When numbers matter: Rethinking the role of gene duplication on short evolutionary timescales

Abstract

The potential roles of genomic structural variations (SVs) in the control of phenotypic traits and in evolution were suggested as early as the 20th century. However, they were then overshadowed by the emphasis put on single nucleotide polymorphisms (SNPs). Recently, SVs have received renewed attention in evolutionary research due to advancements in sequencing technologies and analytical methods.

At the macroevolutionary scale, plant genomes tend to evolve faster than those of other eukaryotes, due to the prevalence of whole genome duplication events (Wendel et al., 2016). Unlike other types of structural variants, such as inversions, copy number variations (CNVs) result from unbalanced mutations that affect the dosage, or amount, of a DNA sequence. When genes are involved, the number of copies of a gene varies from one individual to another. In plants, gene copy number variations (gCNVs) are likely to be abundant due to events such as mating system shifts (the efficacy of purifying selection is reduced in selfing species), hybridization and subsequent genome rearrangements, and whole genome duplications followed by biased retention (Panchy et al., 2016; Wendel et al., 2016; Van de Peer et al., 2017). For example, in Arabidopsis thaliana (L.) Heynh. (Brassicaceae) 10 to 18% of all genes display CNVs (Zmienko et al., 2020; Jaegle et al., 2023). In the genus Picea Mill. (Pinaceae), at least 10% of the protein-coding genes display CNVs (P. abies (L.) H. Karst and P. obovata Ledeb., Q. Zhou et al., 2025; and P. glauca (Moench) Voss and P. mariana (Mill.) Britton, Sterns & Poggenburg, Prunier et al., 2017).

Gene duplications have primarily been studied for their roles in long-term evolution. However, a change in gene dosage usually results in a change in the amount of gene products, such as RNA or proteins (e.g., Shao et al., 2019). Therefore, gCNVs have a unique, multiallelic, and quantitative nature. Fully apprehending their role in short-term evolutionary processes requires studying them as quantitative genotypes rather than in a presence/absence (or biallelic) manner, as is most often done (Figure 1, top panel). Unlike SNPs, the accuracy and resolution of gCNV genotyping are usually dependent on the platform used. From short-read sequencing data, one can use biased allelic ratios (Figure 2A) and changes in the depth of coverage (DoC) caused by the mis-mapping of reads from duplicated regions to the same locus in the reference genome to identify CNVs (Figure 2). However, short reads often fail to capture the underlying genetic structure of gCNVs, and changes in DoC can only be interpreted as relative copy numbers across homologous chromosomes (Figure 1, middle panel). Long-read sequencing is a promising alternative that allows for the phasing of the various alleles to obtain absolute copy numbers (see Figure 1, bottom panel). However, long-read data can still be biased in assembling repetitive regions (Carvalho et al., 2025 [preprint]), they are more computationally demanding, and likely too costly for extensive population-level genomics studies. Nevertheless, continuous advancements in sequencing technologies and CNV analysis methods open the door to more extensive studies focusing on gCNVs, even in non-model species (Karunarathne et al., 2023). In the following sections, we recognize gCNVs as a largely untapped source of genetic variation and explore their potential for studying short-term evolution in plants. We also discuss the main challenges of incorporating this type of polymorphism into population and quantitative genomic frameworks, and how plants, as a study system, offer an opportunity to address them.

Despite all the evidence that gCNVs are a non-negligible source of polymorphism, it is still unclear to what extent they can be used to address broad questions in evolutionary biology, such as better inferring past demographic events or refining predictions of population responses to global change for conservation strategies. This is partly due to difficulties in estimating key evolutionary parameters, which largely prevent their use in population and quantitative genetic models (Mérot et al., 2020). Gene copy number variation is more likely to occur via low-copy repeat mechanisms (e.g., non-allelic homologous recombination). As a result, duplications at the origin of gCNVs may span multiple genes (as illustrated in Figure 2), thus affecting the apparent gene duplication rate, which has been estimated to be higher than substitution rates (Katju and Bergthorsson, 2013). The presence of gCNVs induces mismatches during synapsis and can originate from different molecular mechanisms (Hastings et al., 2009). Thus, variable mutation rates in terms of copy number change are expected across loci and mechanisms. Additionally, there may be asymmetric rates of copy number gain and loss, as well as state-dependent mutation rates, where the rate of change in copy number varies with the actual number of copies. Gene copy number variations can affect patterns of recombination and segregation. Therefore, models developed to study the evolution of microsatellites or gene families, such as the stepwise mutation model (SMM) and the birth–death model, respectively, do not accurately predict gCNV evolution. However, these models could produce relevant diversity summary statistics when applied to a large enough number of gCNVs to avoid gene-specific bias. For example, allele size variance (Valdes et al., 1993) and Goldstein's δμ² (Goldstein et al., 1995), which are based on relative copy number, and CNV entropy (similar to Shannon's diversity index) and R_ST (Slatkin, 1995), which are based on absolute copy number, could be useful for measuring within-population diversity and between-population divergence of gCNVs, respectively. Comparing these estimates with those obtained from neutral SNPs (e.g., nucleotide diversity and F_ST, respectively) would also inform us about the global distribution of fitness effect of gCNVs.

For structural variation, it is often considered that their phenotypic and fitness effects scales with their size. For gCNVs it means that both could increase with copy number and the apparent copy number may therefore be the result of a trade-off between a high mutation rate and large fitness effects. The highly dynamic nature of plant genome evolution makes them excellent models for addressing the knowledge gaps associated with gCNVs with >1400 reference genomes available (Bernal-Gallardo and de Folter, 2024). For example, the Brassicaceae plant family now contains the highest number of genome and transcriptome sequences for any plant lineage, allowing extensive comparative studies to address some of the challenges mentioned above and separate species-specific effects from general properties of gCNVs. Several genera have closely related species with different mating systems (selfing and outcrossing), which could be used to investigate the main evolutionary forces shaping the evolution of gCNVs while controlling for phylogenetic relationships, as has been done for RNA expression previously (e.g., Zhang et al., 2022). The change in ploidy can also be used to study mutation and recombination rates by comparing patterns of gCNVs between the sub-genomes of auto-polyploids as any differences between the sub-genomes will have been acquired after the polyploidization event. In addition, interspecific hybridization and resynthesized polyploids can be generated experimentally, and many plants are model species for functional genomics (Bernal-Gallardo and de Folter, 2024). The use of natural and synthetic resources, together with state-of-the-art sequencing and gene editing technologies, would allow for direct measurements, for example, of the phenotypic and fitness effects of gCNVs.

Despite the limitations exposed above, the multiallelic and quantitative nature of gCNVs makes them excellent markers for quantitative genetics as, in many cases, the phenotypic values of traits and the copy number of causal gCNVs show a quantitative relationship, enabling straightforward genotype-to-phenotype mapping. The prevalence of gCNVs in plant genomes makes them natural candidates in the control of quantitative traits and adaptation along environmental gradients.

Forest trees are particularly relevant models for studying such questions, as they often have extensive ranges with populations connected by long-distance gene flow and show strong patterns of local adaptation (Savolainen et al., 2013). In a recent study, we showed that gCNVs are widespread and involved in local adaptation in both Picea abies (Norway spruce) and Picea obovata (Siberian spruce), two keystone species of the Eurasian boreal forest (Zhou et al., 2025). Importantly, we found no overlap between candidate genes detected purely from SNP variation and genes whose copy number correlates with environment and/or phenotypic variation. This means that, and in contrast with genomic inversions, the explanatory power of gCNVs is not captured by SNPs, and they must be specifically studied to gain a comprehensive view of the genetic architecture of local adaptation patterns and of phenotypic traits.

Recent studies have shown that considering structural variants in addition to SNPs better explains the heritability of complex traits. However, these studies tend to consider all structural variants together (including gCNVs) in a biallelic manner (Figure 1, top panel). In doing so, information about any quantitative relationship between the number of copies of a gene and the phenotypic trait of interest is lost when there are more than two copy-number states. Gene copy number variations can thus be considered largely untapped genetic variation that may even explain some of the so-called ‘missing’ heritability. Such information would be particularly relevant for plant breeding, where genomic selection and phenotypic prediction are increasingly being used to shorten breeding cycles and reduce phenotyping costs.

The recent burst in high-throughput sequencing has revealed a high prevalence of gCNVs in eukaryotes, and the evolutionary significance of this polymorphism largely remains to be determined. We argue that fully comprehending the role of gCNVs in plant evolution requires solving their structure at the genomic level, but even more importantly to study them as quantitative genotypes. The rapid development of long-read sequencing technologies is full of promise for solving their haplotypic structure. However, we also want to emphasize that much can already be done by using short-read sequencing data and dedicated analytical methods as illustrated above. Extensive population-level genomic data have been generated over the past decade—for example for conservation and breeding purposes—and we hope that this paper will serve as an incentive to reanalyze these data with a focus on gCNVs. Finally, plants display valuable features—such as mating system transitions and polyploidization—for designing comparative frameworks to measure population genetics parameters much needed to fully comprehend the role of gCNVs in evolutionary response at short and intermediate timescales.

F.L., P.M., and Q.Z.: Conceptualization, visualization, writing – original draft; writing – review and editing. P.M.: Funding acquisition, supervision.