS ONE (Medicine)
- **Source URL:** [Original Journal Publication](https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0360852)
- **Published At:** 2026-10-22T14:00:00.000Z
- **Evidence Rating:** Journal Feed
## Executive GIST (TL;DR)
- The study evaluated how errors in phylogenetic tree reconstruction contribute to false inference of **reassortment** in segmented RNA viruses, using influenza A H5Nx genomes from avian hosts. - Investigators curated 11,795 complete avian H5Nx genomes and focused on segment 4 (hemagglutinin, **HA**) to build reference trees for simulations. - Subsets of HA sequences with varying expected reconstruction error were sampled; trees from these samples served as scaffolds for simulating evolution of other segments in the absence of reassortment. - Simulations varied sequence length and rates of evolution to mirror the other genomic segments; because no reassortment was simulated, any detected discordance reflects reconstruction error. - Simulated data from random HA samples produced about 32% as many subtree prune and regraft (**SPR**) events as were observed in real segment comparisons, indicating a substantial false-positive contribution. - When simulations used the most genetically divergent HA subset, the false-positive SPR rate fell to 4%, showing sampling composition affects error. - The team introduced SPR rearrangements into simulated trees to quantify false negatives and found a false-negative rate of 8.5%, which rose with the number of SPRs per tree. - Entanglement values and distance-to-SPR metrics (computed with Espalier) were used to compare real and simulated trees; the authors characterized inferred SPRs by subtree size and traversal distance to assess which rearrangements might be genuine. - The authors concluded that reassortment frequency can be substantially overestimated from phylogenetic discordance and recommended accounting for phylogenetic uncertainty and sampling design when analyzing IAV datasets for reassortment.
## Clinical Analysis & Structured Key Points
Assessing the impact of phylogenetic reconstruction error on genetic reassortment inference in segmented viruses | PLOS One Browse Subject Areas ? Click through the PLOS taxonomy to find articles in your field. For more information about PLOS Subject Areas, click here . Article Authors Metrics Comments Media Coverage Peer Review Reader Comments Figures Figures Abstract Recombination is an important process in the evolution of RNA viruses. For viruses with segmented genomes, recombination can occur by reassortment of entire genomic segments. This process has played a significant role in the adaptation and spread of influenza viruses with novel genomic combinations, i.e. , antigenic shifts. Reassortment is usually identified by discrepancies between the topologies of trees reconstructed from different genomic segments, e.g. , subtree prune and regraft (SPR) rearrangements. However, this phylogenetic discordance can also result from errors in tree reconstruction due to insufficient information. The objective of this study was to evaluate the contribution of phylogenetic reconstruction error to the inference of reassortment events from influenza A virus (IAV) genomes from avian hosts. We curated a dataset of 11,795 complete avian H5Nx IAV genomes and sampled subsets of HA sequences with different expected levels of reconstruction error. Next, we simulated molecular evolution along the trees reconstructed from these HA samples, varying rates of evolution and sequence lengths to correspond to the other genomic segments. Because these simulations had no reassortment, any phylogenetic discordance identified from the resulting trees must be caused by reconstruction error. We found that simulations from random samples yielded about 32% as many SPRs as obtained from real data. This false positive rate was reduced to 4% when simulating trees from the most genetically divergent subset of HA sequences. By modifying the simulated trees with SPR rearrangements, we observed a false negative rate of 8.5%, which increased with the number of SPRs per tree. Overall, we found that reassortment can be substantially overestimated from phylogenetic discordance, underscoring the need to account for phylogenetic uncertainty and sampling design when analyzing IAV datasets for reassortment. Citation: Castelán-Sánchez HG, Poon AFY (2026) Assessing the impact of phylogenetic reconstruction error on genetic reassortment inference in segmented viruses. PLoS One 21(10): e0360852. https://doi.org/10.1371/journal.pone.0360852 Editor: Gadissa Bedada Hundie, St Paul's Hospital Millennium Medical College, ETHIOPIA Received: March 10, 2026; Accepted: October 1, 2026; Published: October 22, 2026 Copyright: © 2026 Castelán-Sánchez, Poon. This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability: In this study, genome sequences of the influenza A virus (IAV) were accessed from two databases: NCBI GenBank and GISAID EpiFlu™. All accession numbers, sequence identifiers, and sequences used in this study are included in the Zenodo repository associated with this study. The repository also includes the entanglement results, the distance-to-SPR values computed using Espalier, the distance-to-siblings values, and scripts for this study. The data and additional analysis materials are included in Zenodo under the DOI: 10.5281/zenodo.21985671 . The original sequence records are available in NCBI GenBank, and GISAID EpiFlu™. Funding: The author(s) received no specific funding for this work. Competing interests: The authors have declared that no competing interests exist. Introduction Recombination is an important mechanism of genomic evolution in RNA viruses [ 1 ]. This process involves the exchange of genetic material between different genomic backgrounds, facilitated by coinfection of a host with two or more distinct viral lineages. Recombination uncouples the evolutionary fate of a mutation from the rest of the genome, and under certain conditions [ 2 ], it will increase the diversity of genotypes in the population [ 3 , 4 ]. Many RNA viruses have segmented genomes, including members of the Reoviridae and Orthomyxoviridae families [ 5 ]. There are also many multipartite viruses that package their genomic segments into separate virus particles [ 6 ]. In segmented RNA viruses, recombination can occur by reassortment, in which the genomic segments packaged into a virion are derived from different parental genomes. This wholesale exchange of segments can accelerate the acquisition of combinations of mutations that are required to overcome host barriers more rapidly than the progressive accumulation of individual mutations [ 7 ]. Influenza A virus (IAV; family Orthomyxoviridae , genus Alphainfluenza ) illustrates the potential impacts of genetic reassortment on the health of human and non-human animal populations. This virus has caused numerous severe epidemics and pandemics [ 8 ], several of which are associated with a reassortant virus [ 9 ]. IAV has a negative-sense RNA genome that consists of eight genomic segments that encode 10 major proteins and other accessory gene products [ 10 ]. When a host cell becomes co-infected with two different IAV genomes, there are 2 8 = 256 possible combinations of genome segments that can be packaged into a new virus particle. Only two of these combinations are derived from the same parental genome. With the simplifying assumption of independent assortment of genome segments during packaging, the probability of packaging a reassortant genome is . The co-circulation of multiple IAV subtypes and clades in domesticated animal host species provides abundant opportunities for co-infection and reassortment between divergent lineages [ 11 , 12 ]. Consequently, reassortment can generate novel chimeric IAV genomes, potentially enhancing the virus’s ability to transmit to new host species and establish productive infections [ 13 , 14 ]. The comparative analysis of IAV genome sequences plays an important role in measuring the frequency of reassortment and detecting the emergence of new reassortant viruses [ 15 , 16 ]. To this end, a number of different methods have been developed to identify reassortment events. In the absence of recombination, the evolutionary history of a sample population of genome sequences may be represented by a single phylogenetic tree. Reassortment uncouples the evolutionary histories of the affected genomic segments. Consequently, a common technique to detect reassortment is to reconstruct a phylogeny for each segment and then to identify discrepancies in the topology and/or labels among the resulting trees. Discordant trees are often visualized as tanglegrams [ 17 ], in which two trees sharing a common set of labels are drawn in opposite orientations, with subtrees rotated to maximize concordance of tip labels between the trees. Tanglegrams can be used to visually assess a small number of reassortment events. In addition, the optimization algorithm for untangling pairs of trees provides a rough measure of the extent of discordance that may be due to recombination, i.e. , entanglement. However, tanglegrams do not readily scale to the resolution of large numbers of reassortment events. Evolutionary events that produce discordant trees can be explicitly modeled using an ancestral recombination graph (ARG) [ 18 ]. Although there has been a recent resurgence of interest in ARGs [ 19 ], they remain too computationally complex for large-scale application. A compromise between these two extremes is to infer the frequency of reassortment events by the number of tree rearrangement operations required to convert one segment tree to another [ 20 ], which can be interpreted as an edit distance. Screening for tips that transition between different labeled subtrees in the respective phylogenies is an equivalent approach [ 21 , 22 ]. There are many edit distances available for this purpose, including the Robinson-Foulds distance [ 23 ], tree bisection and reconnection [ 24 ], nearest neighbor interchange [ 25 ], and subtree prune and regraft [ 26 ]. Among these, the SPR distance is frequently used in studies of lateral gene transfer and recombination [ 27 ], possibly because the SPR most closely resembles the effect of recombination on the topologies of trees on either side of the breakpoint. In this work, we performed a phylogenetic analysis starting from nearly 12,000 full-genome sequences of influenza A H5Nx viruses to identify putative rearrangement events from phylogenetic discordance. Our main objective is to evaluate the reliability of this approach, given the uncertainty in reconstructing accurate phylogenies under realistic conditions. To this end, we reconstructed phylogenies from samples of IAV H5Nx segment 4 sequences (encoding hemagglutinin, HA), and then used the resulting trees as the reference for simulating sequence evolution for the other segments. Because these sequences were simulated in the absence of reassortment, any inferred reassortments can be interpreted as false positives. Additionally, we modified the simulated trees with SPR rearrangements to measure false negative rates. We compared the entanglement values and SPR distances between the reference HA tree and the trees for all other segments in both real and simulated datasets to assess the specificity and sensitivity of these methods for detecting reassortment. Furthermore, we characterized the SPR rearrangements inferred from these data in terms of subtree size and the distance traversed in the tree, to determine whether certain types of SPRs are more likely to be genuine. Materials and methods Data collection We retrieved sets of influenza A virus genome segments in which the segment 4 sequence encoding hemagglutinin (HA) was classified as subtype H5, and for which samples were isolated from avian hosts, from the NCBI GenBank and GISAID databases (accessed 5 May 2024). These genomes were retrieved irrespective of subtype classification for segment 6 encoding neuraminidase (N1–N9), pathogenicity classification (HPAI/LPAI), or clade assignment; hence, they are denoted as H5Nx. The GISAID query returned 5,994–6,000 sequences per segment, and the GenBank query returned 6,157–6,170 sequences per segment. We combined these sequence data sets by genomic segment. Only a small proportion of sequences were exact duplicates across databases. For example, out of 12,167 neuraminidase (NA) sequences, there were 8,347 unique sequences, of which 1,417 (17%) had two or more instances — only 171 (12%) of these were represented in both databases, involving a total of 673 sequence records. This low amount of overlap is largely due to a transition from depositing sequences in GenBank to GISAID around 2020, coinciding with recent outbreaks of avian H5N1 IAV ( S1 Fig ). Sequence identity alone was insufficient to determine whether the sequences were derived from the same isolate. The dataset was not stratified by subtype, pathogenicity, host category, or geographic region, as our objective was to evaluate methodological sensitivity rather than lineage-specific epidemiological dynamics. Consequently, we merged these datasets by harmonizing the sequence labels into the World Health Organization [ 28 ] nomenclature format, including host species, country of sampling, isolate name, and year of sampling, and then filtering sequences with duplicate labels. This reduced the combined data to 11,795 sequences per segment. We generated alignments for each segment using MAFFT [ 29 ], v7.490;], removed duplicate sequences, and then manually inspected and cleaned each segment alignment with trimAl [ 30 ], v1.5.0]. We reconstructed a maximum likelihood phylogeny for the HA alignment using the standard generalized time-reversible (GTR) model with gamma-distributed rate variation and empirical base frequencies in IQ-TREE2 [ 31 ]. Next, we applied two subsampling strategies to reduce the number of sequences and modulate the amount of phylogenetic information. First, we progressively removed the shortest terminal branches from the HA phylogeny, using the prune function of the Biopython Phylo module, until a target tree length of 0.08 expected substitutions per nucleotide site was attained. This resulted in a subsample of the n = 213 most genetically divergent HA sequences. In addition, we generated five independent random samples of n = 213 HA sequences to represent more typical levels of phylogenetic information. We used the labels of the sampled HA sequences to retrieve the corresponding sequences for the other seven segments and reconstruct trees. These two sampling strategies provided a means of evaluating the effect of phylogenetic reconstruction error on levels of phylogenetic discordance. Data simulation To quantify the discordance among phylogenies reconstructed from sequences that evolved in the complete absence of reassortment, we simulated sequence alignments for each of the other segments along the tree inferred from the real HA alignment ( T HA ). Since every segment would evolve along the exact same tree topology, any reassortment event inferred from the simulated trees could be interpreted as a false positive. We proportionally rescaled branch lengths in T HA so that the total tree length matched the length of the tree reconstructed from the actual sequences for that segment. Our motivation for adjusting tree lengths in this manner was to accommodate substantial variation in evolutionary rates among segments, which could otherwise confound our assessment of false-positive rates. We recognize that the evolutionary patterns of individual influenza A genome segments can vary markedly in natural populations, reflecting subtype-specific diversification, distinct reassortment histories, and deep lineage structuring (for example, among NA subtypes) [ 32 ]. However, our focus here was purely methodological: to measure the level of discordance that would arise under perfect topological congruence, thereby separating the impact of phylogenetic reconstruction error from that of genuine biological reassortment. For each rescaled tree, we used the Python module Pyvolve [ 33 ] to simulate sequence evolution under the default Goldman-Yang codon substitution model with the dN/dS ratio set to 0.5 and sequence length matched to each genomic segment. Phylogenetic analysis We used two approaches to infer reassortment events. First, we used a tanglegram, a graphical comparison of two trees sharing a common set of tip labels. The trees are rotated and optionally re-rooted to minimize the discordance in the vertical location of labels, as quantified by the entanglement statistic. Our working assumption is that increasing amounts of reassortment will induce greater amounts of entanglement between trees from different segments. We used the step2side method, implemented in the Python module tanglegram [ 34 ], to disentangle the HA tree and the trees from every other segment. We further visualized the tanglegram using the cophylo function from the phytools package in R, which implements a closest-neighbor (NN) tanglegram approach by heuristically rotating the trees to minimize link crossings between matched tip labels [ 35 ]. Second, we used the Espalier library [ 36 ] in Python to efficiently calculate the approximate number of SPR (subtree prune and regraft) operations required to convert one phylogenetic tree into another. This number provides another means of estimating the number of reassortment events between phylogenetic trees. We obtained SPR distances between the HA tree and the trees for every other segment for the actual IAV genome sequence data, and we made the same comparisons for trees reconstructed from simulated data. We hypothesized that false positive reassortment events due to phylogenetic uncertainty would tend to involve SPR moves separated by shorter distances in the respective trees. To evaluate this hypothesis, we quantified the total path length separating the inferred graft points for each SPR operation. Let the tree reconstructed from real sequences for the i -th segment be denoted by . Moreover, let the tree reconstructed from simulated sequences that were evolved along the real HA tree T 4 be denoted by . For each subtree in the maximum agreement forest between trees T 4 and for , we first located in the reference (segment 4) tree T 4 by searching for the internal node with the fewest descendants that match all tip labels in . We call this node the root of the subtree j in T 4 and denote it as . Next, we located the internal node in T 4 with the same direct common ancestor as , which we refer to as its sibling node, . Switching to the other segment tree , we located the same subtree to identify by matching tip labels. Finally, we located the internal node in with the fewest descendants that match all tip labels associated with , which we denote as . The sibling distance is then defined as the total path length ( i.e. , sum of branch lengths) from to . Put another way, is the approximate location from which the SPR operation pruned the subtree, and is the location to which the subtree was grafted. A diagram summarizing this procedure is provided as S2 Fig . Following this method, we calculated the sibling distances for all SPR events between T 4 and both real ( ) and simulated ( ) trees for other segments. To measure the false negative rate (FNR), we simulated reassortment events on the HA tree with maximum divergence ( n = 213). Known reassortment events were introduced by directly applying a fixed number of SPR rearrangements to the reference tree uniformly at random using a custom Python script. We simulated 100 replicates under two different scenarios: one with a single SPR event per replicate, and another with 10 independent SPR events per replicate. To record ‘ground truth’, we tracked the tip labels for all descendants of the donor branch associated with each SPR event. Next, Espalier was used to build the maximum agreement forest (MAF) between the reference and reassortant trees for each replicate. For each MAF component, the descendant-tip set was compared with the donor branch in the previously recorded database. A reassortment event was only deemed to be successfully retrieved when all of the descendant taxa for a MAF were present in the reassortment event. When multiple SPR events occurred in the same replicate, each MAF component was matched one-to-one with a simulated donor branch, to ensure that each MAF was linked to a single donor branch. Event-level sensitivity (1-FNR) was defined as the percentage of donor branches that were fully recovered. Results Data collection We obtained a total of 11,795 H5Nx influenza A virus genome sequences from the NCBI GenBank and GISAID databases. These sequences represented a variety of avian hosts, with the most frequent annotations being ‘chicken’ ( n = 2,968), ‘duck’ ( n = 1,774), and ‘turkey’ ( n = 927). There was a total of 527 unique annotations for host organisms at varying taxonomic levels; for instance, mallard duck ( n = 47) and Muscovy duck ( n = 199) were both sub-classifications below ‘duck’. Similarly, there were 621 unique geolocation annotations, with the three most common being ‘Netherlands’ ( n = 845), ‘Minnesota’ ( n = 569), and ‘Vietnam’ ( n = 557). Many records were labeled by state or province rather than by country, obscuring the relative abundance of samples from China and the United States. We reconstructed a maximum-likelihood phylogeny of all HA (segment 4) sequences, and then we generated a subsample of n = 213 (1.8%) sequences that maximized the total tree length for a given number of sequences by progressively removing the shortest terminal branches from this phylogeny. Next, we extracted the correspon
## Related Clinical Research

- [Pharmacokinetic variability of baloxavir marboxil among 14 rare avian species](https://medichelpline.com/clinical-feed/plos-one-16-pharmacokinetic-diversity-of-baloxavir-marboxil-across-rare-avian-species.md)
- [High-dose versus standard-dose quadrivalent influenza vaccine immunogenicity in people living with](https://medichelpline.com/clinical-feed/pubmed-42753388.md) (DOI: 10.1016/j.vaccine.2026.129166)
- [Engineered Human MxA Achieves Broad, Super-Restrictor Activity Against Human-Adapted Influenza](https://medichelpline.com/clinical-feed/biorxiv-20-targeted-engineering-of-human-mxa-yields-a-broad-spectrum-super-restrictor-of.md)
- [Scrub Typhus Presenting with Typhoid-Like Bone Marrow Changes in an MDS Patient, China 2025](https://medichelpline.com/clinical-feed/cdc-emerging-infectious-diseases-journal-1-typhoid-fever-like-manifestation-of-scrub-typhus-in-patient-with.md)
- [Detection of Zoonotic Pteropine orthoreovirus in Rousettus amplexicaudatus Bats in Indonesia](https://medichelpline.com/clinical-feed/cdc-emerging-infectious-diseases-journal-0-identification-of-zoonotic-pteropine-orthoreovirus-in-rousettus-amplexicaudatus.md)

## Navigation
- [← Back to Infectious Disease Feed](https://medichelpline.com/clinical-feed/infectious-disease.md)
- [← All Clinical Specialties](https://medichelpline.com/clinical-feed.md)
## Medical & Regulatory Disclaimer

> [!CAUTION]
> MedicHelpline content is structured for research, educational, and professional discovery purposes. It does not constitute individual medical advice, clinical diagnosis, or treatment recommendations.
> Always verify dosing, contraindications, and regulatory alerts against official product labeling and primary regulatory sources before clinical decision-making.