The haplotype-resolved genome assembly of the hexaploid kiwifruit Actinidia deliciosa reveals its hybrid origin and polysomic inheritance

Abstract

The kiwifruit (genus Actinidia) exhibits extensive polyploidy, with natural populations ranging from diploid to tetraploid, hexaploid, and even higher ploidy levels. Actinidia deliciosa (green kiwifruit) is typically hexaploid (2n = 6x = 174), whereas its close relative A. chinensis (golden kiwifruit) is primarily diploid (2n = 2x = 58) or tetraploid. This ploidy diversity reflects a reticulate evolutionary history, likely shaped by hybridization and genome duplication involving multiple ancestral lineages. Determining the origin and inheritance of hexaploid A. deliciosa—the dominant kiwifruit species in global cultivation—is critical for understanding the role of polyploidization in the evolution of this species and for the development of breeding strategies.

Dear Editor,

A long-standing debate centers on whether hexaploid A. deliciosa arose through allopolyploidy or autopolyploidy. Early molecular and cytogenetic evidence favored an allopolyploid origin, proposing that diploid A. chinensis contributed an entire subgenome, supported by subgenome-specific repeats and phylogenetic divergence within polyploid genomes (Crowhurst and Gardner, 1991; Testolin and Ferguson, 1997). In contrast, other studies supported an autopolyploid origin, citing non-preferential chromosome pairing and subgenome homogeneity (Mertten et al., 2012). Recent genomic sequencing has further complicated this debate, suggesting an autohexaploid origin with A. chinensis as the principal progenitor. Post-polyploidization chromosomal rearrangements observed in the genome indicate a more complex evolutionary history, potentially classifying A. deliciosa as a segmental allopolyploid (Liu et al., 2024). Although diploid A. chinensis is widely recognized as a primary progenitor, the existence of any additional contributors and the mechanisms governing their genomic inheritance remain unresolved.

To address this issue, we generated a high-quality haplotype-resolved hexaploid genome assembly of A. deliciosa using 52.6 Gb of PacBio HiFi reads (∼15× coverage per haplotype), 206.3 Gb of Hi-C data (∼57× coverage per haplotype), and 21.4 Gb of Nanopore ultralong reads (N50 = 100.6 kb) (Supplemental Table 1). The sequences were assembled into 5154 contigs with a total size of 3.91 Gb and an N50 length of 5.91 Mb. Contigs were filtered for redundancy and scaffolded using YaHS (Supplemental Text 1), followed by manual curation. This process produced 174 long scaffolds, each corresponding to an A. deliciosa chromosome, collectively representing 90.0% of the genome assembly (Supplemental Table 2). The high contiguity and completeness of the assembly were supported by multiple metrics, including a long terminal repeat assembly index of 14.13, a BUSCO score of 99.60%, and strong syntenic alignment with the A. chinensis genome (Figure 1A). These metrics indicate higher overall assembly quality than the recently published genome assembly (Liu et al., 2024). Repetitive elements accounted for 51.33% of the assembly (Supplemental Table 3). We annotated 249 392 protein-coding genes, 83.09% of which were assigned predicted functions.

Figure 1.

Genomic architecture and evolutionary origin of hexaploid A. deliciosa.

(A) Synteny analysis of the hexaploid A. deliciosa and diploid A. chinensis genomes. This panel shows a dot plot of the 174 hexaploid chromosomes (vertical axis) aligned to the 29 diploid reference chromosomes (horizontal axis).

(B) Allelic dosage distribution across the 29 hexaploid A. deliciosa accessions. Boxplots for each accession (horizontal axis) show proportions of bi-allelic SNPs with 4:2 (green), 5:1 (blue), or 3:3 (red) reference-to-alternate (or vice versa) allele ratios, calculated across all 29 chromosomes.

(C) Phylogenetic relationships among haplotypes and progenitors. Tree topology frequencies were inferred from maximum-likelihood phylogenies constructed using syntenic orthologs across six haplotypes (haps) of A. deliciosa, A. chinensis, diploid A. deliciosa, and A. polygama (Apoly; outgroup).

(D) Haplotype origins and phylogeny. Genome segments were assigned to progenitor lineages (A. chinensis [Achin] or diploid A. deliciosa [Adeli]) using sequence similarity with whole-genome alignments (WGAs). White regions on the chromosome maps indicate poorly aligned genomic regions. The tree was constructed from concatenated WGAs using a maximum-likelihood algorithm.

(E) Site-frequency spectrum (SFS) analysis. Empirical A. deliciosa polymorphisms were compared to simulated SFSs under disomic, polysomic, and intermediate inheritance models (dN = 0.4). Statistical significance was evaluated using the Kolmogorov–Smirnov (K–S) test.

(F) Proposed polysomic inheritance model for hexaploid A. deliciosa.

(G) Gene expression by ancestral origin. The expression distributions for alleles derived from the diploid and tetraploid progenitors were compared using the K–S test.

(H) Chromatin accessibility of homoeologs. Normalized ATAC-seq signals show accessibility for diploid A. deliciosa–derived (blue) and A. chinensis–derived (orange) homoeologous regions in fruit tissue. Statistical significance was evaluated using the K–S test.

(I) DNA methylation patterns by genetic ancestry. Methylation levels are shown for genic and flanking regions of progenitor-specific alleles. Differences in promoter region methylation were assessed using the Mann–Whitney test.

(J) Evolutionary model of A. deliciosa polyploidization, illustrating the proposed hybridization and genome duplication events.

We first attempted to resolve the 174 chromosomes into subgenomes using SubPhaser (Jia et al., 2022) and analysis of repeat profiles (Wang et al., 2020). However, this approach was unsuccessful due to insufficient differential k-mers and a lack of distinct repetitive elements among haplotypes (Supplemental Figure 1); these characteristics likely reflect the close evolutionary relationship of the putative genome donors or a loss of heterozygosity due to homoeologous exchange (Han et al., 2025). To further investigate the genome configuration of A. deliciosa, we analyzed resequencing data from 29 hexaploid A. deliciosa accessions (Liu et al., 2024), identifying 33 434 660 high-quality bi-allelic SNPs using the A. chinensis genome (Han et al., 2023) as a reference. Genotype analysis revealed that, on average, 44.8% of SNPs in each accession exhibited a 4:2 allele ratio (reference vs. alternative, or vice versa), with 5:1 and 3:3 ratios observed less frequently (Figure 1B). This pattern supports a hybrid origin involving diploid and tetraploid progenitors. Among the SNPs with a 4:2 ratio, 62.3% had four of the six alleles matching the A. chinensis genome, which suggests that the tetraploid progenitor was closely related to this species. Furthermore, phylogenetic analysis of orthologs showed that the haplotypes of hexaploid A. deliciosa frequently clustered with those of the recently discovered diploid A. deliciosa (Xia et al., 2023) (Supplemental Figure 2). The diploid A. deliciosa is morphologically and genetically distinct from A. chinensis cultivars (Xia et al., 2023; Li et al., 2025). To clarify its role in the formation of the hexaploid variety, we reconstructed phylogenetic trees for 9252 syntenic orthologs using A. polygama as an outgroup. We hypothesized that tree topology and haplotype clustering would reflect the evolutionary history of hexaploid A. deliciosa. The dominant topology grouped four haplotypes of hexaploid A. deliciosa with A. chinensis and two with diploid A. deliciosa (Figure 1C), which confirms that the diploid and tetraploid progenitors were genetically close to diploid A. deliciosa and A. chinensis, respectively.

In polyploids, genetic inheritance typically follows one of two models: disomic (typical of allopolyploids) and polysomic (common in autopolyploids). To determine the inheritance model in A. deliciosa, we analyzed haplotype similarity relative to its two progenitors. Under a strict disomic model, two haplotypes should closely resemble diploid A. deliciosa and four should resemble A. chinensis, whereas polysomic inheritance would preclude such partitioning. We generated whole-genome alignments (WGAs) of all six haplotypes alongside the genomes of the two putative progenitors and A. polygama. Both phylogenetic and genomic synteny analyses of hexaploid A. deliciosa revealed limited regions that were consistent with disomic inheritance; however, most chromosomal regions lacked a distinct 4:2 haplotype ratio, consistent with polysomic inheritance (Figure 1D; Supplemental Figure 3). This pattern was also observed in the previously published genome of hexaploid A. deliciosa (Liu et al., 2024) (Supplemental Figure 4). For further validation, we simulated disomic and polysomic inheritance models and compared them to the resequencing data. Coalescent simulations produced the expected neutral site-frequency spectra under different historical and inheritance scenarios. The observed spectra closely aligned with the polysomic model but diverged significantly from disomic predictions (Figure 1E). Combined with previous evidence of bivalent chromosome associations during meiosis in this species (Mertten et al., 2012), these findings strongly support polysomic inheritance in hexaploid A. deliciosa (Figure 1F), explaining the dosage variability observed in gene phylogenies (Figure 1C) and genomic ancestry patterns (Figure 1D; Supplemental Figure 3).

Allopolyploids frequently exhibit subgenome dominance in gene expression, a phenomenon rarely observed in autopolyploids. Given that hexaploid A. deliciosa originated from hybridization between two closely related but distinct progenitors—diploid A. deliciosa and tetraploid A. chinensis—we investigated whether gene expression dominance occurs in this species. Transcriptome sequencing was performed across multiple tissues, including root, stem, leaf, and fruit at five developmental stages (25, 55, 85, 115, and 145 days post-anthesis) (Supplemental Table 1). Among 138 806 genes with a clear genetic origin, 55.66% were transcribed (FPKM > 0.5) in at least one tissue. Phylogenetic clustering (Figure 1C) classified these expressed genes by ancestral origin. Genes derived from diploid A. deliciosa showed significantly (P < 1.47E−05), albeit modestly, higher expression levels than those from A. chinensis across all tissues and developmental stages (Figure 1G; Supplemental Figure 5), which suggests mild expression dominance between homoeologs in the hexaploid genome.

To investigate potential mechanisms underlying this dominance, we analyzed chromatin accessibility using ATAC-seq (Supplemental Figure 6). Principal component analysis revealed tissue-specific clustering of samples, confirming data reproducibility (Supplemental Figure 7). Between 7966 and 19 593 accessible chromatin regions per tissue were identified with high confidence. However, genes from the two progenitors showed no significant differences in the number of accessible chromatin regions within promoter regions (Supplemental Figure 8) or in overall chromatin accessibility near transcription start sites (Figure 1H; Supplemental Figure 9). Gene expression dominance is often linked to transposable element–associated differences in DNA methylation between subgenomes. We therefore profiled DNA methylomes in leaves (Figure 1I). Global methylation levels were 64.2% for mCG, 34.5% for mCHG, and 10.3% for mCHH. Although mCG levels were similar between genes inherited from different progenitors, genes associated with diploid A. deliciosa displayed lower mCHG and mCHH methylation in promoter regions than those derived from A. chinensis (Figure 1I), implicating methylation variation as a potential regulator of expression dominance among homoeologs.

In conclusion, our study provides strong evidence that hexaploid A. deliciosa originated from hybridization between a tetraploid progenitor closely related to A. chinensis and a diploid progenitor genetically similar to diploid A. deliciosa, occurring approximately 2.1 million years ago (Figure 1J), resulting in a genome that functions as a unified gene pool under polysomic inheritance. Despite the absence of distinct subgenomes, mild expression dominance was observed between homoeologs, potentially influenced by inherited epigenetic differences, such as DNA methylation in regulatory regions.

Data availability

Raw reads for genome sequencing have been deposited in the National Genomics Data Center (NGDC) under BioProject no. PRJCA037936. Genome assemblies and annotations are also available on FigShare at https://doi.org/10.6084/m9.figshare.28689167.

Funding

This study was supported by the Zhejiang Provincial Natural Science Foundation of China (LR23C150001, LQ24C150003, and Q24C150007) and the NSFC Excellent Young Scientists Fund (Overseas) awarded to X.S.

Acknowledgments

The authors have no conflict of interests.

Author contributions

X.S. conceived the project. X.L. and H.K. assembled the genome. S.Z., S.X., L.H., X.H., J.G., and M.W. collected samples and prepared libraries. X.S., X.L., and C.Z. analyzed the data. X.Y. discussed the data. X.S. and X.L. wrote the manuscript.

Published: July 7, 2025