A population derived from a cross between Japanese pear (Pyrus pyrifolia Nakai, 2n =34) breeding lines ‘515-20’ and ‘415-1’ (140 F₁ seedlings) was used for determination of S-haplotype, construction of genetic maps, and investigation of genome duplications. ‘415-1’ was developed by selective fertilization, achieved by pollinating the style of non-irradiated ‘Kosui’ (S₄S₅ ) with pollen from γ-irradiated ‘Kosui’ and obtaining a rare successful pollination, indicating the presence of a self-compatibility mutation in the pollen (Sawamura et al., 2013). The S-genotype of ‘415-1’ was determined to be S₄dS₅S₅ , where d indicates a partly duplicated haplotype. ‘515-20’ (S₁S₃ ) was derived from a cross between ‘Natsushizuku’ and ‘Hatsumaru’ and was among the 6th generation of cultivars produced in the Japanese pear breeding program at the Institute of Fruit Tree and Tea Science, National Agriculture and Food Research Organization. Both lines are homozygous in some regions, a consequence of their origin involving crosses between closely related cultivars.

Amplicon sequencing was based on simplified AmpSeq, which uses a mixture of two primer sets (Nishio et al., 2023), the first consisting of tailed target primers, the second of primers that contain flow-cell binding sites, indexes, and tail sequences complementary to those in the first set. Primers for S-RNase (Ishimizu et al., 1999) and SFBB18 were used for the target first-primer sets. To obtain the SFBB18 primers, we used the SFBB18 sequences found in apple and Chinese pear (Pratas et al., 2018; Huang et al., 2023) as query sequences, and performed a BLAST search against the reference European pear sequence “Bartlett DH Genome v2.0” (Linsmith et al., 2019) to detect the homologue of SFBB18 in European pear. A primer set was designed to specifically amplify the detected SFBB18 sequence. Information on the primer sets and the sequence alignment of S₁ , S₃ , S₄ , and S₅ are presented in Supplementary Table S1 and Supplementary Figure S1 , respectively. PCR amplification was performed in 10 µL containing 5 µL of 2× Green GoTaq G2 Hot Start Master Mix (Promega, Madison, WI, USA), 0.2 µL of the 1st primer set (1 µM), 1 µL of the 2nd primer set (1 µM), 2.8 µL H₂O, and 1 µL of genomic DNA (2.5 ng/µL) with an initial denaturation of 94°C for 5 min; 35 cycles of 94°C for 30 s, 60°C for 1 min, and 72°C for 30 s; and a final extension at 72°C for 10 min. The products were mixed equally by volume in a single tube and purified with AMPure XP beads (Beckman Coulter, Inc, Bree, CA, USA) following the AMPure XP PCR Purification protocol. The libraries were sequenced by PE 300-bp sequencing on an Illumina MiSeq platform (Illumina, Inc., San Diego, CA, USA). The reads from the Illumina MiSeq were demultiplexed to each cultivar or seedling on the basis of the index sequences in the 2nd primer set, and paired fastq files of each cultivar were obtained. The fastq files were trimmed of adapter sequences and low-quality bases in Trimmomatic v. 0.39 software (Bolger et al., 2014). The paired fastq files were merged in flash2 software with parameters “-M 150 -X 0.05 –allow-outies” (Magoč and Salzberg, 2011). The merged reads were aligned and stacked to obtain the sequences of S-RNase and SFBB18. For each marker, the allele frequencies of the three most common haplotypes were calculated for each individual by using a custom script ( Supplementary File 1 ).

Libraries for ddRAD-Seq were constructed as described in Shirasawa et al. (2016). A total of 200 ng of genomic DNA from each seedling or cultivar was double-digested with PstI and MspI (FastDigest restriction enzymes; Thermo Fisher Scientific, Waltham, MA, USA), ligated to adapters in the LigaFast Rapid DNA Ligation System (Promega, Madison, WI, USA), and purified with Agencourt AMPure XP reagent (Beckman Coulter, Brea, CA, USA) to eliminate short (<300 bp) DNA fragments. Purified DNA was diluted with H₂O and amplified by 20 cycles of PCR with indexed primers. Amplicons were pooled and separated in a BluePippin 1.5% agarose cassette (Sage Science, Beverly, MA, USA), and fragments of 300–900 bp were purified with a Mini Elute Kit (Qiagen). The library was then sequenced on an MGI DNBSEQ-G400 sequencer. The reads were de-multiplexed on the basis of 8-bp dual index sequences. Raw reads were analyzed in Trimmomatic v. 0.39 software (Bolger et al., 2014) to remove reads with poor-quality ends (Q < 30). Sequences were aligned to Bartlett DH Genome v2.0 by using the “mem” command of Burrows–Wheeler aligner v. 0.7.17 software (BWA; Li and Durbin, 2009) with default parameters and avoiding multiple-mapping reads. SNPs and indels were mined by following the Genome Analysis Toolkit 4 (GATK4) best practices pipeline (Van der Auwera et al., 2013). GATK’s Base Quality Score Recalibration (BQSR) was implemented to improve the accuracy of variant calling. No SNPs and indels were filtered using QD, MQ, FS, SOR, HaplotypeScore, MQRankSum, or ReadPosRankSum, because we prioritized the analyses of depth of coverage. The GATK’s HaplotypeCaller was used to call variant with options “–max-reads-per-alignment-start 1000” and “–disable-read-filter NotDuplicateReadFilter” to calculate the depth at each locus, because reads starting at the same position above 50 would be downsampled if the default settings were used. To confirm the SNP distribution mapped to Bartlett DH Genome v2.0, following filtering in VCFtools v. 0.1.16 software (parameters: –minDP 20 –maxDP 100 –min-meanDP 30 –max-meanDP 80; Danecek et al., 2011), we visualized the density of SNPs in ChromoMap software (Anand and Rodriguez Lopez, 2022).

The homozygous regions in both parental individuals (‘515-20’ and ‘415-1’) were estimated by examining the SNPs mapped onto Bartlett DH Genome v2.0 within every 200 kb. If a 200-kb region contained no heterozygous SNPs, that region was presumed to be homozygous. The regions that were homozygous in ‘415-1’, in ‘515-20’, and in both ‘415-1’ and ‘515-20’ were visualized in ChromoMap.

For construction of the genetic map, SNPs and indels were filtered in VCFtools v. 0.1.16 software (parameters: –minDP 15–max-alleles 2 –max-missing 0.9 –maf 0.05 –max-maf 0.95). A genetic map was constructed in LepMap3 software (Rastas, 2017). SNPs showing distorted segregation and no information were removed by using the Filtering2 module with parameter dataTolerance = 0.01. The module SeparateChromosomes2 was used to split the retained SNPs into 17 presumed linkage groups with a LOD score limit between 6 and 13 depending on the linkage group. Subsequently, the JoinSingles2All module was used to assign singular markers to the presumed linkage groups. Finally, the OrderMarker2 module was used for each group. To map both male and female informative markers, the option informativeMask=123 was used for the SeparateChromosomes2, JoinSingles2All, and OrderMarker2 modules.

To estimate the duplicated regions in ‘415-1’ and in seedlings derived from the cross of ‘515-20’ and ‘415-1’, differences in depth of coverage (ΔDP) were estimated. SNPs and indels were filtered with VCFtools (parameters: –minDP 20 –maxDP 100 –min-meanDP 30 –max-meanDP 80). VcfR (Knaus and Grünwald, 2017) was used to extract the depth of coverage (DP defined by GATK). ΔDP was calculated by subtracting the depth of coverage of ‘515-20 (diploid)’ from that of each target individual (diploid or partly triploid). The sliding-window average was applied using the R package windowscanr (Tavares, 2021) based on a window size of 20 SNPs and a step size of 10 SNPs. The 95% and 99% confidence intervals (CIs) were calculated as follows. The sliding average of DP was randomly extracted for each individual and ‘515-20’, and ΔDP was calculated by subtracting each moving average of depth of coverage of ‘515-20’ from that of each individual. The operation was performed 10,000 times, and the top 5% and 1% of ΔDP were defined as 95% and 99% CIs, respectively.

‘415-1’, ‘515-20’, ‘Seedling047’, ‘Seedling053’, ‘Seedling075’, and ‘Kosui’ were used for whole-genome sequencing. PCR-free DNA libraries for 150-bp paired-end DNA sequencing of the six individuals were constructed by using the TruSeq DNA library preparation kit (Illumina). DNA of the six individuals was sequenced on a NovaSeq sequencer. The same GATK best practice pipeline used for ddRAD-seq was used for SNP calling. Different SNP filtering criteria were applied for ΔDP, SNP index, and ΔSNP index analyses.

After SNPs were filtered in VCFtools (–minDP 20 –maxDP 80 –min-meanDP 30 –max-meanDP 70 –max-missing 1), ΔDP was calculated by subtracting the depth of ‘Kosui’ (diploid) from that of each target individual. A sliding-window average of ΔDP based on a window size of 1 Mb and a step size of 0.5 Mb was used to identify the duplicated regions in each individual. To estimate 95% and 99% CIs, the sliding average of depth of coverage was randomly extracted for each individual and ‘Kosui’, and ΔDP was calculated by subtracting each moving average of depth of coverage of ‘Kosui’ from that of each individual. The operation was performed 5,000 times, and the top 5% and 1% of ΔDP were defined as 95% and 99% CIs, respectively.

The SNP index, defined as the ratio of allelic depth of the reference allele to the total depth at each locus, was used for estimating homozygous regions, bias of allele frequencies, and haplotype structures. First, SNPs were filtered by VCFtools with the following options: –minDP 20 –maxDP 80 –min-meanDP 30 –max-meanDP 70 –max-missing 1 –maf 0.05 –max-maf 0.95 –remove-indels –thin 10000. At each locus, the allelic depth for the reference allele and total depth was extracted in vcfR, and the SNP index was calculated and plotted in ggplot2 software (Wilkinson, 2011).

The ΔSNP indexes, defined as the absolute value of the difference between the SNP index of two individuals, were calculated using ‘415-1’, ‘Seedling047’, ‘Seedling053’, and ‘Seedling075’. The SNPs were filtered by VCFtools with following options: –minDP 20 –maxDP 80 –min-meanDP 30 –max-meanDP 70 –max-missing 1 -remove-indels –maf 0.2 –max-maf 0.8. Plotting of the sliding-window average of ΔSNP index was based on a window size of 1 Mb and a step size of 0.5 Mb in the R packages windowscanr and ggplot2.

Seeds resulting from a cross between breeding lines ‘515-20’ and ‘415-1’ were initially sown in Jiffy Pot Strips in January 2022 and later transferred to a nursery in April 2022. The plants were cultivated under natural light and temperature conditions. In December 2022, when shoot elongation had ceased, the height of one-year-old seedlings was measured from the soil level to the tip of the meristem. Significance in plant height between normal diploid seedlings and partly triploid seedlings was assessed using a one-way analysis of variance (ANOVA) test.

To identify S-genotypes and haplotype duplication, we conducted amplicon-sequence-based genotyping for S-RNase and SFBB18 using ‘515-20’, ‘415-1’, and their progeny seedlings ( Supplementary Table S2 ). The respective allele frequencies of S₁ and S₃ in ‘515-20’ (S₁S₃ ) were 0.40 and 0.52 for S-RNase and 0.42 and 0.49 for SFBB18, whereas those of S₄ and S₅ in ‘415-1’ (S₄dS₅S₅ ) were 0.13 and 0.79 for S-RNase and 0.26 and 0.61 for SFBB18 ( Supplementary Table S2 ). Those results support that ‘515-20’ has genotype S₁S₃ and that ‘415-1’ has genotype S₄dS₅S₅ . Because amplicons of S-RNase included more SNPs and indels among S-haplotypes than did those of SFBB18 ( Supplementary Figure S1 ), amplification bias among alleles was higher for S-RNase than for SFBB18. For example, the ratios of S₅ /S₄ frequencies were 6.20 in S-RNase and 2.39 in SFBB18, though the expected ratio is 2.00 for ‘415-1’ because of the duplicated S₅ region. Likewise, the S₃ /S₁ ratio and for ‘515-20’ was 1.32 in S-RNase and 1.15 in SFBB18 (the expected ratio was 1.00). Because of the smaller allele bias in SFBB18, we used the SFBB18 amplicon to detect S-haplotype duplication. Among the seedlings derived from ‘515-20’ and ‘415-1’, 22 out of 140 were determined to have third alleles, because the frequency of the third SFBB18 allele was ≥0.2 in those seedlings. In the seedlings in which only two alleles were detected and the ratio of first/second allele frequencies exceeded 1.5, seedlings were determined to carry a duplication of the first allele and thus to show a partially triploid genotype (e.g., S₃S₄dS₄ in Seedling053 and S₃S₅dS₅ in Seedling075) ( Supplementary Table S2 ).

The segregation estimated from the amplicon sequencing of SFBB18 is summarized in Table 1 . Interestingly, 30 seedlings with triploid genotypes—S₁S₄dS₅ (7 seedlings), S₃S₄dS₅ (15), S₁S₅dS₅ (3), S₃S₅dS₅ (4), and S₃S₄dS₄ (1)—were detected out of the 140 seedlings derived from ‘515-20’ × ‘415-1’. The segregation of female gametes derived from ‘515-20’ was S₁ :S_{3 =} 72:68, which is not significantly different from the 1:1 ratio expected for monogenic inheritance. On the other hand, the segregation of male gametes derived from ‘415-1’ was S₅ :S₅dS₅ :S₄ :S₄dS₄ :S₄dS_{5 =} 107:7:3:1:22, which is presumed to have deviated from Mendelian inheritance. Direct genotyping of single pollen grains in a previous study showed S₄ :S₅ :S₄dS_{5 =} 10:135:28 for S-RNase frequency (Mase et al., 2014b). As that study could not detect allele duplications (i.e., S₅dS₅ and S₄dS₄ gametes were presumed to be included with S₅ and S₄ gametes, respectively), the results obtained here and in Mase et al. (2014b) are similar. The microscopic analysis revealed that ‘415-1’ contained sterile pollen with a smaller size compared to ‘Kosui’ (Mase et al., 2014b). The abortion of pollen likely results from abnormal meiosis during gamete formation caused by genome duplication, leading to significant segregation distortion in male gametes.

In total, 247.9 M reads were obtained from the MGI DNBSEQ-G400 sequencer. The numbers of reads per sample ranged from 1.09 M to 2.86 M with an average of 1.77 M. After cleaning and trimming, the retained sequences were mapped to Bartlett DH Genome v2.0, and 433,922 SNPs and indels were obtained through the GATK best practices pipeline. To confirm SNP density mapped on the reference map, we extracted 27,554 SNPs that showed mean depth of >30× and counted them in each 200 kb ( Figure 1A ). Except for several regions in which no SNPs were mapped, the SNPs were almost equally distributed among the chromosomes. Because recent breeding lines, including ‘515-20’ and ‘415-1’, were developed from ‘Nijisseiki’ and its relatives (Nishio et al., 2016), those lines include some homozygous regions. Before constructing a genetic map, we determined the homozygous regions of ‘515-20’ and ‘415-1’. When a parent had at least one heterozygous SNP in a given 200-kb region, the parent was classified as heterozygous in that region. Interestingly, 46.1% of the genome was homozygous in both ‘515-20’ and ‘415-1’ ( Figure 1B ), 19.9% was heterozygous in both, 12.3% was heterozygous in only ‘515-20’, and 21.7% was heterozygous in only ‘415-1’. After removal of SNPs that had >10% missing data, <15 depth, and <0.05 allele frequency, 11,451 SNPs were retained and used to construct a consensus genetic map of ‘515-20’ and ‘415-1’. The 17 linkage groups were created in accordance with the order of Bartlett DH Genome v2.0 with 6,890 SNPs ( Figure 1C ). The total length of the map was 810 cM, which was smaller than the consensus maps in other studies (1,433 to 3,266 cM; Wu et al., 2014; Yamamoto et al., 2014; Chen et al., 2015; Li et al., 2017; Gabay et al., 2018; Nishio et al., 2018). The lengths of chromosomes 12 and 16 were 1.8 and 15.2 cM, respectively, owing to the long homozygous regions from both parents, which were non-informative for constructing the map. There were also some gaps in several regions, including the first parts of chromosomes 01 and 10, the middle part of chromosome 05, and the end parts of chromosomes 03 and 15. Those regions also corresponded to those that were homozygous in both ‘515-20’ and ‘415-1’ ( Figure 1B ).

Before conducting ΔDP analyses, we extracted the SNPs that showed an average depth of 30 to 80. The average depth of all filtered SNPs from the breeding lines and seedlings was 48.1. The read depths of ‘415-1’ and progeny seedlings were subtracted from those of ‘515-20’, and the sliding-window average based on a window size of 20 SNPs was used to confirm the duplicated region. The plot of ΔDP of ‘415-1’ showed a long duplicated region from positions 15 Mb to 26.5 Mb on chromosome 17 ( Supplementary Figure S2 ). Although the delta depth with window size of 20 SNPs around the 22-Mb position of chromosome 2 in ‘415-1’ exceeded the 99% CI, most other regions, including those on other chromosomes, had no duplicated regions. The plots of ΔDP on chromosome 17 of ‘415-1’ and eight progeny seedlings are shown in Figure 2 . The seedlings determined to carry three S-haplotypes by amplicon sequencing targeting SFBB18 and S-RNase have a long genome duplication in the region from 15 to 26.5 Mb on chromosome 17. Interestingly, the length of the duplicated region in ‘415-1’ and its progeny that had three S-haplotypes was constant, indicating that the duplicated region maintained a certain length over generations. Results from only eight progeny seedlings (representing eight different S-genotypes) are plotted in Figure 2 , but the other 132 individuals showed patterns on chromosome 17 similar to those that had the same S-genotype.

Using PCR-free DNA libraries for 150-bp paired-end DNA sequencing, we obtained 140 M to 168 M total reads for ‘415-1’, ‘515-20’, ‘Kosui’, and three seedlings representing different triploid genotypes (‘Seedling047’, ‘Seedling053’, and ‘Seedling075’; Table 2 ). After trimming and cleaning of whole-genome sequence data, we mapped 135 M to 164 M reads to Bartlett DH Genome v2.0, representing 40.7×to 49.5× genome depth. Variant calling obtained 18,716,743 SNPs for the six individuals. After selecting of SNPs with a mean depth of 30 to 70, 4,867,289 SNPs were retained. ΔDP analyses using whole-genome sequencing showed results similar to those obtained by ddRAD-seq ( Supplementary Figures S2 , Figure 3 ). The limits of the duplicated region on chromosome 17 was estimated more accurately than with ddRAD-seq, and ranged from positions 15.3 Mb to 26.3 Mb in ‘415-1’ ( Supplementary Figure S3 ), ‘Seedling047’, ‘Seedling053’, and ‘Seedling075’, all of which had three S-haplotypes, whereas ‘515-20’ did not show any duplication anywhere in the genome ( Supplementary Figure S4 ). Whole-genome delta depth analysis did not detect any other duplicated or deleted region in ‘415-1’, ‘Seedling047’, ‘Seedling053’, and ‘Seedling075’ ( Supplementary Figure S4 ), including the region around 22 Mb on chromosome 2, where ΔDP of ddRAD-seq detected a duplication ( Supplementary Figure S2 ). Considering the larger number of markers used for estimating ΔDP for whole-genome sequencing than for ddRAD, we conclude that the genome duplication was limited to an 11-Mb region of chromosome 17.

To detect homozygous regions and to confirm the duplicated region on chromosome 17, we calculated the SNP indexes for the six individuals used in the whole-genome sequence analysis ( Figure 4 ). The distribution of SNP indexes showed that there were two homozygous regions common to all six individuals, including positions 4.0–8.2 Mb and 9.8–14.0 Mb. In ‘415-1’, those homozygous regions extended to positions 2.6–8.5 Mb and 9.8–19.6 Mb, and the region from position 15.3 Mb to 19.6 Mb overlapped with the duplicated region. In ‘Seedling047’, ‘Seedling053’, and ‘Seedling075’, the SNP index of position 15.3–26.3 Mb on chromosome 17 was around either 0.33 or 0.67, indicating that this region was triploid, with each locus carrying either one reference allele and two alternative alleles or two reference alleles and one alternative allele.

We calculated ΔSNP indexes to infer regions identical among ‘515-20’, ‘415-1’, ‘Seedling047’, ‘Seedling053’, and ‘Seedling075’ ( Figure 5 ), which would help to clarify the haplotype structure around the duplicated region on chromosome 17. On the basis of the ΔSNPs and ΔSNP indexes, we constructed putative haplotype structures of ‘415-1’, ‘515-20’, ‘Seedling047’, ‘Seedling053’, and ‘Seedling075’ ( Figure 6 ). All five individuals shared the same genotypes at positions 4.0–14.0 Mb on chromosome 17, while some of them had different genotypes at positions 0.0–4.0 Mb and 14.0–15.3 Mb. In duplicated regions, all five individuals had different genotypes at position 19.3–26.3 Mb, in accordance with their having different S-genotypes (S₁S₃ , S₄dS₅S₅ , S₃S₄dS₅ , S₃S₄dS₄ , and S₃S₅dS₅ , respectively). On the other hand, ‘Seedling047’, ‘Seedling053’, and ‘Seedling075’ shared the same genotypes at position 16.3–19.3 Mb, whereas they had different genotypes at position 15.3–16.3 Mb. The result suggests a recombination event between S₁ and S₃ haplotypes at around position 16.3 Mb in a gamete giving rise to ‘Seedling075’ ( Figure 6 ).

The inheritance of duplicated segments in ‘Seedling047’, ‘Seedling053’, and ‘Seedling075’ was estimated as follows ( Figure 6 ). ‘Seedling047’ had an S₃S₄dS₅ genotype, and the duplicated S₅ segment would have been connected to the original S₄ haplotype of ‘415-1’. ‘Seedling053’ had an S₃S₄dS₄ genotype, in which the new duplicated S₄ segment would have been developed through recombination between the duplicated S₅ segment and the original S₄ haplotype. ‘Seedling075’ had an S₃S₅dS₅ genotype, indicating that the duplicated S₅ segment was transferred from the original S₄ haplotype to the original S₅ haplotype.

The average height of normal diploid seedlings was 149.3 cm, whereas partly triploid individuals measured 136.2 cm ( Supplementary Table S3 ), with no significant difference found by ANOVA. Although some partly triploids tended to germinate later, there were no clear distinctions observed between diploids and triploids in subsequent growth, including tree vigor and stem diameter.

Several methods exist to determine S-genotype, including allele-specific PCR and PCR-RFLP (Ishimizu et al., 1999; Takasaki et al., 2004; Kakui et al., 2007; Mase et al., 2014a; Nashima et al., 2015). However, none of these methods can detect S-gene duplication. Here, we determined the S-genotype of individuals, including partly triploid individuals, using amplicon sequencing targeting S-RNase and SFBB18. This method can detect allele duplication effectively. Notably, it revealed novel genotypes that were not identified by the previous method (Mase et al., 2014a)—including S₃S₄dS₄ , S₃S₅S₅ , and S₁S₅S₅ —among seedlings derived from crossing ‘515-20’ (S₁S₃ ) and ‘415-1’ (S₄dS₅S₅ ). We observed minimal bias in amplification of the SFBB18 amplicon, allowing us to confirm allele duplications with a high degree of precision. Nevertheless, it is worth noting that SFBB18 loci are positioned at 448 kb from the S-RNase locus in the genome of ‘Bartlett’ and 264 kb in the genome of ‘Nijisseiki’ (Shirasawa et al., 2021). The detection of one seedling exhibiting recombination between SFBB18 and S-RNase (‘Seedling059’; Supplementary Table S2 ), evidenced by the fact that SFBB18 had one genotype (S₁S₅ ) and S-RNase had another (S₃S₅ ), implies that SFBB18 likely lies outside the S-haplotype. Similar to the findings of this study, some SFBB genes in European pear also exhibited an incomplete linkage to the S-RNase (de Franceschi et al., 2011). These genes and SFBB18 in this study would have placed in close proximity of S-haplotype but not belonging to it. It is possible that some SFBB genes were escaped from S-haplotype in the long term evolution of S-locus.

Due to the presence of recombination between SFBB18 and S-RNase, it is better to use S-RNase for determination of the genotype of the S-haplotype in normal diploid cultivars. On the other hand, our method using both SFBB18 and S-RNase amplicons would be preferable for marker-assisted selection of self-compatibility in practical breeding programs because it is useful for detecting duplicated loci. Individuals carrying three distinct alleles (S₃S₄dS₅ , S₁S₄dS₅ ) are typically self-compatible. However, it is noteworthy that partially triploid individuals with genotypes such as S₁S₄dS₄ , S₃S₄dS₄ , S₁S₅dS₅ , and S₃S₅dS₅ may not necessarily be self-compatible, as they may not produce heteroallelic pollen (e.g., S₁dS₄ , S₃dS₄ , S₁dS₅ , and S₃dS₅ ), which is needed to break down self-incompatibility by detoxification of non-self S-RNase proteins. In other words, pollen carrying the same two alleles (S₄dS₄ pollen and S₅dS₅ pollen) would be rejected by their S-RNase counterparts (S₄ and S₅ , respectively) in the pistil.

Whole-genome sequencing analyses using sliding-window techniques have found extensive application in quantitative trait locus sequencing (QTL-seq), employing both SNP index and ΔSNP index methodologies (Takagi et al., 2013; Xue et al., 2017; Yamakawa et al., 2021). Furthermore, investigations of genome duplications and deletions within γ-irradiated mutants have used depth-of-coverage analysis (Komura et al., 2022). This approach facilitates the identification of relatively large genomic duplications and deletions by subtracting the depth of coverage observed in mutants from that of the original cultivar, within a sliding-window framework. We showed here that this method works well for pear, a heterozygous perennial crop. And we modified it to be applicable to sequence data obtained by ddRAD-seq. For sliding-window analysis in ddRAD-seq, instead of using a physical distance of 200 kb, we defined a region as including 20 SNPs. This modification makes it easier and more cost-effective to detect duplicated regions in a large number of individuals.

Mutations produced by γ-irradiation include insertions, deletions, translocations, duplications, and structural variations (Komura et al., 2022; Abe et al., 2023; Hase et al., 2023). Because ‘415-1’ arose through pollination of a self-incompatible cultivar (‘Kosui’) with γ-irradiated pollen of the same cultivar, a mutation in the S-haplotype was presumed (Sawamura et al., 2013). The duplication around S-haplotypes in ‘415-1’ was substantiated by the detection of a third set of alleles for RNase, SFBB^−γ , and the flanking SSRs (Mase et al., 2014a). This finding was further corroborated through single-pollen genotyping techniques (Mase et al., 2014b). In this study, we clarified that the duplicated segment in ‘415-1’ spanned ~ 11.0 Mb, considerably larger than anticipated, and that it remained consistent in size when inherited by partially triploid progeny. In most cases, the original (non-mutant) S₄ segment was transmitted to the next generation in combination with the duplicated S₅ segment. But the duplicated S₅ segment was sometimes lost from association with the S₄ original haplotype, resulting in transmission of only the original S₄ haplotype to the next generation. This separation of the original S₄ haplotype and the duplicated S₅ segment was also confirmed by single-pollen genotyping (Mase et al., 2014b). Interestingly, recombination between the duplicated S₅ segment and the original S₄ haplotype appears to have occurred in a gamete giving rise to ‘Seedling053’ (S₃S₄dS₄ ), suggesting that the duplicated segment participated in pairing and recombination despite being less than half the size of a typical chromosome 17. Because other chromosomes did not show any duplication or deletion in ‘415-1’ or its progeny, the duplicated segment remained in chromosome 17 (i.e., it had not been translocated to another chromosome). The normal appearance of the genetic linkage map constructed using ddRAD-seq also supports the absence of translocations involving the duplicated segment.

Given the pivotal role of centromeres in mitosis and meiosis (Pluta et al., 1995), it is reasonable to infer that this duplicated region encompasses the centromeric region. Triploids in plant species often arise from the fusion of a reduced gamete (n) and an unreduced gamete (2n) (Pelé et al., 2018; Howard et al., 2023). Unreduced gametes are produced by failure of disjunction of homologous chromosomes in meiosis. There are two types of nondisjunction: first- and second-division restitution (FDR and SDR, respectively). The nondisjunction in FDR produces heterozygous gametes that may accompany a recombination event, whereas that in SDR produces homozygous gametes. Since ‘415-1’ seemed to carry heterozygous gametes (S₄ and S₅ ), the duplication may be derived from FDR.

Within the Pyreae tribe, the occurrence of triploid cultivars has been well documented in various fruit species, including apple, European pear, and loquat (Watanabe et al., 2008; Ferreira dos Santos et al., 2011; He et al., 2012; Lassois et al., 2016; Larsen et al., 2017; Baccichet et al., 2020). Among Asian pears, relatively few triploid accessions have been identified (Kadota and Niimi, 2004; Postman et al., 2015), despite their advantage of self-compatibility. The scarcity of triploid Asian pear cultivars is likely due to challenges related to pollen and seed fertility and overall viability. In the case of ‘415-1’ and its progeny, the triploid region of chromosome 17 is limited to an 11-Mb segment, and this limitation may help to overcome issues with reduced fertility and viability often associated with triploids. In fact, we did not observe significant differences in plant height between partially triploid and diploid seedlings ( Supplementary Table S3 ). ‘415-1’ and these partly triploid seedlings will be useful materials for Asian pear breeding programs. An efficient method for obtaining self-compatible seedlings is selective pollination using partially triploid individuals to pollinate diploid cultivars that share common S-haplotypes. For example, pairing an S₄dS₅S₅ pollen parent and an S₄S₅ seed parent can yield self-compatible offspring without the need for marker-assisted selection techniques.