Allelic structural variation at the NLR25-1 locus enhances defense against Pseudomonas syringae in kiwifruit

Abstract

Background

Canker disease caused by Pseudomonas syringae pv. actinidiae (Psa) poses a major threat to cultivated kiwifruit, and utilization of wild relatives are key to improve resistance. However, comprehensive comparative genomic analyses between cultivated kiwifruit and their wild relatives with enhanced resistance to Psa remain limited.

Results

Here we generate chromosome-scale genome assemblies for eleven wild Actinidia eriantha accessions and one interspecific hybrid between Actinidia eriantha and cultivated Actinidia chinensis var. chinensis. Integrating these with twelve previously released genomes including three Actinidia eriantha and nine Actinidia chinensis var. chinensis, we construct a reference-unbiased graph-based pangenome. These datasets reveal extensive genomic variation, including 31,790,044 SNPs, 13,512,079 InDels and 623,478 structural variations, and provide a landscape of structural variations within and between the two species. Leveraging these datasets, we identify a wild allele showing allele-specific expression, AeNLR25-1, which enhances Psa resistance in cultivated kiwifruit. Genetic and molecular analyses demonstrate that a transposable element-induced structural variation in the AeNLR25-1 promoter introduces a species-specific WRKY binding site, conferring enhanced defense against Psa.

Conclusions

Pangenome across cultivated species and wild relatives provides a theoretical framework for accelerating kiwifruit genetic improvement through pangenome-enabled identification of favorable wild alleles.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13059-026-04068-0.

Background

The successful domestication of plants and animals was one of the most important events in the history of human civilization. As one of the most recently domesticated fruit crops, kiwifruit (Actinidia ssp.) has gained popularity worldwide due to its high vitamin C content, balanced nutritional composition of minerals, dietary fiber, and other beneficial metabolic products. Native to China, kiwifruit seed was initially introduced to New Zealand in 1904 and since then the domestication of kiwifruit began [1]. In recent years, kiwifruit cultivation has rapidly expanded to the world’s temperate regions and the crop currently covers approximately 286,348 hectares and produces nearly 4.433 million tons of fruit in 2023 (http://faostat.fao.org).

However, the biggest challenge for the cultivated kiwifruit (A. chinensis var. chinensis/A. chinensis var. deliciosa complex, with diploid, tetraploid and hexaploid varieties) is the outbreak of canker disease caused by Pseudomonas syringae pv. actinidiae (Psa) [2, 3]. Psa was first reported in Japan as the causal agent of the bacterial canker of kiwifruit [4]. Subsequently, this bacterial pathogen was recorded and prevalent in all major kiwifruit-producing countries [5–9], resulting in substantial economic losses [10, 11]. The cultivated kiwifruit lacks genetic diversity particularly for Psa resistance-related haplotypes, possibly due to intensive artificial selections in fruit yield and quality.

By contrast, wild relatives of Actinidia have adapted to diverse ecological environments and manifest broad genetic and phenotypic diversity [12]. These wild species are characterized with a rich source of allelic variation and contain genes for biotic and abiotic stress tolerance, as well as consumer preferred traits, such as colorful flesh, high levels of soluble solids content, flavor compounds, vitamin C content, and other nutritional metabolites [13–18]. Therefore, effectively utilizing the natural diversity of these wild germplasms is crucial for promoting genetic improvement of kiwifruit.

The assembly of several haplotype-resolved Actinidia reference genomes has allowed comprehensive characterization of genetic diversity in terms of SNPs and small insertion/deletions (indels) between individual haplotypes, unraveling favorable trait domestication and adaptive evolution [19–23]. Nonetheless, accumulating evidence have implicated that large structural variants (SVs), such as copy number variants (CNVs) and presence/absence variants (PAVs), also play pivotal roles in plant functional diversity and adaptive evolution [24]. Recent advances in kiwifruit genomics include a pangenome constructed by using 14 chromosome-scale haplotype-resolved genome assemblies from seven diploid representative accessions in A. chinensis var. chinensis [20], and a super-pangenome build from 15 high-quality assemblies of eight Actinidia species [25]. These analyses demonstrated that SVs contribute to phenotypic variance/diversity and can be utilized in functional genomic studies and accelerating breeding design.

Indeed, the genus Actinidia includes 54 species and 75 taxa and exhibits tremendous genetic variation/diversity within or between Actinidia species in its botanical or horticultural characteristics, such as organ morphology, bud germination time, flowering time, fruit yield/quality and adaptability to various environments [26–28]. Except for the cultivated A. chinensis var. chinensis/A. chinensis var. deliciosa, the vast majority of wild relatives are unexplored. Among them, A. eriantha is of particular interest to breeders due to its distinct resistance to Psa, exceptionally high vitamin C content, relatively short juvenile phase and interspecific crossability [29]. This species is widely distributed in the large area of the middle and lower reaches in Yangtze River Basin of China, and has been subjected to breeding selection and interspecific hybridization and introgression [12, 30–32]. Currently, only a limited number of genome assemblies for A. eriantha are available, which hinders the dissection and utilization of its genetic variants [14, 31, 33, 34]. To fully explore and utilize the favorable genetic resources of A. eriantha, it is necessary to assemble additional genomes to facilitate effective SVs detection and their uses in genetic improvement of kiwifruit.

In this study, combining with the previously released chromosome-level reference genomes from 12 cultivated A. chinensis var. chinensis accessions [20, 22, 23, 35], we constructed a reference-unbiased graph-based pangenome by de novo assembling chromosome-level genomes of an additional eleven wild A. eriantha accessions collected from its natural distribution area and one artificial hybrid individual between A. eriantha and A. chinensis var. chinensis. Comparative analyses reveal the distinct structural variations across the tested species, enabling the discovery of a wild haplotype harboring a highly conserved WRKY-binding cis-element in an SV embedded in the promoter of an NLR gene that has the potential to increase resistance against Psa in cultivated kiwifruit. Our findings provide insights into the exploitation of wild genetic resources by constructing interspecific pangenome across diverse species for crop improvement.

Results

Chromosome-level genome de novo assemblies and annotation of 12 kiwifruit accessions

To capture the genomic variations, 11 wild A. eriantha accessions (LD19 and YX01 are male, and the rest are female) representing its natural distribution areas and an interspecific hybrid F1 individual (named ‘HH-1’) between an A. eriantha accession ‘White’ [33] and an A. chinensis var. chinensis accession ‘H0809’ [35] were selected for de novo genome assembly (Table 1). All the tested accessions are diploid and their fruit characteristics are highly variable (Additional file 1: Table S1, Additional file 2: Fig. S1). Using the PacBio Revio sequencing platform, a total of 14.1 to 64.4 Gb of HiFi reads (approximately 21.7 to 92.0 ×) were generated (Additional file 1: Table S2) and then these reads were assembled to contigs of two haplotypes (primary haplotype and alternate haplotype) using hifiasm (v0.16.1) [36]. For the 11 wild A. eriantha accessions, contigs were anchored onto 29 chromosomes using the recently released telomere-to-telomere (T2T) genomes of MDHAPA (A. eriantha) [31]. For ‘HH-1’, contigs were assigned to two chromosome-scale haplotypes (haplotype 1 sourced from A. eriantha, here after HH-1.hap1; and haplotype 2 sourced from A. chinensis var. chinensis, here after HH-1.hap2) using quarTeT software [37] based on their collinearities with the T2T reference genomes [31, 35]. Finally, a total of 24 de novo haplotype assemblies were generated and the final assembly sizes ranged from 581.2 to 630.7 Mb with contig N50 ranged from 5 to 21 Mb (Table 1). The BUSCO completeness of these assemblies was estimated to be 98.62% on average, with the lowest score 94.00% (Table 1). All the long terminal repeat (LTR) assembly index (LAI) scores exceed 18.94 and thus meet the “reference” level [38], ensuring the quality of the 24 haplotype assemblies (Table 1). Next, we utilized the EDTA toolkit [39] to annotate transposable elements (TE) and reveal the repetitive sequences’ contents ranging from 38.14% to 41.92% (Table 1, Additional file 1: Table S3). Combining the previously published genomes (Additional file 1: Table S4) [22, 23, 31, 35, 40, 41], we found that the average genome size of A. eriantha group was approximately 12.5 Mb larger than that of A. chinensis var. chinensis group (Additional file 2: Fig. S2a, Additional file 1: Table S5). Additionally, the average TE content in A. eriantha (40.9%) was slightly higher than in A. chinensis var. chinensis (38.6%) (Additional file 2: Fig. S2b, Additional file 1: Table S3). Obviously, the increased genome size can be primarily explained by the corresponding increase of TE size (Additional file 2: Fig. S2c). By integrating transcriptome-based, ab initio, and homologous protein-based prediction methods, a total of 42,742 to 48,555 genes across the 24 assemblies were finally predicted (Table 1).

Table 1.

Statistics of the assembly and annotation of 12 newly sequenced kiwifruit genomes in this study

Sample	Species	City	Location	Accession	Total size of contigs (Mb)	Number of contigs	N50 (Mb)	Total size of genomes (Mb)	gaps	TE size (%)	Number of genes	Genome BUSCOs (%)	LAI
CY38	A. eriantha	Ganzhou City, Jiangxi Province, China	25.549876 N, 114.307938 E	CY38.hap1	646	460	19	629.4	23	41.92	47,199	99.20	20.43
				CY38.hap2	629	216	17	621.1	31	40.95	45,477	98.60	20.19
DG01	A. eriantha	Qiandongnan City, Guizhou Province, China	26.149891 N, 109.051666 E	DG01.hap1	772	3,777	13	613.1	75	40.72	45,108	98.60	20.17
				DG01hap2	688	1,257	14	620.4	47	40.85	46,552	99.00	20.56
GZ26	A. eriantha	Nanping City, Fujian Province, China	27.521293 N, 117.260915 E	GZ26.hap1	660	707	21	630.7	6	40.64	46,457	98.90	21.04
				GZ26.hap2	617	211	19	607.5	21	40.36	45,294	99.00	20.32
HN01	A. eriantha	Zhuzhou City, Hunan Province, China	27.068227 N, 113.5619189 E	HN01.hap1	639	376	20	623.2	21	41.42	45,872	99.40	21.90
				HN01.hap2	626	226	18	616.9	42	40.93	45,907	98.70	21.52
LD19	A. eriantha	Lishui City, Zhejiang Province, China	28.045319 N, 120.064087 E	LD19.hap1	622	573	5	603.5	194	40.66	44,858	94.00	20.67
				LD19.hap2	627	343	10	581.2	75	41.38	42,742	97.70	21.34
MQ32	A. eriantha	Fuzhou City, Fujian Province, China	26.053724 N, 118.882756 E	MQ32.hap1	640	590	20	619.9	24	40.96	46,402	99.20	21.11
				MQ32.hap2	627	344	17	612.6	33	40.89	45,467	99.00	20.73
QS41	A. eriantha	Shangrao City, Jiangxi Province, China	28.297126 N, 117.745972 E	QS41.hap1	643	573	21	622.6	17	40.74	45,910	99.30	21.36
				QS41.hap2	625	286	20	612.8	11	40.99	45,301	98.90	20.79
RY10	A. eriantha	Shaoguan City, Guangdong Province, China	24.674474 N, 113.056183 E	RY10.hap1	756	3,884	11	615.9	54	41.00	45,543	99.10	21.15
				RY10.hap2	676	1,367	14	593.2	43	40.79	43,704	97.30	20.78
YL01	A. eriantha	Wenzhou City, Zhejiang Province, China	27.518015 N, 119.814148 E	YL01.hap1	642	488	21	623.3	5	41.29	46,553	99.00	21.82
				YL01.hap2	613	262	19	602.9	29	40.47	45,423	98.60	20.82
YX01	A. eriantha	Sanming City, Fujian Province, China	26.165815 N, 118.197579 E	YX01.hap1	639	485	21	619.1	26	41.30	45,596	98.50	21.31
				YX01.hap2	624	198	19	615.3	22	40.74	45,589	99.20	20.33
YX02	A. eriantha	Sanming City, Fujian Province, China	26.156546 N, 118.192313 E	YX02.hap1	638	370	21	621.7	6	41.03	45,923	99.00	21.19
				YX02.hap2	615	225	20	604.2	8	40.79	44,667	98.60	20.94
HH-1	interspecific hybrid between A. chinensis var. chinensis and A. eriantha			HH-1.hap1	655	751	20.7	612.5	2	40.10	48,555	98.90	24.19
				HH-1.hap2	612	278	20.8	600.9	2	38.14	48,034	99.10	18.94

Gene family-based pangenome architecture

First, we constructed a pangenome of A. eriantha by utilizing 15 accessions from different geographical locations with 28 haplotypes/assemblies, including the 23 present de novo haplotypes and 5 previously published haplotypes/assemblies (Additional file 1: Table S4) [31, 35, 41]. Orthologs investigation allowed classification of annotated genes from the 28 assemblies into 54,770 gene families (Additional file 2: Fig. S3a). The total number of gene families increased gradually by adding an additional assembly, and approached a plateau at n = 15 (Additional file 2: Fig. S3a), indicating the pangenome reached a saturated state. Among these 54,770 gene families, 14,071 gene families (25.7%) were defined as core gene families (present in all 28 assemblies), 8,121 gene families (14.8%) as softcore families (present in 26 to 27 assemblies), 32,416 gene families (59.2%) as dispensable gene families (present in 2 to 25 assemblies), and 162 gene families (0.3%) as private gene families (present in only a single assembly) (Additional file 2: Fig. S3b).

Since an interspecific pangenome construction allows integrating the entire genetic variation across cultivated species and wild relatives [42], we subsequently add 18 A. chinensis var. chinensis haplotypes/assemblies (including hap2 of ‘HH-1’ and 17 previously published haplotypes/assemblies from 10 diploid accessions) (Additional file 1: Table S4) into the constructed pangenome of A. eriantha, creating an actually interspecific pangenome across both species [20, 22, 23, 31, 35, 41]. Phylogenetic analysis revealed that the sampled assemblies of A. eriantha and A. chinensis var. chinensis are clustered into distinct groups and those haplotypic assemblies derived from the F1 hybrid are separately clustered into their corresponding parental groups (Additional file 2: Fig. S4). All the annotated genes from the 24 accessions (i.e., 14 A. eriantha, nine A. chinensis var. chinensis and one hybrid) with 46 assemblies were classified into 56,572 gene families by orthologs analysis (Fig. 1a), slightly more than those in the pangenome of A. eriantha alone. The saturation curve analysis showed that the number of gene families ceased to increase while the number of assemblies approached 30, indicating that the sample size has reached or is beyond saturation (Fig. 1a). After incorporating A. chinensis var. chinensis genomes, we observed a reduction in core gene families to 21.4% and softcore gene families to 11.1%, alongside an increase in dispensable gene families to 67.2% and a decrease in private gene families to 0.3% (Fig. 1b). Meanwhile, we found that the exon count, CDS length, and gene expression levels of core gene families were higher than those of non-core genes (Additional file 2: Fig. S5), indicating the crucial role of core genes in plant growth and development. GO enrichment analysis revealed that core gene families are primarily involved in fundamental biological processes, such as cell wall biogenesis, anion transmembrane transport, shoot system morphogenesis, and vegetative to reproductive phase transition of meristem (Additional file 2: Fig. S6a), whereas the variable gene families are enriched in biological processes in terms of immune response and/or secondary metabolism (Additional file 2: Fig. S6b).

Fig. 1.

Construction and analysis of an interspecific pangenome for A. eriantha and A. chinensis var. chinensis accessions. a Variation of gene families in the pan-genome and core-genome with the increase in number of genome assemblies. b Compositions of the interspecific pan-genome and individual genomes. The histogram shows the number of gene families in the 46 assemblies with different frequencies. The pie chart shows the proportion of the gene family marked by each composition. c SV density across all 22 accessions. d Length distribution of SVs in the graph pangenome. e The ratio of TE-induced SVs. f The ratio of highly divergent SVs between A. chinensis var. chinensis and A. eriantha accessions

Structural variations of the assembled genomes

A reference-unbiased pangenome is crucial for identifying structural variations (SVs) and capturing the genetic diversity distributed in different accessions. Thus, we constructed a reference-unbiased pangenome using the PGGB pipeline [43], based on 44 haplotype genomes derived from 22 accessions (excluding the ‘Donghong’ and ‘wild’ accessions due to their mosaic genome assemblies) [23, 41]. The graph pangenome constructed is 2.24 Gb in size, 3.6 times as large as that of the MDHAPA genome (Additional file 1: Table S6) [31]. A total of 31,790,044 SNPs, 13,512,079 InDels (< 50 bp) and 623,478 SVs (≥ 50 bp) were identified (Fig. 1c). The size of most SVs is smaller than 2 Kb (Fig. 1d). These SVs constitute a total of 1.5 Gb of sequences, with an average of 34.1 Mb per genome. In addition, we found that approximately 48.05% of SVs overlap with TEs’ locations (Fig. 1e), suggesting that TEs might be a significant driving force for the formation of SVs. Out of these SVs, 269,784 SVs are highly differentiated between A. chinensis var. chinensis and A. eriantha (Fig. 1f), which likely represent important genomic determinants underlying the interspecific differentiation between the two species.

Construction of pan-NLRome

The nucleotide-binding domain and leucine-rich repeat (NLR) proteins have been shown to play a crucial role in plant immunity by specifically recognizing pathogen effectors [44]. To explore the relationship between NLR genes and immunity difference, we identified 7,692 NLR genes across the 46 assemblies, with significant variation in the number of NLR genes among different accessions, ranging from 130 to 210 (Fig. 2a, Additional file 1: Tables S7 and S8). On the basis of domain and motif structures [45], the identified NLRs were classified into six categories: TN (Toll/interleukin-1 receptor (TIR) and NB-ARC), CN (coiled-coil (CC) and NB-ARC), NL (NB-ARC and leucine-rich repeat (LRR)), CNL (CC, NB-ARC, and LRR), NB (NB-ARC), and TNL (TIR, NB-ARC, and LRR) (Additional file 1: Table S7). We found that the number of NLR genes in A. chinensis var. chinensis was slightly larger than in A. eriantha (Fig. 2a). In fact, larger number of N, CN, and CNL-type NLR genes in A. chinensis var. chinensis was present than in A. eriantha, whereas significantly smaller number of TNL-type NLR genes was annotated in A. chinensis var. chinensis compared to A. eriantha (Fig. S7a). If the NB group is not included, we found that the total NLR counts in A. eriantha and A. chinensis var. chinensis showed no significant difference, but that A. eriantha exhibited significantly greater variation than A. chinensis var. chinensis (Additional file 2: Fig. S7b).

Fig. 2.

Characterization of the pan-NLRome in pangenome across A. eriantha and A. chinensis var. chinensis. a Comparison of the total number of NLR genes between A. eriantha and A. chinensis var. chinensis. b The percentage of core, softcore, and dispensable NLR orthogroups in the pan-NLRome. c Comparison of the expression values among core, softcore, and dispensable NLR gene orthogroups. d Presence and absence of the NLR gene orthogroups in A. chinensis var. chinensis and/or A. eriantha. The bars in the upper panel represent the expression levels of different non-redundant NLR gene orthogroups. Red bars indicate NLR gene orthogroups exclusive to A. eriantha and absent in all accessions of A. chinensis var. chinensis

Next, we used the OrthoFinder software [46] to cluster the 7,692 NLR genes, and 99 non-redundant NLR orthogroups were identified (Additional file 1: Table S9). In the constructed pan-NLRome, core NLR orthogroups account for 9.1%, softcore NLR orthogroups for 26.3%, and dispensable NLR orthogroups for 64.6% (Fig. 2b). Then we selected leaves from 16 accessions (eight from either A. chinensis var. chinensis or A. eriantha, see materials and methods) for transcriptome sequencing. The expression levels of core, softcore, and dispensable NLR orthogroups showed no significant differences (Fig. 2c), but several NLR orthogroups belonging to the dispensable group exhibited exceptionally high expression (Fig. 2d). Most NLRs are shared by both A. eriantha and A. chinensis var. chinensis accessions, and a total of nine NLR orthogroups that are present only in A. eriantha and absent in all accessions of A. chinensis var. chinensis (Fig. 2d). In addition, genomic comparison analysis revealed that 448 out of 269,784 SVs highly differentiated between A. chinensis var. chinensis and A. eriantha are associated with 69 NLR genes (Additional file 1: Table S10).

Diverged Psa resistance between A. chinensis var. chinensis and A. eriantha and species-specific SV associated with allele-specific expression of NLR25-1 in the interspecific F1 hybrid

It has been known that the vast majority of A. chinensis var. chinensis varieties are susceptible to Psa, while A. eriantha accessions are resistant [47, 48]. To explore the genetic basis of the differential immune abilities against Psa, an interspecific hybridization was made using an A. eriantha ‘White’ [33] as female pollinated by a male A. chinensis var. chinensis ‘H0809’ [35]. A female F1 individual (named ‘HH-1’) and its parents were employed to evaluate the phenotypic performance in response to Psa-inoculation. Their leaf discs or shoots at uniformly developmental stages were inoculated with Psa race ‘ScBcH’ isolated previously from the infected field plants of A. chinensis var. chinensis cv. ‘Hongyang’ [49]. Disease symptoms appeared in leaf discs at three days post-inoculation (dpi) or shoots at 21 dpi were observed (Fig. 3a) and evaluated by measuring lesion size (Fig. 3b) or length (Fig. 3c) of the infected tissues. This analysis showed that ‘H0809’ tissues were extensively infected, while ‘White’ tissues remained almost innate immune, and their hybrid was in an intermediate state. In addition, by using leaf discs inoculation with Psa race ‘ScBcH’, we also evaluated the phenotypic performance of Psa resistance/susceptibility for all other accessions employed for pangenome construction at 3 dpi. As a result, all tested A. eriantha accessions were highly resistant to the pathogen, while the A. chinensis var. chinensis accessions displayed different magnitudes of susceptibility (Additional file 2: Fig. S8).

Fig. 3.

Structural variation (SV) in promoter led to allele-specific expression (ASE) at NLR25-1 locus in F1 hybrid between A. eriantha ‘White’ and A. chinensis var. chinensis ‘H0809’. a-c The phenotypic symptom (a) and quantification through measuring lesion area in leaf discs (b) or lesion length in shoots (c) of ‘White’, ‘H0809’, and ‘HH-1’ at 3 or 21 dpi with Psa. Each data point represents the mean (± SD) of three replications, and asterisk refers to significant difference (Student’s t-test; **: p < 0.01; ***: p < 0.001). d The gene counts of allele-specific expression genes (ASEGs) of ‘HH-1’ upon Psa-inoculation at different hpi. ASEG1s or ASEG2s indicate gene counts that are preferentially expressed in HH-1.hap1 or HH-1.hap2. The blue or pink bar represents non-pathogen-resistant or pathogen-resistant genes in ASEGs, respectively. e Expression profiles of 87 allelic gene pairs with ASE pattern in individual haplotypes of ‘HH-1’ upon Psa-inoculation. f Expression patterns of NLR25-1 in ‘HH-1’ and its parents (g) PCR validation of allelic structure variation. h Diagram of structural variation at NLR25-1 locus of ‘HH-1’. AeNLR25-1 is derived from HH-1.hap1 (inherited from A. eriantha ‘White’), while AcNLR25-1 is derived from HH-1.hap2 (inherited from A. chinensis var. chinensis ‘H0809’). Blue indicates syntenic regions upstream the start codon (ATG), while yellow represents the structural variation (SV) derived from individual haplotypes. The top or bottom black solid lines represent upstream promoter sequences that are designated as ProAeNLR25-1 or ProAcNLR25-1, respectively. The red, blue, green, and dark green vertical lines on the black horizontal lines corresponding to the conserved binding motifs recognized by WRKY (binding to TGACC/T), MYB (binding to TAACTG/CAACTG), bHLH (binding to CAGGTG/CACGYG), and AP2 (binding to GCCGCC/ACCGC/CCGAC) transcription factors, respectively. All the experiments and measurements have been independently performed with three biological replicates and three technical replicates

Incorporating beneficial alleles or haplotypes from wild relatives into breeding program is a crucial step for crop improvement [50]. To this end, identification of allele-specific expression (ASE, referring to the preferential expression of a parental allele in the hybrid due to variations in regulatory sequences from the parental genomes) is essential [51, 52]. To explore ASE in the A. eriantha × A. chinensis var. chinensis F1 hybrid, we performed transcriptome sequencing on the leaves of ‘HH-1’ at 0-, 12-, and 72-h post-inoculation (hpi) with Psa race ‘ScBcH’ (Additional file 1: Tables S11 and S12). As compared to the control (0 hpi), 24,361 (11,932 for HH-1.hap1 and 12,429 for HH-1.hap2) and 28,694 (14,047 for HH-1.hap1 and 14,647 for HH-1.hap2) differentially expressed genes (DEGs) were identified in samples at 12 and 72 hpi (Additional file 1: Tables S13 and S14), respectively. Using JCVI [53], 27,553 pairs of allelic genes were identified between HH-1.hap1and HH-1.hap2 (Additional file 1: Tables S15 and S16). A total of 8,649, 9,546 and 9,717 pairs of significant ASE genes (ASEGs) were filtered in samples at 0, 12 and 72 hpi, respectively (Additional file 1: Tables S17, S18 and S19). Subsequently, 3,139 or 2,985 pathogen-resistant genes including salicylic acid pathway related genes and immune receptor-like genes (Additional file 1: Table S20) were annotated in HH-1.hap1 or HH-1.hap2 utilizing DRAGO and HMMER respectively [54, 55]. Among ASEGs, 3,999, 4,520 and 4,589 ASEGs were biased expressors of HH-1.hap1 derived from A. eriantha (designated as ASEG1) at 0, 12 and 72 hpi, respectively, including 321, 364 and 340 pathogen-resistant genes (Fig. 3d, Additional file 1: Table S21). And 62 allelic pathogen resistant gene pairs with a significant up-regulation after Psa-inoculation and a continuous expression bias to HH-1.hap1 were identified (Fig. 3e, Additional file 1: Table S22). Meanwhile, 4,650, 5,026 and 5,128 gene pairs were biased expressors of HH-1.hap2 derived from A. chinensis var. chinensis (designated as ASEG2) at 0, 12 and 72 hpi, respectively, including 348, 377 and 377 pathogen-resistant genes (Fig. 3d, Additional file 1: Table S23). And 25 allelic pathogen-resistant gene pairs with a significant up-regulation after the Psa inoculation and a continuous expression bias to HH-1.hap2 were identified (Fig. 3e, Additional file 1: Table S24). Similar analyses were conducted on the parental tissues to explore the differently expressed orthologous genes at 0, 12 and 72 hpi with Psa-inoculation (Additional file 1: Tables S25, S26 and S27). Consistently, transcriptomic profile comparison between the parental lines showed that expression of over 70% ASEG1 orthologs in ‘White’ was significantly higher than that of ‘H0809’ and that expression of over 70% ASEG2 orthologs in ‘H0809’ was significantly higher than that of ‘White’ (Additional file 2: Fig. S9).

Out of ASEG1s with a significant up-regulation after the Psa-inoculation, a CNL member from the NLR gene family, designated as NLR25-1 located on chromosome 25 with a pair of allelic genes (AeNLR25-1/Aisf1h1a25g408720 and AcNLR25-1/Aisf1h1b25g410370), was resident in an interspecies-specific SV (Table S10) and showed extremely biased expression toward HH-1.hap1 in hybrid ‘HH-1’ (Fig. 3f, Additional file 1: Table S16). To validate the allele-specific expression pattern at the NLR25-1 locus, qRT-PCR analysis was performed on the Psa-inoculated leaves of ‘HH-1’, ‘H0809’ and ‘White’ at 0, 12 or 72 hpi. Consistently, abundant AeNLR25-1 transcripts were detected at 0 hpi and its expression was strongly reinforced at 12 and 72 hpi, while AcNLR25-1 transcripts were hardly detected either before or after Psa-inoculation (Fig. 3f). However, both AeNLR25-1 and AcNLR25-1 encode a protein with identical size of 1040 amino acids, and thirty-four SNPs are present across the allelic coding sequences, leading to 20 amino acid substitutions (Additional file 2: Fig. S10).

To dissect the mechanism underlying the allele-specific expression at the NLR25-1 locus, our subsequent investigation focused on the structure variations in their promoter regions. Consistent with the SVs detected by pangenome analysis (Additional file 1: Table S10), sequence comparison between the allelic regions revealed a conserved interspecific SV resident in the upstream regulatory region at the NLR25-1 locus, i.e. unique fragments ranging from 3,409 to 3,720 bp were consistently identified at 46 bp upstream of the AeNLR25-1 gene's translation start site across individual A. eriantha accessions, whereas this region was replaced by distinct sequences ranging from 549 to 622 bp in individual A. chinensis var. chinensis accessions, creating a significant interspecies divergence in the NLR25-1 promoter architecture (Additional file 2: Fig. S11a). In accordance with pangenome analysis, this difference is confirmed in multiple A. chinensis var. chinensis and A. eriantha accessions through PCR amplification and agarose gel electrophoresis, in which all A. eriantha accessions showed a larger amplified fragment than from A. chinensis var. chinensis (Fig. 3g). More specifically, a unique 3,523 bp sequence (namely AeSV) was identified at 46 bp upstream of the AeNLR25-1 translation start site in ‘White’ and HH-1.hap1 from ‘HH-1’, whereas this region was replaced by a distinct 601 bp fragment (namely AcSV) in ‘H0809’ and HH-1.hap2 from ‘HH-1’ (Fig. 3h). Further investigation revealed that the sequence divergence is primarily due to the insertion of six additional transposable elements (TEs), including one LTR and five terminal inverted repeats, in the promoter region of AeNLR25-1 compared to that of AcNLR25-1 with a single terminal inverted repeat (Additional file 2: Fig. S11b).

NLR25-1 functions in immune response against Psa

To unravel the biological function of NLR25-1, CRISPR-cas9-mediated mutants of ‘HH-1’ were constructed by Agrobacterium-mediated genetic transformation, and three independent NLR25-1-knockout lines (nlr25-1–1/nlr25-1–2/nlr25-1–3) were obtained (Additional file 2: Fig. S12). Then leaf discs, tissue culture seedlings with uniform developmental stage, or 5-leaf-age potted seedlings derived from wild-type ‘HH-1’ (WT) or three mutants were inoculated with Psa race ‘ScBcH’. Increased lesion area was observed on leaf discs of mutant lines compared to WT at 3 dpi (Additional file 2: Fig. S13a and S13b). The average in planta bacterial growth of Psa bacteria in tissue culture seedlings was higher in mutants compared to WT at 7 dpi (Additional file 2: Fig. S13c and S13d). The potted seedlings of mutant lines displayed a severe wilting phenotype with almost entire leaf areas of expanded lesion compared to WT that exhibited mild wilting symptoms with restricted area of lesion expansion (Fig. 4a), and the average in planta bacterial growth in WT was less than in the mutants at 7 dpi (Fig. 4b). NBT staining, DAB staining [56, 57], as well as O₂⁻ and H₂O₂ quantification were conducted on leaf samples of the potted seedlings at 1 dpi to detect the immunity performance, and the results showed that the immunity responses of mutants were significantly weaker than WT (Fig. 4c-e). In addition, to further assess the disease symptom, leaf samples of the potted seedlings at 7 dpi were collected and subjected to physiological index measurement. Trypan blue staining, NBT staining and DAB staining [56, 57] was conducted, and the results showed that dead cells’ number and hydrogen peroxide content were significantly increased in mutant leaves (Fig. 4c). Consistently, the average contents of O₂⁻, H₂O₂ or MDA significantly increased in mutant lines compared to WT (Fig. 4d-f), and the activities of defense enzymes including catalase (CAT), peroxidase (POD) or superoxide dismutase (SOD) in WT were significantly higher than mutants (Additional file 2: Fig. S13e-g). These results indicated that the nlr25-1 mutants displayed more severe cellular damage (necrosis) resulting from Psa-mediated abiotic stress than WT. Meanwhile, the expression of defense-related genes from salicylic acid (SA) pathway including homologs of PR1 (Aisf1h1a19g304391/Aisf1h1b19g310001, Additional file 2: Fig. S13h), TGA9 (Aisf1h1a12g195220/Aisf1h1b12g198870, Additional file 2: Fig. S13i), PR9 (Aisf1h1a28g469250/Aisf1h1b28g469530, Additional file 2: Fig. S13j) and NPR1 (Aisf1h1a09g139020/Aisf1h1b09g139210, Additional file 2: Fig. S13k) was detected at 0 or 72 hpi through qRT-PCR analysis, and the result showed that all the tested genes were significantly induced and the reinforced expression magnitude in WT was much higher than in mutant plants. Taken together, these results suggested that NLR25-1 plays a vital role in resistance to Psa.

Fig. 4.

Functional characterization of NLR25-1 using knockout mutants or overexpression transgenics in responses to Pseudomonas syringae pv. actinidiae (Psa) inoculation. a Phenotypic comparison of ‘HH-1’ and its CRISPR-cas9-mediated NLR25-1-knockout plants in potted seedlings at 7 dpi (days post inoculation) with Psa (‘ScBcH’) (lower panel). The upper panel represents control without Psa-inoculation. b The in planta bacterial growth measured for the Psa-inoculated seedlings in (a). c Staining of leaves from Psa-inoculated seedlings at 1 or 7 dpi with trypan blue to assess cell death, or NBT and DAB to assess superoxide production, respectively. The depth of color represents the quantity or content. d-f Physiological changes measured for the Psa-inoculated seedlings at 1 or 7 dpi, including superoxide (O₂⁻) content (d), hydrogen peroxide (H₂O₂) content (e), malondialdehyde (MDA) content (f). g Phenotypic comparison of ‘H0809’ and NLR25-1-OE plants in potted seedlings at 7 dpi with Psa (‘ScBcH’) (lower panel). The upper panel represents control without Psa-inoculation. h The in planta bacterial growth measured for the Psa-inoculated seedlings in (g). i Staining of leaves from Psa-inoculated seedlings at 1 or 7 dpi with Trypan blue to assess cell death, or NBT and DAB to assess superoxide production, respectively. The depth of color represents the quantity or content. j-l Physiological changes measured for the Psa-inoculated seedlings at 1 or 7 dpi, including superoxide (O₂⁻) content (j), hydrogen peroxide (H₂O₂) content (k), malondialdehyde (MDA) content (l). Each data point represents the mean (± SD) of three replications. Different letters on error bar referred to significant difference based on one-way ANOVA and Tukey's test (p < 0.05). All the experiments and measurements have been independently performed with three biological replicates and three technical replicates

To further assess the immune activity of NLR25-1, three independent transgenic lines constitutively overexpressing AeNLR25-1 (OE1/OE2/OE3) were obtained through Agrobacterium-mediated genetic transformation using ‘H0809’ as transgene host (Additional file 2: Fig. S14a) [58]. Upon Psa inoculation, leaf discs at 3 dpi showed reduced lesion area in three OE lines compared to WT (Additional file 2: Fig. S14b and S14c). Similarly, in planta bacterial growth of Psa bacteria for tissue culture seedlings or potted seedlings in OE lines at 7 dpi were significantly lower than that of WT (Fig. 4g, h, Additional file 2: Fig. S13d and S13e). NBT staining, DAB staining, as well as O₂⁻ and H₂O₂ quantification were conducted on leaf samples of the potted seedlings at 1 dpi to measure the immunity response, and the results showed that the immunity responses of OE lines were stronger than WT (Fig. 4i-k). Disease severity was further accessed through physiological index measurement for potted seedlings at 7 dpi. Dead cells and hydrogen peroxide were reduced in OE leaves compared to WT (Fig. 4i). The contents of O₂⁻, H₂O₂ and MDA in WT were significantly higher than that in OE lines (Fig. 4j-l), while the activities of CAT, POD and SOD in WT were significantly lower than that in OE lines (Additional file 2: Fig. S14f-h). Consistently, the defense genes’ transcription levels were significantly higher in OE plants after Psa inoculation (Additional file 2: Fig. S14i-l). These results further suggest a crucial role conferred by AeNLR25-1 in resistance to Psa.

To provide more evidence supporting that the SV-associated AeNLR25-1 confers the resistance to Psa, a backcross population (BCF1) was constructed using ‘HH-1’ as female pollinated by the male parent ‘H0809’. Leaf tissues from 120 BCF1 offspring were challenged by the Psa infection and subjected to phenotypic assessments (Additional file 2: Fig. S15a). To determine the genotype of the tested BCF1 plants, a pair of distinct PCR primers was designed to distinguish the allelic difference at the NLR25-1 locus (Additional file 2: Fig. S15b). Genotyping was then conducted on the 120 plants, resulting in 44 heterozygotes (AeNLR25-1/AcNLR25-1) and 76 homozygotes (AcNLR25-1/AcNLR25-1) (Additional file 2: Fig. S15c). The phenotypic analysis showed that the average lesion area of homozygous or heterozygous plants was 12.10% or 7.32% (Additional file 2: Fig. S15d), respectively, indicating that the heterozygous plants were more resistant to Psa-inoculation and further supporting the immune function by AeNLR25-1 preferential expression in response to Psa-inoculation.

Both AeWRKY75 and AcWRKY75 specifically bind to AeSV and activate AeNLR25-1 transcription and immune responses

To study how the identified SV affects the activity of promoter of AeNLR25-1 (ProAeNLR25-1) or AcNLR25-1 (ProAcNLR25-1), a GUS staining assay was conducted in ‘HH-1’ leaves by Agrobacterium-mediated infiltration, and the result indicated that the activity of ProAeNLR25-1 was much stronger than ProAcNLR25-1 (Fig. 5a). These results are consistent with the transcriptome profile dataset and qRT-PCR analysis (Fig. 3e and f).

Fig. 5.

AeWRKY75/AcWRKY75 specifically binds to and activates the promoter of AeNLR25-1 and confers resistance to Pseudomonas syringae pv. actinidiae (Psa). a GUS staining assay validating promoter activity at the NLR25-1 locus by detecting differential GUS activities in ‘HH-1’ leaves transiently expressing different constructs. The expression level in ProAeNLR25-1::GUS was designated as “1”. b Dual-LUC assay validating activity of AeWRKY75/AcWRKY75 by activating AeProNLR25-1/AcProNLR25-1 in Nicotiana benthamiana leaf (**: p < 0.01; ns: not significant). c Y1H assay showing binding ability of AeWRKY75/AcWRKY75 to the promoters (ProAeNLR25-1/ProAcNLR25-1). The putative binding sites of WRKY family transcription factors on truncated segments (AeP2.1, AeP2.2 and AeP2.3) are marked by red vertical lines. The blue colony in lower panel shows the ability of binding. d The electrophoretic mobility shift assay (EMSA) validating binding ability of AeWRKY75 and AcWRKY75 to AeSV/ProAeNLR25-1. e The expression alteration of AeNLR25-1/AcNLR25-1 in ‘HH-1’ or WRKY75-knockout plants with Psa-inoculation. f Phenotypic comparison of ‘HH-1’ and WRKY75-knockout plants (potted seedlings) at 7 days post inoculation (dpi) (lower panel). The upper panel represents control without Psa-inoculation. g The in planta bacterial growth measured for the Psa-inoculated plants in (f). h Staining of leaves from Psa-inoculated seedlings at 1 or 7 dpi with trypan blue to assess cell death, or NBT and DAB to assess superoxide production, respectively. The depth of color represents the quantity or content. i-k Physiological changes in ‘HH-1’ and mutant plants at 1 or 7 dpi, including superoxide (O₂⁻) content (i), hydrogen peroxide (H₂O₂) content (j), malondialdehyde (MDA) content (k). Each data point represents the mean (± SD) of three replications. Different letters on error bar referred to significant difference based on one-way ANOVA and Tukey's test (p < 0.05). All the experiments and measurements have been independently performed with three biological replicates and three technical replicates

To explore the underlying mechanism of differential activities between the allelic promoters (ProAeNLR25-1/ProAcNLR25-1), a total of 2,632 or 2,569 transcription factors were predicted in HH-1.hap1 or HH-1.hap2, respectively (Additional file 1: Table S28), and the transcriptome profiles with 5,201 annotated transcription factors in ‘HH-1’ genome were inspected. In ‘HH-1’ samples at 12 hpi with Psa when AeNLR25-1 expressed at the most abundant level (Fig. 3f), we found a total of 841 transcription factors were significantly upregulated (Additional file 1: Table S29) and enriched in WRKY transcription factor family (Additional file 2: Fig. S16). The twelve WRKY members with top Pearson correlation coefficient in expression with NLR25-1 were screened as candidates (Additional file 1: Table S30) and validated utilizing Dual-LUC assay, and one of them, namely AeWRKY75 (Aisf1h1a29g472580), was eventually verified to be able to activate the promoter of AeNLR25-1 (Additional file 2: Fig. S17).

To demonstrate the function of WRKY75 homologs in Psa resistance, the allelic gene of AeWRKY75, namely AcWRKY75 (Aisf1h1b29g473170) was also identified and isolated (Additional file 2: Fig. S18a and S18b). Moreover, combined with the transcriptome profile analysis and qRT-PCR analysis, without Psa-inoculation, both AeWRKY75 and AcWRKY75 showed similarly low level of expression. By contrast, both AeWRKY75 and AcWRKY75 transcripts were abundantly induced upon Psa-inoculation (Additional file 2: Fig. S18c). Next, dual-luciferase reporter assay suggested either AeWRKY75 or AcWRKY75 is able to activate the AeNLR25-1 promoter containing AeSV but neither of them can activate the AcNLR25-1 promoter containing AcSV (Fig. 5b). Yeast-one-hybrid (Y1H) assay indicated that the binding site was present in the AeP2.3 fragment derived from AeSV (Fig. 5c). The electrophoresis mobility shift assay (EMSA) showed a W-box motif (TGAC/GTCA) in the AeP2.3 fragment is the precise binding site (namely AeW-box) (Fig. 5d). Although autoactivation was observed on AeP1 and AcP1, further EMSA assays ruled out the possibility of additional functional binding site(s) available for WRKY75 binding (Additional file 2: Fig. S19). These results suggest that both AeWRKY75 and AcWRKY75 specifically binds to AeW-box in AeSV/ProAeNLR25-1 to activate AeNLR25-1 expression, while the AcNLR25-1 cannot be activated due to the ProAcNLR25-1 lacking the corresponding cis-element.

To validate the possible function of WRKY75 in Psa resistance, CRISPR-cas9-mediated mutants of ‘HH-1’ (AeWRKY75/AcWRKY75) were constructed by Agrobacterium-mediated genetic transformation [58], and three independent knockout lines (wrky75-1/2/3) were obtained (Additional file 2: Fig. S20). Then, we evaluated the expression level of NLR25-1 in wrky75 mutants. qRT-PCR analysis showed that the expression level of AeNLR25-1 was significantly lower in mutant plants than WT without Psa-inoculation (Fig. 5e). Moreover, the expression of AeNLR25-1 was significantly induced upon Psa-inoculation in both WT and mutants, but the increased level in mutant plants was much lower than WT. By contrast, the expression level of AcNLR25-1 in wrky75-1/2/3 mutants and WT plants remained low level or unchanged before or after Psa-inoculation. These together suggest that AeWRKY75/AcWRKY75 plays a critical role in Psa resistance probably via binding to AeW-box and activating AeNLR25-1 expression.

Meanwhile, phenotypic evaluation showed larger lesion area was observed on leaf discs of mutants than WT upon Psa-inoculation at 3 dpi (Additional file 2: Fig. S21a and S21b). The in planta bacterial growth in the treated seedlings is also higher in mutants than WT at 7 dpi (Fig. 5f, g, Additional file 2: Fig. S21c and S21d). NBT staining, DAB staining, as well as O₂⁻ and H₂O₂ quantification were conducted on leaf samples of the potted seedlings at 1 dpi to detect the immunity response, and the results showed that the immunity responses of mutants were weaker than WT (Fig. 5h-j). Disease severity was further accessed through physiological index measurement for potted seedlings at 7 dpi. The dead cells and hydrogen peroxide were significantly increased in mutant leaves compared to WT (Fig. 5h). We also measured the contents of O₂⁻, H₂O₂ and MDA or activities of POD, SOD and CAT. The results indicated that the average contents of O₂⁻, H₂O₂ and MDA in the mutants were significantly increased compared to WT (Fig. 5i-k), while the average activities of CAT, POD, and SOD in mutants were significantly decreased compared to WT (Additional file 2: Fig. S21e-g). In addition, the expression of defense-related genes was also tested. All the tested genes were expressed at a low level in both WT and the three mutant lines at 0 hpi, while their expression was substantially induced at 72 hpi, but the induction was substantially inhibited by AeWRKY75/AcWRKY75-defective mutations compared to WT (Additional file 2: Fig. S21h-k).

Discussion

Despite the protracted process that canonical crops underwent from wild forms to domesticated forms, accumulating evidence suggests that the rapid reduction in diversity of modern cultivated varieties is a result of modern breeding and cultivation practices [59–61]. This is exactly the case for kiwifruit whose domestication only began in the early twentieth century and no more than 100 years of successive rounds of artificial selection has generated numerous distinctive cultivars, forming a huge kiwifruit industry worldwide [1]. However, after decades of intensive cultivation, these domesticates are highly susceptible to canker pathogens of Psa. In the present study, a reference-unbiased pangenome construction allowed identification of 623,487 SVs. Our genomic data indicates that TE-induced genome size expansion in the wild A. eriantha accessions may also contribute to the increased SVs number and genomic diversity. We combined pangenome data with transcriptome analysis of interspecific samples between the cultivated individual and the wild relative to discover the association between many allele-specific expression (ASE) gene loci and SVs. What is particularly interesting is the drastic changes in expression patterns caused by the highly differentiated regulatory regions of disease resistant genes.

To confront the distinct defense abilities against Psa between the cultivated A. chinensis var. chinensis and wild A. eriantha, we characterized the ASE profile of all annotated NLR genes that can be paired one on one. A total of 76 pairs of allelic NLR genes were matched, and 52 pairs were validated as ASE genes for at least one timepoint among 0, 12 or 72 hpi in our transcriptomic data (Additional file 1: Table S31). Considering the abundance of structure variations between the two species, the upstream 5,000 bp of each pair of allelic NLR genes were retrieved and aligned. Thirty-one pairs of allelic NLR genes showed low identity in upstream region, while 20 or 17 pairs showed small or large segment alignment, and 8 pairs showed a highly matched upstream sequence similarity. The proportion of ASE genes increased with the difference of upstream sequences enlarged (Additional file 2: Fig. S22). Further investigation showed that the large variations were owing to multiple TE insertions (Additional file 1: Table S31).

We conducted a detailed genetic and molecular analysis of this change at the NLR25-1 locus that was initially identified through transcriptome profiling and ASE analysis (Fig. 3e-f; Additional file 1: Table S16). Our analysis indicated that it is a typical ASE gene and its regulatory region was associated with a TE-induced SV whose length and structure are highly conserved within parental accessions. We provide evidence showing the wild allele AeNLR25-1 is predominantly expressed in the hybrid and confers resistance to Psa. We discovered that the preferential expression of AeNLR25-1 in response to Psa-inoculation is owing to the presence of a unique A. eriantha-specific AeW-box (TGAC/GTCA) that could be bound and activated specifically by WRKY transcription factors AeWRKY75 or AcWRKY75.

To inspect the specificity of the conserved structure variation at the NLR25-1 locus, a total of 75 sequenced accessions of various Actinidia species were examined including all 46 accessions in the interspecific pan-genomics in this study, as well as 29 accessions retrieved from Kiwifruit Pan-Genome Database (KPGD) [62] consisting of two A. eriantha, one A. chinensis var. chinensis, nine A. arguta, six A. chinensis var. deliciosa, one A. hemsleyana, one A. latifolia, one A. longicarpa, one A. macrosperma, two A. polygama, one A. reticulata, two A. rufa and two A. zhejiangensis (Additional file 1: Table S32). We found only a single allele at the NLR25-1 locus was missing in A. chinensis var. deliciosa haplotype ‘MeiweiW2.hap5’ and A. chinensis var. chinensis haplotype ‘Jinpai.hap2’, suggesting a highly conserved existence of the NLR25-1 locus. Importantly, the AeW-box cis-element is exclusively present in all tested A. eriantha assemblies and an A. zhejiangensis haplotype ‘ZhejiangW1.hap1’, which has been postulated as a hybrid derived from a cross between A. eriantha and A. hemsleyana [63], demonstrating the specificity of AeW-box cis-element uniquely existing in A. eriantha (Additional file 2: Fig. S23, Additional file 1: Table S32). Interestingly, a mutation on the translation start site of the NLR25-1 locus occurs in five accessions including A. arguta accessions ‘Longcheng2V1.hap3’ and ‘RuanzaoW2.hap1’, A. chinensis var. deliciosa accession ‘MeiweiW2.hap1’, A. hemsleyana accession ‘ChangyeW1’ and A. zhejiangensis ‘ZhejiangW1.hap2’, probably resulting in a delayed translation initiation at 88 bp, causing a skip of 29 amino acid in the translation products. Furthermore, all A. arguta accession consists of one or two duplicated segments sized 81 bp at the NLR25-1 locus (Additional file 2: Fig. S24 and S25, Additional file 1: Table S32). However, these structural variations do not alter regions coding for either CC, NB-ARC or LRR domains at the NLR25-1 locus.

It is worth noting that although NLR25-1 gene was not detected in haplotype 2 of A. chinensis var. chinensis ‘Jinpai’, two identical NLR25-1 genes were detected in haplotype 1. Further analysis revealed an asymmetric recombination wherein a large fragment containing an entire NLR25-1 locus was probably translocated from haplotype 2 into haplotype 1 (Additional file 2: Fig. S26a). A Copia LTR resides in the junction point of the recombination, which probably contributed to the translocation (Additional file 2: Fig. S26b). Intriguingly, when validating this recombination via PCR amplification across the junction point in various A. chinensis var. chinensis accessions, an unexpected band appeared for A. chinensis var. chinensis cv. ‘Biyu’ (Additional file 2: Fig. S26c). Further analysis revealed a similar junction point in haplotype 2 of ‘Biyu’, suggesting a symmetric recombination might result in an inversion (Additional file 2: Fig. S26a and S26d). As the same junction point appeared in different types of structure variations, it could be assumed that this Copia LTR is relatively active. This may suggest a possible causal agent for the expansion or contraction of pathogen-resistant genes, while also enlightening an approach to enhance resistance utilizing natural source of duplicated pathogen-resistant gene loci.

Taken together, our work demonstrates that there are a large number of TE-mediated structural variations and cis-regulatory elements useful for resistance to canker disease in wild kiwifruit germplasm resources, highlighting that once these structural variations are transferred from wild kiwifruit to cultivated varieties, their cis-regulatory elements can be activated and utilized by trans-factors derived from both wild and cultivated ones.

Materials and methods

Genome sequencing and assembly

To comprehensively represent the genetic diversity of kiwifruit, a total of eleven samples of A. eriantha representing its natural distribution areas and an interspecific hybrid F1 individual (named ‘HH-1’) between an A. eriantha accession ‘White’ [33] and an A. chinensis var. chinensis accession ‘H0809’ [35] were selected for de novo genome assembly. The 12 newly sequenced plants were grown at the farm of Anhui Agricultural University. Fresh and healthy leaves were harvested upon sprouting, promptly frozen in liquid nitrogen, and subsequently used to construct HiFi SMRTbell libraries. The libraries were then sequenced on the PacBio Revio platform using the Circular Consensus Sequencing (CCS) model. Then the HiFi reads of each accession were assembled to contigs of two haplotypes using hifiasm (v0.16.1) [36] with the default parameters. Employing the quarTeT (v1.2.4) [37], the contigs were oriented and anchored to chromosomes by the reference-guided strategy using the “HY4P” (A. chinensis var. chinensis) and “MDHAPA” (A. eriantha) genomes as references [22, 31], respectively. Then the completeness of assemblies was evaluated using BUSCO (v3.0.2) [64] with the Embryophyta odb10 dataset. The continuity was assessed using the LTR Assembly Index (LAI) [38].

Genome annotation

The transposable elements (TEs) were annotated using the Extensive De Novo TE Annotator (EDTA) pipeline [39] with parameters of "–anno 1 –force 1". The protein-coding genes were annotated using the BRAKER3 pipeline [65] through a combination of de novo prediction, homology prediction and transcript evidence. Only those genes that contained both start and stop codons and were at least 100 amino acids (aa) in length were reserved. The function of all protein-coding genes was annotated by eggNOG-mapper [66] against a series of protein sequence databases to obtain the functional description, Gene Ontology (GO) numbers, and PFAM domains.

Gene family-based pangenome construction

By integrating five previously published kiwifruit assemblies, we constructed a pangenome of A. eriantha comprising 28 assemblies derived from 15 accessions (Additional file 1: Table S4). Ortholog groups among the 28 assemblies were identified using OrthoFinder [46] with default parameters (-M msa -S diamond -A mafft -T fasttree -I 1.2). Based on the clustering analysis, the resulting families were divided into core (those gene families shared across all 28 assemblies), softcore (those presence > 90% of all accessions, 26–27), dispensable (those present in 2 to 25 assemblies) and private gene families (present in only one assembly). Similarly, all genes from the 46 assemblies were clustered into gene families using the OrthoFinder software [46]. The resulting families were divided into core (those shared across all 46 assemblies), softcore (those presence > 90% of all assemblies, 42–45), dispensable (those shared in 2 to 41 assemblies) and private gene families (those existing in only one assembly).

Detection of genomic variations

A graph-based, reference-unbiased pangenome was constructed using 44 haplotype genomes derived from 22 accessions (comprising thirteen A. eriantha, eight A. chinensis var. chinensis, and one F1 hybrid). The process involved extracting chromosome sequences from all assemblies and renaming them in accordance with PanSN specification rules (e.g., “MD31#1#Chr02” denotes Chromosome 2 on Haplotype 1 of the ‘MD31’ accession). Following the Pangenome Graph Builder (PGGB v0.7.4) pipeline [43], we initially partitioned all chromosomal sequences into 29 communities, wherein each community represented one chromosome, and graph-based pangenomes were constructed for each chromosome. The graph construction process followed a standard pipeline: aligning assemblies with wfmash (v0.14.0); generating an initial graph from alignments using seqwish (v0.7.11); refining the graph with smoothxg (v0.8.2) and gfaffix (v0.2.1); finally analyzing and visualizing with odgi (v0.9.2). The key parameters for constructing the graph pangenome were set as “-s 50,000 -p 90 -k 23 -n 44”, and the variants were decomposing using MD31.hap1 as reference (-V “MD31#1:1000”). The odgi stats command were used to produce the final statistics of the graph. Based on the VCF file generated from the PGGB, we categorized all variants into three types: SNPs, InDels (< 50 bp), and SVs (≥ 50 bp). Then, SVs with fully discordant genotypes between A. chinensis var. chinensis and A. eriantha accessions, termed highly differentiated SVs, were selected for downstream analysis. The annotated SVs’ impact was detected using the SnpEff software [67].

Transcriptome sequencing and data analysis

To evaluate the impact of structural variations (SVs) on gene expression, we performed transcriptome sequencing using young leaves from 16 accessions, including eight A. eriantha accessions (MD31, CY38, DG01, GZ26, HN01, MQ32, QS41, YX02) and eight A. chinensis var. chinensis accessions (Biyu, Donghong, Hongyang, Hort16A, Huangyang, Jinmi, Jinpai, Zps). Three biological replicates were sequenced for each sample. Total RNA was extracted using TIANGEN RNAprep Pure Kit. And then mRNA libraries were constructed and sequenced using the NovoSeq 6000 platform of illumina. The clean reads were mapped to ‘MDHAPA’ genome [31] using the hisat2 [68] with default parameter. Transcripts per million (TPM) values were calculated using featureCounts [69]. The differential expression analysis was performed using the DESeq2 package [70] in R and the screening criteria was set as ‘| log₂(fold change) |> 1 and the p-value < 0.05’.

Pan-NLRome construction and analysis

Genes containing the NB-ARC domain were identified by pfamscan (https://github.com/aziele/pfam_scan) with E-value < 1e-5 as candidate NLRs. Then the candidate NLRs were validated and classified by NLR-annotator [71]. Only genes containing the NB-ARC domain were retained as the final NLR genes. Based on the distribution of NB-ARC, Toll/interleukin-1 receptor (TIR), coiled-coil (CC), and leucine-rich repeat (LRR) domains within each protein, these members were classified into six subgroups including TN (TIR and NB-ARC), CN (CC and NB-ARC), NL (NB-ARC and LRR), CNL (CC, NB-ARC, and LRR), NB (NB-ARC), and TNL (TIR, NB-ARC, and LRR). The number of members in the six NLR subgroups was then compared between A. chinensis var. chinensis and A. eriantha. We clustered NLR genes from all 46 assemblies using OrthoFinder [46] with default parameters (-M msa -S diamond -A mafft -T fasttree -I 1.2) to obtain non-redundant NLR orthogroups. Based on the distribution of each non-redundant NLR orthogroup across all assemblies, they were categorized into core (those shared in all 45 assemblies), soft-core (those presence > 90% of all assemblies, 41–45), dispensable (those shared in 2 to 40 assemblies), and private (those existing in only one assembly) NLR orthogroups. The presence-absence patterns of each non-redundant NLR orthogroup across the 46 assemblies were visualized using a heatmap. Using the previously sequenced leaf transcriptomes from 16 accessions, we aligned the reads to their respective reference genomes and quantified NLR gene expression levels. Expression levels of the 99 non-redundant NLR orthogroups were calculated as the mean TPM value of all NLR genes within each orthogroup.

Transcriptome sequencing and allele-specific expression (ASE) analysis

For Pseudomonas syringae pv. actinidiae (Psa) inoculation, five-leaf-stage kiwifruit seedlings were needle-punctured and immersed in bacterial suspension (OD₆₀₀ = 1.0, 0.025% Silwet L-77) for 10 min. The inoculated seedlings were maintained in a controlled-environment chamber with the following parameters: 16/8 h photoperiod, 18 °C/14 °C day/night temperature regime, and 95% relative humidity. Leaves of ‘HH-1’, ‘White’ and ‘H0809’ were collected at 0-, 12- and 72-hour-post-infection (hpi) for transcriptome sequencing with at least three biological replicates. All samples were frozen in liquid nitrogen and stored at −80℃. Then total RNA was isolated and purified from sampled tissues and sequenced by Illumina NovoSeq 6000 platform. The RNA-seq raw reads retrieved from ‘HH-1’ samples (0, 12 and 72 hpi) were mapped against allele-aware annotated gene models according to previous procedure [40], and that from ‘White’ or ‘H0809’ samples were aligned to HH-1.hap1 or HH-1.hap2 gene models respectively using hisat2 (v2.1.0) software [68]. The alignment was counted by featureCounts (v2.0.6) software [69] with additional option (-M –fraction) to infer allele-aware expression. More specifically, the reads that contain haplotype-specific SNP were assigned to corresponding haplotype, and reads without SNP were averagely distributed to two haplotypes. The read counts were normalized by the sequencing depth and the transcript length to transcript per kilobase of exon model per million mapped reads (TPM).

The pathogen-resistant genes were annotated though DRAGO and HMMER [54, 55], and the allelic gene pairs were identified utilizing ‘jcvi.compara.catalog ortholog’ in JCVI library [53]. The transcription factors were identified via PlantTFDB [72]. The differentially expressed genes (DEGs) between 0 and 12 hpi or 0 hpi and 72 hpi sample for ‘HH-1’, ‘White’ or ‘H0809’, the allele-specific expression genes (ASEGs) in 0 hpi, 12 hpi or 72 hpi sample of ‘HH-1’, as well as DEGs between ‘White’ and ‘H0809’ at 0 hpi, 12 hpi or 72 hpi sample were identified using DESeq2 (v1.42.0) software [70] with a standard of | Log₂FoldChange |> 0.5, and adjusted P-value < 1e-5. HH-1.hap1 was used as reference when conducting ortholog comparison. Transcription factors enrichment was conducted utilizing clusterProfiler package [73]. Co-expression transcription factors of up-regulated genes were filtered with a standard of Pearson correlation coefficient > 0.5.

RNA/DNA extraction and qRT-PCR

The total RNAs were isolated from kiwifruit leaves using the Plant Total RNA Isolation Kit (Vazyme, RC411) following the manufacturer's protocol. First-strand cDNA was synthesized using an All-in-One First-Strand Synthesis MasterMix (with dsDNase) (EG15133S, BestEnzymes Biotech). The CTAB method was used to extract high-quality genomic DNA from the young kiwifruit leaves. Taq SYBR® Green qPCR Premix (Universal) (EG20117M, BestEnzymes Biotech) was used for qRT-PCR on a CFX Connect real-time PCR Detection system (Bio-Rad), the ACTIN was used as reference genes for calibration of qRT-PCR data. The primers of qRT-PCR are included in Additional file 1: Table S33.

Genetic transformation of kiwifruit

For gene knocking out, the editing sites for the target genes were designed through the CRISPOR web tool [74]. Agrobacterium-mediated genetic transformation of kiwifruit was performed following the previously described procedure [58]. Briefly, the complete PTG structure of U6-29-tRNA-sgRNA containing targeting sequences was synthesized through gene synthesis service (Sangon Biotech, shanghai, China). The constructed plasmid was transferred into A. rhizogenes K599. The buds from ‘HH-1’ were completely submerged in inducted bacterial solution for 10–12 min, then transferred to a co-cultivation medium for 2 days in the dark. Until the seedlings grew, the induced hairy roots were planted in media containing Timentin. DNA amplification was conducted on transgenic plants to check out positive plants.

For gene overexpression, the complete coding sequence of target gene was amplified from cDNA of resistant maternal plant ‘White’, and integrated into pBI121 vector. The constructed plasmid was transferred into A. rhizogenes K599, and the subsequent operation was performed same as described above.

Psa inoculation and evaluation

Shoots were inoculated through an injury length 1 mm with Psa suspension culture of OD₆₀₀ values 0.1 and phenotypic evaluation was performed at 20 days-post-infection (dpi). Leaf discs were inoculated by vacuum osmosis [75] with Psa suspension culture with OD₆₀₀ values 0.01 and phenotypic evaluation was performed at 3 dpi. Tissue culture seedlings with uniformly developmental stage were inoculated by flood assay [76] using Psa suspension culture with OD₆₀₀ values 0.6 for 15 min and phenotypic evaluation was conducted at 7 dpi. Potted seedlings were inoculated by flood assay using Psa suspension culture with OD₆₀₀ value 0.6 for 15 min and the phenotypic evaluation was conducted at 7 dpi.

Physiological indexes measurement

Trypan blue staining (C0040), NBT staining (G4816), DAB staining (G4815) was measured using corresponding assay kits (Beijing Solarbio Science & technology Co., Ltd, Beijing, China) following the manufacturer’s instructions. POD activity (AKAO005C), SOD activity (AKAO001C), CAT activity (AKAO003), MDA content (AKFA013C), H₂O₂ content (AKAO009C) and O₂⁻ content (AKAO008C) were measured using corresponding assay kits (Beijing Boxbio Science & technology Co., Ltd, Beijing, China) following the manufacturer’s instructions. The in planta bacterial growth was measured by counting the number of bacterial colonies on leaf discs using the spread-plate method [77]. GraphPad Prism 9.5 was used for statistical analyses and visualization.

DNA–protein interaction assays

For dual-luciferase assay, the promoters of AeNLR25-1 or AcNLR25-1 were cloned into the pGreen0800-LUC vector and the ORFs of AeWRKY75 or AcWRKY75 were inserted into the pBI121 vector, respectively. The corresponding constructs of pGreenII 0800-LUC and pBI121-HA vector were co-injected into Nicotiana benthamiana leaves after transforming into Agrobacterium tumefaciens strain GV3101. Luminescence was detected 48 hpi (hour-post-infiltration) using 1 mM luciferin with a Tanon 4600SF system. LUC/REN activity was measured using corresponding assay kits purchased from Yeasen Biotechnology (Shanghai, China) following the manufacturer’s instruction.

To construct the plasmid for Yeast-one-hybrid, the full length of AeWRKY75 or AcWRKY75 was amplified from the cDNA of ‘White’ and ‘H0809’ and cloned into pJG4-5 vector. The promoters of AeNLR25-1 or AcNLR25-1 were amplified from the genomic DNA of ‘White’ and ‘H0809’ and cloned into pLacZi vector. The corresponding plasmids were co-transformed into the yeast strain EGY48 and were cultured on the SD/-Ura/-Trp selection medium for 48 h, then monoclonal colonies were transferred to the SD/-Ura/-Trp/+ X-gal selection medium for 24 h.

EMSA was conducted using purified AeWRKY75/AcWRKY75 protein and 3’ biotin-labeled probes containing the putative WRKY-binding cis-element (TGAC/GTCA) from the AeNLR25-1 promoter. The full-length CDS of AeWRKY75/AcWRKY75 was inserted into the pMal-c2X vector for the MBP fusion protein expression. The fusion plasmid was transformed into Rosetta Escherichia coli cells for AeWRKY75/AcWRKY75 protein expression. The probes were 3’ biotin end labeled and converted to dsDNA probes by annealing complementary oligonucleotides (Sangon Biotech.) The experiment was performed using the LightShift Chemiluminescent EMSA kit (Beyotime GS009). The binding specificity was examined by competition probe. For mutant probe, the cis-element (TGAC) was replaced with TTTT. The primers used were listed in Additional file 1: Table S33.

Agrobacterium-mediated transient expression in kiwifruit

To analyze the promoter activity of the A. eriantha ‘White’ and A. chinensis var. chinensis ‘H0809’, the promoter sequence of NLR25-1 was amplified from the DNA of ‘White’ and ‘H0809’ and cloned into the pCAMBIA-GUS vector. Vacuum infiltration was used to perform transient expression in kiwifruit plants. Briefly, seedlings separated from ‘HH-1’ were infiltrated with suspension culture of Agrobacterium strain GV3101 (OD₆₀₀ = 0.8) with infection buffer (10 mM MgCl₂·6H₂O) for 10 min. Before infiltrating, Silwet L-77 was added to buffer with a concentration of 0.025%.

The GUS staining was used to detect promoter activity. Leaves from transiently expressing kiwifruit plants were vacuum immersed in staining buffer for 30 min and then incubated overnight at 37 °C. The stained leaves were incubated in 70% ethanol to remove chlorophyll.

Statistical analysis

All the experiments and measurements have been independently performed at least three times. One-way ANOVA (analysis of variance) was used for statistical analysis, and statistical differences were indicated by the different letters (p < 0.05, Tukey's test). p-value for Student's t-test is presented by asterisk (**: p < 0.01; ***: p < 0.001).

Authors’ contributions

Y. Liu, S. Wang and J. Yue planned and designed the research. Y. Wu, Y. Wang and Y. Lin conducted experiments and contributed equally to this work. Y. Wang, Y. Lin and Y. Wu analyzed data. M. Zhao, Y. Huo, T. Zhang, H. Wang and Y. Zhu participated in performing experiments. Y. Wang, Y. Wu, Y. Lin and Y. Liu wrote the manuscript. L. Wang, P. Zheng, J. Yue, S. Wang, Y. Wang and Y. Liu revised the manuscript. All authors read and approved the final manuscript.

Funding

This research was supported by the National Natural Science Foundation of China (Grant Nos. U23A20204, 32472680 and 32472703) and Lishui City's Key R&D Program Project (2025zdyf14).

Data availability

The genome sequencing data and assemblies have been deposited in NGDC (https://ngdc.cncb.ac.cn/) and the accession number is PRJCA046593 [78]. Genome assembly and annotation files can also be accessed in Figshare [79]. The variation data of reference-unbiased interspecific graph pangenome has also been deposited in Figshare [80]. The collected materials in the study are available from Yongsheng Liu (liuyongsheng1122@ahau.edu.cn) upon request. The custom scripts and code used in the study are available in a GitHub repository under the MIT License [81] and in a Zenodo repository [82]. Additionally, we utilized publicly available genome assembly of A. eriantha ‘Blank’ [35], A. eriantha ‘MD31’ [31], A. eriantha ‘Wild’ [41], A. chinensis var. chinensis ‘Biyu’ [20], A. chinensis var. chinensis ‘Donghong’ [23], A. chinensis var. chinensis ‘H0809’ [35], A. chinensis var. chinensis ‘Hongyang’ [22], A. chinensis var. chinensis ‘Hort16A’ [20], A. chinensis var. chinensis ‘Huangyang’ [20], A. chinensis var. chinensis ‘Jinmi’ [20], A. chinensis var. chinensis ‘Jinpai’ [20] and A. chinensis var. chinensis ‘Zps’ [20], downloaded from Kiwifruit Pan-Genome Database (KPGD) [62].

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.