Polyploidy, also known as whole-genome duplication (WGD), is a major evolutionary force in plants (Soltis et al., 2015; Van de Peer et al., 2021; Morris et al., 2024). WGDs generate genetic, phenotypic, and metabolic diversity and are associated with increased evolvability and diversification (Wendel, 2015; Soltis and Soltis, 2016; Landis et al., 2018; Doyle and Coate, 2019; Fox et al., 2020; Van de Peer et al., 2021; Morris et al., 2024). All living angiosperms are of ancient polyploid ancestry (Jiao et al., 2011), and 35% of extant vascular plant species may have originated via polyploidy (Wood et al., 2009). In addition, most crops are polyploids (Renny-Byfield and Wendel, 2014), and polyploidy plays an important role in plant breeding (Udall and Wendel, 2006; Sattler et al., 2016). Therefore, a better understanding of polyploid genome function and evolution is essential for both comprehending plant diversity and facilitating crop improvement.

In newly formed allopolyploids (those that arose through hybridization of closely related species and associated genome doubling), there are duplicate gene copies (homeologs) with redundant or overlapping expression and function. Over time, these duplicates experience various fates, ranging from retention of both copies with original function to homeolog divergence to loss of one copy via fractionation (Papp et al., 2003; Freeling et al., 2012; Tang et al., 2012; De Smet et al., 2013; Wendel et al., 2018). Some genes are consistently conserved as singletons, implying their single-copy status is advantageous and favored by selection (Paterson et al., 2006; De Smet et al., 2013; Li et al., 2016). In addition, following WGD, genes from one parent may be more highly retained than those from the other (i.e., subgenomic dominance) (Chaudhary et al., 2009; Evangelisti and Conant, 2010; Wendel et al., 2012).

For many other genes, duplicate copies are preserved following WGD, a phenomenon that may be explained by the dosage balance hypothesis, which argues that genes encoding subunits of protein complexes tend to be retained following polyploidy; loss of one copy of these genes may be selected against because of the disrupted stoichiometry of members of multi-subunit protein complexes (Birchler and Veitia, 2007; 2012; 2021). Additionally, genes encoding transcription factors are usually dosage-sensitive and are more likely to be retained in duplicate following WGD: loss of one copy of transcription factor genes may have a pleiotropic effect on many downstream genes involved in the same pathway; this effect would not be observed when losing one copy of genes acting at the termini of genetic networks (Blanc and Wolfe, 2004; Freeling et al., 2012; Birchler and Veitia, 2021).

Despite these observed gene retention patterns, the mechanisms underlying these patterns remain elusive, and the phenotypic consequences of changes in gene dosage are still largely unknown. This is primarily due to: 1) the lack of a homeolog-specific editing system for manipulating gene dosage across polyploid plants, except in a few crop species, including hexaploid wheat (Triticum aestivum) (Zhang et al., 2016; Liang et al., 2017) and tetraploid cotton (Gossypium hirsutum) (Chen et al., 2025; Yu et al., 2025); and 2) the absence of functional studies in organisms that best exemplify the earliest phases of WGD, especially species from natural systems. A deeper insight into gene fate following recent WGD is critical for understanding the genetic basis of the success of polyploids.

The diploid-polyploid system in North American Tragopogon (Asteraceae) represents an evolutionary model for studying the immediate consequences of polyploidy (Ownbey, 1950; Soltis et al., 2004; 2012). The naturally occurring allotetraploids Tragopogon miscellus and T. mirus formed in the last 95–100 years. The diploid parents of T. miscellus are T. dubius and T. pratensis, and those of T. mirus are T. dubius and T. porrifolius. Previous studies have demonstrated that novel arrays of karyotypes, gene content and expression, and epigenetic regulation were generated in these newly formed Tragopogon polyploids (Tate et al., 2009; Buggs et al., 2011; Chester et al., 2012; Spoelhof et al., 2017; Shan et al., 2020; 2024; Yoo et al., 2024). In addition, an efficient CRISPR/Cas9-mediated gene editing system has been recently developed in T. porrifolius (2x) and T. mirus (4x), allowing simultaneous editing of both alleles in diploid Tragopogon and all four alleles in allotetraploid Tragopogon (Shan et al., 2018; Shan et al., 2025).

In this study, we developed a homeolog-specific editing system in allotetraploid Tragopogon using MYB10 and DFR (dihydroflavonol 4-reductase) genes as examples. Both genes are involved in the well-characterized anthocyanin biosynthesis pathway (Elomaa et al., 2023). This represents the first such system established in a non-model polyploid plant. Our work provides the foundation for future investigations into gene retention patterns following polyploidy, with broad implications for both agricultural applications and evolutionary biology.

Although CRISPR/Cas-mediated gene editing has been applied in many allopolyploid plants (e.g., Braatz et al., 2017; Sánchez-Gómez et al., 2023; Xu et al., 2024; Zhang et al., 2025), these studies knocked out all copies of the target gene without differentiating between individual homeologs. Homeolog-specific editing has only been reported in a few polyploid crop species, including hexaploid wheat (T. aestivum) and allotetraploid cotton (G. hirsutum), to study function of genes associated with agronomic traits (Liang et al., 2017; Chen et al., 2025; Yu et al., 2025). GW2 controls grain weight in wheat. Liang et al. (2017) designed a sgRNA that perfectly matched the GW2-B1 and GW2-D1 homeologs but had a single nucleotide mismatch at the target site in the GW2-A1 homeolog. Following biolistic delivery of the CRISPR/Cas9 ribonucleoprotein complexes, 0%, 2.2%, and 4.4% of the T₀ wheat plants showed editing events at the A, B, and D homeologs, respectively (Liang et al., 2017). By leveraging SNPs between the two GhMML3 homeologs, sgRNAs were designed to specifically knock out either the A-subgenome copy (GhMML3_A12) or the D-subgenome copy (GhMML3_D12) in G. hirsutum (4x) (Chen et al., 2025). This study demonstrated that GhMML3_A12 and GhMML3_D12 regulate fiber development in a dosage-dependent manner. Yu et al. (2025) used CRISPR to edit either the GhHD1A or the GhHD1D homeolog in G. hirsutum and found functional divergence between the two homeologs in regulating trichome and fiber initiation.

Beyond plant systems, CRISPR-mediated allele-specific editing in diploid species (Yoshimi et al., 2014; Smith et al., 2015; Hashizume et al., 2025) and homeolog-specific editing in polyploid species (Gan et al., 2021; 2023) have been reported in several animal models. In an allele-specific editing experiment in the diploid laboratory rat (Rattus norvegicus), sgRNAs were designed to target either the Tyr ^C or Tyr ^c allele by exploiting one SNP at the CRISPR/Cas9 target site. Editing was found only in the targeted allele, with editing efficiencies of 30.4% and 28.6% for the Tyr ^C and Tyr ^c alleles, respectively (Yoshimi et al., 2014). Smith et al. (2015) showed that CRISPR/Cas9 can distinguish between two alleles in human cells based on a single nucleotide difference. In polyploid carp (Carassius gibelio), sgRNAs were designed to target either the CgRunx2b-A or CgRunx2b-B homeolog to study their respective roles in intermuscular bone development (Gan et al., 2023).

Currently, homeolog-specific editing studies in polyploid species employ several approaches to differentiate between homeologs. First, the non-PAM approach takes advantage of presence/absence variation in the PAM sequence–only the homeolog containing the PAM site will be edited. In the current study, we used this approach to differentiate the two DFR homeologs at the target site in exon 3 (Figure 1A). This method has also been applied in the study of cotton by Chen et al. (2025). The second approach aims to maximize mismatches within the 10 nucleotides immediately upstream of PAM (also known as the seed sequence) in the sgRNA, as targeting specificity is largely determined by this region. Studies in tetraploid cotton (Chen et al., 2025) and hexaploid wheat (Zhang et al., 2016) have employed this strategy. In T. mirus, a single SNP within the seed sequence successfully differentiated the T. dubius and T. porrifolius MYB10 homeologs (Figures 1A,B). Finally, the use of premixed CRISPR ribonucleoprotein (RNP) has been shown to improve homeolog editing specificity compared to expression of the CRISPR DNA construct. In wheat, Liang et al. (2017) reported that expression of the CRISPR/Cas9 DNA construct led to a 3.8% editing rate in the untargeted homeolog, whereas the RNP approach showed no off-target edits.

Despite the novelty of applying homeolog-specific gene editing in allotetraploid cotton (G. hirsutum), neither Chen et al. (2025) nor Yu et al. (2025) reported homeolog editing efficiency or whether any off-target events (i.e., modification of the untargeted homeolog) were detected. In both cases, homozygous mutant lines were often obtained in subsequent generations, implying monoallelic mutation (i.e., editing one of two alleles of the target homeolog) may have been present in the T₀ generation (Chen et al., 2025; Yu et al., 2025). In model crop species such as cotton, substantial resource investment has enabled the development of streamlined and highly efficient gene editing platforms, where reporting metrics such as editing efficiency and off-target rate may be considered less critical. However, for researchers working on non-model polyploid plants, including those of evolutionary significance, such information is essential for guiding future genome editing efforts.

In the present study, we demonstrated the feasibility of performing homeolog-specific editing in a non-model plant polyploid system. In allotetraploid Tragopogon mirus, the homeolog-specific editing efficiencies were 35.7% and 45.5% for MYB10 and DFR genes, respectively (Figure 1). Biallelic mutation of the target homeolog was detected in the T₀ generation, and importantly, no transgenic plant showed editing of the untargeted homeolog. This detailed report of homeolog-specific editing in T. mirus (4x) provides a valuable reference for researchers interested in applying similar technologies to their own species of interest.

One limitation of the current work is the relatively low percentage of biallelic mutants in the T₀ generation: 0% and 9.1% for MYB10 and DFR, respectively (Figure 1B). In diploid T. porrifolius (2x), the efficiency of creating DFR null mutants–where both alleles are edited–was 100% (Shan et al., 2025). Moreover, when using sgRNAs targeting conserved sequences shared by the T. dubius and T. porrifolius homeologs, 71.4% of the transgenic T. mirus showed edits in all four alleles (from both homeologs) (Shan et al., 2025). Therefore, the relatively low efficiency of homeolog-specific editing was unexpected.

Homeolog-specific editing efficiency in polyploid Tragopogon can be improved through several strategies. First, incorporating additional sgRNAs may enhance editing efficiency. Our study showed that, compared to using one sgRNA (targeting MYB10), using two sgRNAs (targeting DFR) increased the editing efficiency from 35.7% to 45.5% (Figure 1B). Future constructs could include three or four sgRNAs to further increase the frequency of biallelic mutants. Second, while only CRISPR/Cas9 was used here, combining CRISPR/Cas9 and CRISPR/Cas12a systems could broaden targeting options: Cas12a recognizes a TTTN PAM sequence, in contrast to the NGG PAM required by Cas9 (Zhang et al., 2023). Additionally, Cas12a crRNA arrays are more compact than Cas9 sgRNA arrays, allowing the incorporation of more guide RNAs in the construct (Port et al., 2020). Third, monoallelic mutants could serve as explants for a second round of transformation, increasing the likelihood of obtaining biallelic mutants. Finally, additional biallelic mutants are expected in the T₁ generation, as 25% of progeny from a self-fertilized monoallelic T₀ plant should carry biallelic mutations.

The availability of homeolog-specific editing in the evolutionary model T. mirus (4x) now enables us to address key questions in polyploid genome evolution and fractionation. Although the current work does not include a systematic phenotypic analysis of T. mirus mutants with modified DFR or MYB10 homeologs, such analyses will be performed in our future work. Several important questions will be addressed, including those listed here. For genes in which retention of both homeologs is predicted to be favored by selection (such as genes encoding subunits of protein complexes or those that are dosage-sensitive), what are the effects of losing one versus the other homeolog on co-expression network and phenotype? Does the parental copy matter in terms of resulting network and phenotypic response? Additionally, for genes for which reversion to singleton status is predicted to be favored by selection, both homeologs may still be present and expressed in the early stages of polyploidy, including 95–100-year-old T. mirus. How do the phenotypes and transcriptomes differ between plants retaining both homeologs and genome-edited individuals with only one homeolog? Finally, in cases where one parental homeolog is preferentially retained or expressed in polyploids, what are the consequences of retaining the non-preferred copy? In summary, homeolog-specific editing opens new avenues for investigating polyploid genome evolution, with significant implications for both basic and applied plant science.