The research team, led by Zhaobo Hu and Heath Blackmon of Texas A&M University with colleagues at the University of Houston and the University of Louisiana at Lafayette, compiled genome assemblies from the Ensembl database spanning mammals, ray-finned fishes, sauropsids (birds and reptiles), and other vertebrate lineages. For each species, they measured how evenly genes are spread across chromosomes using three complementary statistics: the coefficient of variation of gene density, a size-weighted Gini coefficient, and the coefficient of determination from a linear model relating chromosome size to gene count. Higher values on all three measures indicate greater homogeneity — a genome in which each chromosome carries roughly its proportional share of genes.

The logic behind the study rests on competing evolutionary forces. Insertions and deletions tend to accumulate in gene-poor, heterochromatic regions, where DNA repair is less efficient, while purifying selection quickly removes disruptive indels from gene-rich areas. A widespread deletion bias then preferentially shrinks noncoding DNA, pulling gene-poor chromosomes toward the genome-wide average. Chromosomal fissions and fusions similarly mix gene-rich and gene-poor blocks, gradually evening out differences between chromosomes. Repeat expansion works in the opposite direction: transposable elements and satellite repeats often accumulate in hotspots such as pericentromeric regions, inflating the length of some chromosomes far beyond what their gene content would predict and amplifying inequality.

To capture repeat dynamics, the team built assembly-specific repeat libraries with RepeatModeler and masked each genome with RepeatMasker, condensing the results into six categories: LTR retrotransposons, LINEs, SINEs, DNA transposons, unidentified repeats, and a residual category of repeats not annotated in every species, which includes simple repeats, satellites, and various RNA repeats. The scale of the underlying dataset deserves emphasis. The researchers drew on every vertebrate genome available in a single release of the Ensembl database, then applied strict quality filters: a species had to retain at least three sequences of at least ten megabases, and the retained sequences had to account for at least eighty percent of the total assembly. These thresholds were designed to weed out fragmented assemblies and unplaced scaffolds that would distort estimates of how genes are distributed among genuine chromosomes.

Out of the initial pool of 262 vertebrate genomes, measures of gene density homogeneity could be computed for 117 species, and complete repeat landscape data paired with homogeneity measures were available for 104. Where assemblies carried named chromosomes, those were used directly; otherwise the sixty longest qualifying sequences served as a proxy chromosome set, a compromise that still captures microchromosomes, the tiny gene-rich chromosomes characteristic of birds and other reptiles. Quality control extended beyond assembly size. Among the seventy-three assemblies with available completeness scores on NCBI, only seven fell below ninety-five percent on the BUSCO benchmark, and none dropped below ninety percent, suggesting that gene discovery was reliable across most of the sample. Two species, the pufferfish Tetraodon nigroviridis and the prairie vole Microtus ochrogaster, were nonetheless excluded despite having named chromosomes, because their assigned sequences covered too little of the total assembly to support trustworthy inference.

The choice of three distinct homogeneity metrics was deliberate. The coefficient of variation captures dispersion in gene density regardless of any relationship with chromosome size. The size-weighted Gini coefficient, borrowed from economics, quantifies inequality in how gene density is distributed among chromosomes, with values near one indicating that a handful of chromosomes dominate the variation. The coefficient of determination from a regression of gene count on chromosome size asks a subtly different question: how tightly does gene count track chromosome length? A genome can score well on one metric and poorly on another, and the sauropsids illustrate this vividly. Their genomes show the highest median Gini coefficient (0.22) and coefficient of variation (0.60) of any major clade examined, yet also the highest linearity between chromosome size and gene count (median R2 of 0.94). Gene-dense microchromosomes, which are dramatic outliers in density but contribute little to the size-gene relationship because of their small size, likely explain this apparent contradiction. Ray-finned fishes showed the most even gene density distributions, with the lowest median values for both the coefficient of variation (0.13) and the Gini coefficient (0.07), while mammals fell intermediate between fishes and sauropsids. A standard analysis of variance found strong differences among clades, but a phylogenetically informed version found no evidence for among-group differences after accounting for their evolutionary relationships.

The statistical machinery behind the study was unusually thorough. Rather than testing a handful of preselected predictors, the team generated every valid combination of eighteen candidate variables, derived from six repeat classes each characterized by abundance, sequence divergence, and their interaction, producing up to 15,625 models per dataset. Models were ranked by the corrected Akaike Information Criterion, and the smallest set of top models whose cumulative weight reached ninety-five percent formed a confidence set from which coefficients were averaged. A predictor was treated as supported only if its relative importance exceeded one half and its averaged confidence interval excluded zero. The influence of individual species was assessed using Cook’s distance, a diagnostic that measures how much fitted values change when a single species is removed. Excluding the top ten percent most influential species changed the picture in informative ways. Across all datasets, no supported effect ever reversed direction, and sixteen of twenty-three supported predictor combinations survived. But removal also unmasked new associations, including effects of LINE divergence, SINE abundance, and DNA transposon interactions that had been hidden by influential species in the full data. In sauropsids the signal proved especially fragile, with only two supported combinations retained and three new ones introduced after removing just three species, a warning that conclusions in that clade rest on a narrow empirical base.

A permutation test provided an additional safeguard. For each dataset, predictors were shuffled two thousand times and the full modeling procedure was repeated, generating a null distribution of coefficients against which the observed estimates could be judged. Twenty-one of thirty-one supported estimates fell outside the ninety-five percent null interval, with mammals showing the strongest support: all six of its supported estimates exceeded the null. In sauropsids, only the SINE abundance effect survived, prompting the authors to limit their interpretation of that clade’s results. The interaction between SINE abundance and sequence divergence predicted homogeneity consistently across all three response variables, with the strongest contribution coming from mammals. The direction of the effect aligns with what is known about SINE biology: young, abundant SINEs disrupt gene density uniformity, while old, diverged SINEs do not, consistent with observations in cattle where recently expanded ruminant-specific repeats preferentially insert into gene-rich regions whereas ancestral repeats show the opposite pattern. A divergence-threshold analysis reinforced this interpretation, showing that including repeats older than roughly two percent divergence added almost nothing to the estimated effects, implying that the youngest elements exert an outsized influence. By contrast, across all species combined, the proportion of young repeats was positively associated with gene density homogeneity under both the Gini coefficient and the coefficient of variation at every cutoff examined, a pattern that may be driven by ray-finned fishes.

The weakness of the signal in ray-finned fishes is itself scientifically meaningful. Amniote genomes are organized into gene-rich and gene-poor compartments that provide the structural backdrop against which SINEs accumulate differentially. Teleost genomes lack this compositional compartmentalization, so transposable elements may simply have no preferential target zones to amplify. The teleost-specific whole-genome duplication and subsequent rediploidization may have further reorganized genome architecture independently of repeat dynamics, obscuring any residual association. In other words, the absence of a repeat effect in fishes is not a failure of the hypothesis but a clue that the hypothesis depends on genomic features that not all lineages share. The authors also propose a mechanistic bridge between repeat age and chromosome-scale structure involving non-allelic homologous recombination. When recombination occurs between highly similar young repeats, double Holliday junctions tend to be resolved without crossovers, repairing breaks without changing gene counts. As repeats diverge, joint-molecule interfaces become unstable and structure-selective nucleases preferentially cleave them into crossovers, which between chromosomes can generate translocations that shuffle gene-rich and gene-poor segments and thereby increase uniformity. Old repeats additionally recruit DNA methylation and heterochromatin that can silence or erode neighboring genes, potentially amplifying local disparities. The authors are careful to note that this connection between junction resolution and gene density remains to be tested directly.

Several limitations temper the conclusions. Taxonomic sampling is uneven: only four non-avian reptile genomes passed the quality thresholds, amphibians were represented by just two anuran assemblies, and early-branching vertebrates such as jawless fishes were excluded entirely for poor contig recovery. Unplaced scaffolds were excluded from homogeneity measures even when large. Standard benchmarks like BUSCO completeness do not guarantee that repetitive regions are fully assembled, and alternative metrics such as the LTR Assembly Index, developed for plant genomes, are not applicable to many vertebrates that fall below required LTR abundance thresholds. The authors also emphasize that the observed associations reflect statistical relationships and do not necessarily imply causal links between repeat parameters and gene density homogeneity, which may instead be driven by unmeasured genomic features that covary with repeat divergence or abundance. Situated within the accordion model of genome size evolution, in which transposable element expansion is balanced by segmental deletion, the findings suggest that lineage-specific repeat activity shapes not just total DNA content but the internal proportional distribution of genes across chromosomes, a dimension of genome architecture that future work, especially explicit tests of chromosomal rearrangement histories, is well positioned to explore.

Subject of Research: Comparative genomics of gene density variation across vertebrate chromosomes in relation to repetitive DNA expansion

Article Title: Gene density homogeneity is disrupted by repetitive DNA Expansion

Article References: Hu, Z., Chien, S., Laird, K., & Blackmon, H. (2026). Gene density homogeneity is disrupted by repetitive DNA Expansion. PLOS One, 21(10), e0361209. https://doi.org/10.1371/journal.pone.0361209

Image Credits: AI Generated

DOI: 10.1371/journal.pone.0361209

Keywords: genomics, transposable elements, gene density, SINEs, vertebrates, chromosome evolution, genome architecture, RepeatMasker, mammals, ray-finned fishes, comparative analysis, PLOS One