The research team, led by Xing-Xing Shen of Zhejiang University together with co-first authors Xianfeng Mi and Guo-Zheng Ou and colleague Yixiao Zhu, leveraged a large-scale dataset originally assembled by Tao and colleagues comprising 472 insect species spanning 19 orders. From each species they extracted amino acid sequences of 13 mitochondrial protein-coding genes and 1,367 single-copy nuclear protein-coding genes. Using concatenation-based maximum likelihood inference with the IQ-TREE software and 1,000 ultrafast bootstrap replicates, they reconstructed two genome-scale phylogenies, one from the mitochondrial data under the best-fitting mtInv+F+G4 model and one from the nuclear data under the LG+G4 model. Both trees were exceptionally well supported: 98.30 percent of nodes in the nuclear phylogeny and 88.11 percent of nodes in the mitochondrial phylogeny carried bootstrap values of 90 percent or higher, and within the five largest insect orders, Coleoptera, Diptera, Hemiptera, Hymenoptera, and Lepidoptera, highly supported nodes remained above 95 percent in nuclear trees and above 80 percent in mitochondrial trees.

That high support made the conflict all the more striking. When the researchers compared the two topologies using the normalized Robinson-Foulds distance, a metric ranging from 0, indicating identical trees, to 1, indicating maximal disagreement, they obtained a value of 0.306 for the full dataset, a considerable level of topological discordance that could not be dismissed as an artifact of weak phylogenetic signal. The degree of conflict also varied substantially among the major orders. Lepidoptera, the order containing butterflies and moths, showed the greatest incongruence with an nRF distance of 0.446, while Hymenoptera, the bees, wasps, and ants, showed the least at 0.158. Because both trees were statistically robust, the discordance had to reflect genuine differences in the evolutionary histories or compositional properties of the two genomes rather than simple failure to resolve branching order.

One obvious suspect was model choice. Mitochondrial genomes evolve faster than nuclear genomes and display strong compositional heterogeneity, so critics might argue that standard models simply fail to capture the true substitution process in mitochondrial sequences, generating misleading trees. To test this, the team rebuilt the mitochondrial phylogeny from concatenated amino acid sequences using a site-heterogeneous mixture model, mtInv+F+G4+C60, which allows substitution patterns to vary across sites in far more flexible ways. The result was unambiguous: the more complex model did not eliminate the mito-nuclear discordance. This finding aligns with a growing body of work suggesting that increasing evolutionary model complexity improves fit to the data but does not necessarily resolve gene tree conflict, meaning the source of the disagreement had to lie elsewhere.

The researchers then cast a wide net, systematically quantifying nine gene properties for every mitochondrial and nuclear gene: alignment length, GC content, amino acid substitution saturation, effective number of amino acids, proportion of constant sites, proportion of parsimony-informative sites, external branch length, average bootstrap support value, and treeness, defined as the proportion of internal branch lengths over all branch lengths. Their central analytical innovation was a property-matching strategy. For each property, they selected 13 nuclear genes whose values closely matched those of the 13 mitochondrial genes, for example nuclear genes whose alignment lengths fell within plus or minus 5 percent of the corresponding mitochondrial genes, concatenated these mito-like nuclear genes, built a phylogeny, and measured its topological distance to the mitochondrial reference tree. This sampling and tree-building procedure was repeated 20 times for each property and compared against a background in which 13 nuclear genes were chosen at random, which yielded a median nRF of 0.334.

Against that background, only one property made a difference. When nuclear genes were selected to match the GC content of mitochondrial genes, the resulting trees became significantly more similar to the mitochondrial phylogeny, while matching on alignment length, saturation, amino acid diversity, or any of the other properties failed to reduce the discordance. The compositional gulf between the two genomes is dramatic: mitochondrial genes exhibited significantly lower GC content than nuclear genes, with mitochondrial codons showing a strong bias against codons ending in G or C and a marked preference for A/T-ending codons, whereas nuclear codons displayed a more balanced pattern across the first, second, and third codon positions. This compositional bias cascades into protein composition, since the mitochondrial genome favors amino acids encoded by low-GC codons such as leucine, isoleucine, phenylalanine, methionine, asparagine, and tyrosine, while the nuclear genome favors amino acids encoded by high-GC codons such as alanine, arginine, glutamate, and aspartate.

The GC effect extends beyond the boundary between the two genomes and into the structure of the nuclear genome itself. When the researchers compared pairwise topological differences among mitochondrial genes, they found that mitochondrial genes were more topologically similar to one another than nuclear genes were to each other, consistent with their shared maternal inheritance. Yet even within the mitochondrial genome, low-GC genes produced trees more similar to each other than high-GC genes did. To probe the nuclear side, the team divided 1,367 nuclear genes into three non-overlapping groups by GC content, a low-GC group at 38 to 40 percent, a medium-GC group at 42 to 44 percent, and a high-GC group at 46 to 48 percent, with each group spanning exactly a 2 percent range to ensure comparable variation. Pairwise comparisons of gene trees within each group revealed a clear trend: topological differences increased steadily with GC content, meaning low-GC nuclear genes are more topologically consistent with one another than high-GC nuclear genes are.

Intriguingly, GC content also tracks functional differentiation within the nuclear genome. Using Gene Ontology analysis, in which detailed GO terms were consolidated into 18 broader biological process categories, the researchers found that high-GC nuclear genes were disproportionately involved in complex biological processes including biological regulation, developmental processes, response to stimulus, multicellular organismal processes, growth, and homeostatic processes, while low-GC nuclear genes represented the greatest proportion of genes involved in metabolic processes. This functional stratification suggests that GC content is not merely a technical nuisance for phylogenetic inference but a biologically meaningful axis of genome organization, one that correlates with gene expression patterns and evolutionary dynamics documented in prior studies of codon usage bias and compositional evolution across insect genomes.

What mechanism might link nucleotide composition to tree topology? The authors point to GC-biased gene conversion, a process in which recombination preferentially fixes G and C alleles over A and T during DNA repair, as a plausible driver. GC-rich genomic regions are thought to experience higher recombination rates, and biased conversion can distort substitution patterns in ways that mislead tree-building algorithms, potentially generating greater topological discordance among high-GC genes. The study’s conclusions are notably conservative in what they rule out: gene alignment length, inheritance pattern, and evolutionary model complexity all failed to explain the observed conflicts, overturning several common assumptions in the field. Because mitochondrial genomes are shorter than nuclear genomes, differences in alignment length have long been considered a contributing factor, yet matching on length had no effect, and the persistence of discordance among mitochondrial genes grouped by GC content shows that maternal inheritance alone cannot guarantee topological agreement.

The authors are careful to acknowledge the limits of their analysis. Multiple biological processes are known to generate nuclear phylogenomic incongruence, including incomplete lineage sorting, introgression, hybridization, sex-biased dispersal, and horizontal gene transfer, and future work integrating these factors will be essential to test whether they also contribute to mito-nuclear discordance. The team emphasizes that their aim was to elucidate the cause of discordance rather than to declare which genome tells the truer evolutionary story. Even so, the practical implications are immediate for anyone reconstructing insect phylogenies: sampling nuclear genes whose GC content resembles that of mitochondrial genes, so-called mito-like nuclear genes, can measurably reduce topological conflict, and researchers interpreting mitochondrial barcodes or mitogenomic trees should now weigh compositional bias as a first-order consideration. By bridging molecular composition and evolutionary inference, the study transforms GC content from a footnote in methods sections into a central character in one of phylogenetics’ most persistent mysteries.

Subject of Research: Mito-nuclear phylogenetic discordance and the role of GC content in insect genome evolution

Article Title: Dissecting discordance of mitochondrial and nuclear phylogenetic trees in insects

Article References: Mi, X., Ou, G.-Z., Zhu, Y., & Shen, X.-X. (2025). Dissecting discordance of mitochondrial and nuclear phylogenetic trees in insects. Crop Health, 3(1), Article 23. https://doi.org/10.1007/s44297-025-00062-3

Image Credits: AI Generated

DOI: 10.1007/s44297-025-00062-3

Keywords: phylogenetics, mito-nuclear discordance, insects, GC content, mitochondrial genome, nuclear genome, insect genomics, GC-biased gene conversion, Gene Ontology, Robinson-Foulds distance, codon usage bias, evolutionary models