Chromosome-level haplotype-resolved genome assembly of kiwifruit (Actinidia eriantha) Authors: Quan Jiang, Sheng Zhang, Dongmei Tang, Weiming Zhong, Qianming Zheng, Qing Liu, Yong Qi, Binbin Shi, Xiaohong Yao, Jia Zhou Published: 2026 ABSTRACT AbstractKiwifruit (Actinidia spp.) is an economically important plant that has undergone rapid development in the 20th century. Among these species, A. eriantha holds considerable agronomic and evolutionary significance. Here, we present a haplotype-resolved, chromosome-level genome assembly for a wild A. eriantha individual. The assembled genome sizes were 663.76 Mb for HAP1 and 633.05 Mb for HAP2, with contig N50 values of 21.82 Mb and 21.39 Mb, respectively. A total of 95.9% of HAP1 (636.55 Mb) and 99.25% of HAP2 (628.28 Mb) sequences were successfully anchored to 29 pseudochromosomes, resulting in scaffold N50 values of 21.89 Mb and 21.50 Mb, respectively. We annotated a total of 39,983 and 40,099 high-confidence protein-coding genes for two haplotypes, respectively. Genome evaluations showed high completeness and accuracy, with BUSCO completeness of 99.5% for both haplotypes as well as QV of 43.11 for HAP1 and 43.42 for HAP2. This high-quality genome provides a valuable reference for comparative and population genomics studies, and will facilitate the identification of genes controlling key agronomic traits, thereby accelerating molecular breeding in kiwifruit. Background & Summary Kiwifruit, renowned as the ‘King of Fruits’ for its exceptionally high vitamin C content and considerable economic value, has transitioned from a regional specialty to a globally important horticultural crop in recent decades1,2. This genus comprises 54 species with extensive natural diversity, among which A. chinensis, A. arguta, and A. eriantha represent the cornerstone of commercial cultivation and breeding programs3. Recently, the rapid advancement of genomic sequencing technologies, particularly the third-generation sequencing platforms such as PacBio HiFi and Oxford Nanopore Technologies (ONT), has provided unprecedented opportunities to unravel the complex genetic architectures underlying agronomically important traits in non-model species like kiwifruit. To date, more than 15 kiwifruit genomes have been sequenced and assembled, including wild accessions and cultivars2,4–6. Among these genomes, several genome assemblies of A. eriantha have been reported, including two wild individuals4,7 and three cultivated varieties8–10. These genomic resources have substantially advanced our understanding of A. eriantha, including its genome structure, evolutionary history, and functional phenotypes, facilitating future studies on molecular function, adaptation, and conservation. However, they collectively fail to capture the full spectrum of genomic variation and genetic diversity within the species. A critical gap exists in the genomic representation of unique wild populations from specific ecological niches and geographic origins. The genetic composition of these local populations often harbors novel alleles and structural variations shaped by distinct evolutionary pressures, which are indispensable for comprehending the species’ adaptive potential and enriching future breeding reservoirs11,12. In our previous study, we identified a specific wild population of A. eriantha that exhibits remarkable genetic divergence and phenotypic plasticity (characterized by a unique array of fruit and leaf colors) compared with other populations13,14. Notably, this population inhabits the marginal edge of the species’ suitable distribution area, where it likely faces heightened environmental stresses. Populations at such range margins are often hotspots for local adaptation and can serve as natural laboratories for studying evolutionary processes15. Therefore, generating a high-quality reference genome from a representative individual of this distinct population is a crucial step toward elucidating the molecular mechanisms underlying local adaptation, phenotypic evolution, and trait diversification. To address this need, we assembled a haplotype-resolved, chromosome-level genome for a wild individual from this unique A. eriantha population using PacBio HiFi sequencing. The primary genome sizes were 663.76 Mb for HAP1 and 633.05 Mb for HAP2, with contig N50 values of 21.82 Mb and 21.39 Mb, respectively. BUSCO analysis indicated high completeness, with 99.5% complete BUSCOs for both haplotypes, while QV scores of 43.11 (HAP1) and 43.42 (HAP2), respectively. The long-terminal repeat assembly index (LAI) further supported the high continuity of the assemblies, with values of 16.37 and 16.17, respectively. Collectively, these results demonstrate that the A. eriantha genome is of high quality, providing a valuable resource for identifying genes underlying adaptation and agronomic traits and accelerating molecular breeding efforts in kiwifruit. Methods Materials collection and sequencing The wild population was found in Hua’an country in Fujiang province. Fresh leaves from one wild mature individual in provenance location (24.86°N, 117.43°E) were collected in 2025 and immediately freezing in the liquid nitrogen (Fig. 1). Total genomic DNA was extracted with a modified CTAB method16. A SMRTbell library with 15–20 kb fragments was constructed and sequenced on the PacBio Sequel II platform. In addition, a 150 bp paired-end library with a 350 bp insert size was constructed for second-generation sequencing and sequenced on the DNBSEQ T7 platform. The sample location and morphological characteristics in natural environment of A. eriantha. Genome survey A total of 39.45 Gb (~60×) of raw short-read data were generated for the genome survey. After quality control with fastp v1.0.117, 39.36 Gb of clean reads were retained for downstream analysis. Genome characteristics were estimated based on a 21-mer analysis implemented in Meryl18 and GenomScope219 with ploidy = 2. The estimated genome size of A. eriantha was approximately 646.6 Mb with the heterozygous rate of 1.02% (Fig. 2). Genome survey of wild A. eriantha in this study. Chromosome-level genome assembly The genome was assembled using 45 Gb (~ 70×) of PacBio HiFi data with hifiasm v0.25.0-r72620 under default parameters. We generated two haplotype-resolved genome assemblies (HAP1 and HAP2) with high continuity and completeness. The total assembly size of HAP1 was 663.76 Mb, while HAP2 was slightly smaller at 633.05 Mb (Table 1). HAP1 consisted of 691 contigs, whereas HAP2 showed a much lower contig number (96), indicating a more contiguous assembly. Both haplotypes exhibited high contiguity, with contig N50 values of 21.82 Mb for HAP1 and 21.39 Mb for HAP2, and a contig N90 values of 18.13 Mb and 12.14 Mb, respectively. The largest contigs reached 28.34 Mb in HAP1 and 28.64 Mb in HAP2. 636.55 Mb (95.9%) of the HAP1 and 628.28 Mb (99.25%) of the HAP2 were anchored to 29 pseudochromosomes using RagTag v2.1.021 with a previously published A. eriantha genome (GWHFPWN00000000.1)4 serving as the reference, with 51 and 47 contigs anchored to chromosomes in HAP1 and HAP2. Scaffold-level statistics were consistent with contig continuity, with scaffold N50 values of 21.89 Mb for HAP1 and 21.50 Mb for HAP2. The number of assembly gaps was limited to 22 in HAP1 and 18 in HAP2, and the length of the largest scaffold was identical to that of the largest contig in both assemblies. GC content was highly consistent between the two haplotypes (35.72% in HAP1 and 35.71% in HAP2). Furthermore, we performed pairwise alignment between each haplotype and the reference genome using minimap222, followed by visualization with D-GENIES (https://dgenies.toulouse.inra.fr/)23. The results demonstrated a high degree of genome sequence similarity (Fig. 3). Table 1Statics and estimation of genome assembly of A. eriantha.Genome assemblyHAP1HAP2Primary assembly size (Mb)663.76633.05Number of contigs69196Contig N50 (Mb)21.8221.39Contig N90 (Mb)18.1312.14Length of the largest contig (Mb)28.3428.64Anchor size (Mb)636.55628.28Anchor rate (%)95.999.25Pseudochromosome number2929Number of contigs anchored to chromosome5147Length of the largest scaffold (Mb)28.3428.64Scaffold N50 (Mb)21.8921.50Number of gaps2218GC content (%)35.7235.71Number of protein-coding genes39,98340,099LAI16.3716.17QV43.1143.42Completeness (%) with Merqury8281.41Short reads mapping rate (%)98.9598.56HiFi reads mapping rate (%)99.8699.86 Statics and estimation of genome assembly of A. eriantha. Synteny between the different haplotypes and the reference genome. Gene prediction and functional annotation Repetitive elements were identified and grouped with the Extensive de novo TE Annotator (EDTA) v2.2.024 with the–anno 1 option. EDTA integrates structural features and homology-based searches to detect major transposable element (TE) types, including long terminal repeats (LTRs), long interspersed nuclear elements (LINEs), short interspersed nuclear elements (SINEs), and DNA transposons. Repetitive elements accounted for a substantial portion of the A. eriantha genome, with total repeat content of 44.19% in HAP1 and 43.91% in HAP2 (Table 2). LTR retrotransposons were the most abundant class, representing ~27.6% of the genome, followed by TIR elements (~9%) and Helitrons (~4%). Other repeat types, including LINEs, pararetroviruses, and fragmented repeats, contributed smaller proportions to the genome (Fig. 4). Table 2Repeat sequence statics of A. eriantha.ClassLength (bp) in HAP1Percent in HAP1Length (bp) in HAP2Percent in HAP2LINE3,543,0920.56%3,566,1410.56%LTR175,512,73627.57%175,036,59527.86%TIR59,370,6539.32%55,436,9378.83%Pararetrovirus102,3270.02%239,1520.04%Helitron25,406,4463.99%25,183,3044.01%Repeat_fragment17,343,8692.72%16,381,6442.61%Total281,279,12344.19%275,843,77343.91% Repeat sequence statics of A. eriantha. Genome features for two haplotypes of A. eriantha. From the outermost to the innermost circle, the tracks represent: chromosomes, GC content, gene density, Gypsy density, and Copia density. All density tracks were estimated using a 1-Mbp sliding window. The soft-masked genome generated by RepeatMasker v4.1.225 based on the results from EDTA. Gene structure prediction was performed using Braker326, integrated with transcriptome prediction, homology search and ab initio prediction. The RNA-seq data of five tissues (fruit, flower, stem, leaves and shoots) downloaded from the NCBI (SRP314429)27 from the previous study7, was utilized to train gene prediction model. According to the Braker3 pipeline, RNA-seq reads were aligned to the soft-masked genome using HISAT2 v2.2.128 with the–dta option and the file with SAM format was convert into BAM format with SAMtools v1.22.129. Proteins evidence from the Viridiplantae dataset of OrthoDB v1230 were also used to guide gene model prediction. A total of 39,983 and 40,099 high-confidence protein-coding genes were retained for HAP1 and HAP2, respectively, after filtering out gene models shorter than 50 amino acids or with incomplete structures using AGAT31 (Fig. 4; Table 1). Functional annotation was performed using multiple databases, including eggNOG5.0 database32, CDD33 and PFAM34 via InterProScan35, and SwissProt36 and UniProt37 via DIAMOND38 (E-value ≤ 1e−5) (Table 3). For HAP1, 76.90% and 34.59% of genes were annotated by Pfam and CDD, respectively, while 71.51% and 71.42% showed homology to SwissProt and UniProt entries. Similar annotation rate was observed for HAP2, with 77.44% (Pfam), 34.65% (CDD), and ~71% (SwissProt/UniProt) of genes annotated. EggNOG assigned functional categories to over 90% of genes in both haplotypes, indicating comprehensive and consistent functional characterization. Table 3The summary of functionally annotated protein-coding genes in the haplotype-resolved genome of A. eriantha.TypeDatabaseHAP1HAP2InterproscanPfam30,745 (76.90%)31,054 (77.44%)CDD13,830 (34.59%)13,896 (34.65)DiamondSwissprot28,591 (71.51%)28,517 (71.12%)Uniprot28,555 (71.42%)28,489 (71.05%)Eggnog-36,357 (90.93%)36,340 (90.63%) The summary of functionally annotated protein-coding genes in the haplotype-resolved genome of A. eriantha. Data Records The PacBio HiFi reads and raw short reads were deposited to NCBI Sequence Read Archive with the SRA accession number SRP66307939 (run accession numbers SRR3686142740 and SRR3686142841) under the BioProject accession number PRJNA1403621. The genome assembly, protein sequences and annotation file were submitted to the European Nucleotide Archive (ENA) with the accession numbers GCA_982292455.142 (HAP1) and GCA_982292445.143 (HAP2), and the Figshare database44 as well. Technical Validation Quality assessment The qualities of our genome assembly and annotation were assessed based on multiple key features. BUSCO v6.0.028 with the eudicots_odb12 database was used to estimate the completeness of both the genome and gene annotation. The results revealed that both HAP1 and HAP2 assemblies are highly complete, with 99.5% complete BUSCOs detected in the assemblies and ~97% completeness retained at the annotation level (Table 4). The low fractions of fragmented (≤0.9%) and missing BUSCOs (≤2%) also indicated overall high assembly and annotation quality for both haplotypes. Short- and long-read sequencing data were mapped to the assemblies with bwa v0.7.17-r118845 and Minimap2 v2.30-r128731, respectively. The short-read mapping rates reached 98.95% (HAP1) and 98.56% (HAP2), while HiFi long reads showed consistently high mapping rates of 99.86% for both haplotypes, indicating excellent assembly accuracy and completeness. Table 4BUSCO analysis statistics of A. eriantha.HAP1HAP2AssemblyAnnotationAssemblyAnnotationProteins (%)Proteins (%)Proteins (%)ProteinsComplete BUSCOs2,790 (99.5%)2,722 (97%)2,791 (99.5%)2,726 (97.2%)Complete Single-Copy BUSCOs1,668 (59.5%)1,780 (63.5%)1,768 (63.0%)1,776 (63.3%)Complete Duplicated BUSCOs1,122 (40%)942 (33.6%)1,023 (36.5%)950 (33.9%)Fragmented BUSCOs9 (0.3%)26 (0.9%)4 (0.1%)22 (0.3%)Missing BUSCOs6 (0.2%)57 (2%)10 (0.4%)57 (2%)Total BUSCO groups searched2,8052,8052,8052,805 BUSCO analysis statistics of A. eriantha. Assembly continuity was further evaluated using the LTR assembly index (LAI)46. The analysis was performed using predicting outputs of LTRharvest47 (implemented in Genometools v1.6.248) and LTR_FINDER_parallel v1.349 with their recommend parameters. LAI values were subsequently calculated with LTR_retriever v3.0.450. The resulting LAI scores were 16.37 for HAP1 and 16.17 for HAP2, indicating reference-level assemblies51. Consensus quality was assessed using Merqury v1.318, which yielded high QV scores of 43.11 for HAP1 and 43.42 for HAP2, indicating a very low base-level error rate. In addition, Merqury estimated genome completeness values of 82.0% and 81.41% for HAP1 and HAP2, respectively. These k-mer–based metrics might be affected by heterozygosity and repetitive content. In summary, taken together with the high BUSCO completeness, mapping rates, and LAI scores, these results demonstrate that the A. eriantha haplotype-resolved assemblies are of high accuracy and continuity, and are well suited for downstream comparative and functional genomic analyses. Genomic synteny analysis Comparative analysis between two haplotypes was performed using Minimap2 v2.30-r128731, SYRI v1.7.152 and then visualized with plotsr v1.1.053 (Figure S1). The results revealed 561 syntenic regions, spanning approximately 559.78 Mb in HAP1 and 553.02 Mb in HAP2, indicating that the majority of the genome is collinear between the two haplotypes. In addition, 278 inversions were detected, covering 20.49 Mb in HAP1 and 19.87 Mb in HAP2. Moreover, 235 translocations were identified, involving 2.29 Mb and 2.32 Mb in HAP1 and HAP2, respectively. These results further support a high level of structural conservation between the two haplotypes. Supplementary information Supplementary information Acknowledgements This work was supported by Guizhou key laboratory of molecular breeding for characteristic horticultural crops (grants Qiankehepingtai ZSYS [2025] 027), the National Natural Science Foundation of China (grants 32500318, 32570443) Guizhou provincial science and technology support program (grants Qiankehe [2023] 045). Author contributions Q.J., X.Y. and J.Z. designed this project. D.M.T helped to revise the manuscript. Q.J. collected the samples and performed data analyses with the help of S.Z., W.Z., Q.Z, Q. L., B.S. and Y.Q. Q.J. X.Y. and J.Z. wrote the manuscript. All authors approved the final manuscript for publication. Data availability The PacBio HiFi reads and raw short reads were deposited to NCBI Sequence Read Archive with the SRA accession number SRP663079 (run accession numbers SRR36861427 and SRR36861428) under the BioProject accession number PRJNA1403621. The genome assembly, protein sequences and annotation file were submitted to the European Nucleotide Archive (ENA) with the accession numbers GCA_982292455.1 (HAP1) and GCA_982292445.1 (HAP2), and the Figshare database (10.6084/m9.figshare.31034008.v1) as well. The version of the software employed in this work was detailed and can be accessed through a public database. Furthermore, all relevant parameters, with the exception of default settings, are explicitly listed in the manuscript. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Xiaohong Yao, Email: yaox@wbgcas.cn. Jia Zhou, Email: 616503691@qq.com. Supplementary information The online version contains supplementary material available at 10.1038/s41597-026-07414-w. References Associated Data Data Citations Supplementary Materials Supplementary information Data Availability Statement The PacBio HiFi reads and raw short reads were deposited to NCBI Sequence Read Archive with the SRA accession number SRP663079 (run accession numbers SRR36861427 and SRR36861428) under the BioProject accession number PRJNA1403621. The genome assembly, protein sequences and annotation file were submitted to the European Nucleotide Archive (ENA) with the accession numbers GCA_982292455.1 (HAP1) and GCA_982292445.1 (HAP2), and the Figshare database (10.6084/m9.figshare.31034008.v1) as well. The version of the software employed in this work was detailed and can be accessed through a public database. Furthermore, all relevant parameters, with the exception of default settings, are explicitly listed in the manuscript.