Genetic diversity and structure of diploid Italian ryegrass (Lolium multiflorum Lam.) cultivars and breeding materials in Japan based on genome-wide allele frequency

2.1 Materials

We used 89 Italian ryegrass accessions: 38 Japanese cultivars, including some bred overseas and sold in Japan, such as ‘Gulf’; 29 Japanese public breeding lines, including the inbred line ‘ILW1’ used as a check for heterozygosity; 7 traditional Japanese landraces adapted to local regions (developed by farmers); 3 Japanese ecotypes (locally adapted populations developed by natural selection); and 11 cultivars and 1 ecotype from other countries (Table 1). Table 1 also shows the origins of cultivars and breeding lines in Japan for which information is publicly available as far as we know. It also shows year of application for Japanese Plant Variety Protection (PVP, http://www.hinshu2.maff.go.jp/en/en_top.html) or its forerunner (before 1978) for cultivars. When the original seeds of cultivars were available, we sowed those; when they were not available, we sowed commercial seeds.

2.2 DNA preparation, sequencing and calculation of allele frequency

Sampling methods for DNA preparation followed the method of Byrne et al. (2013). For germination, 0.5 g of seeds from each accession was put on wet filter paper in a rectangular plastic plate (14 cm × 10 cm) and incubated under light at 25°C for 8–10 days. Then, 2-cm fragments of shoots >1 cm above the seeds were cut and gathered in one bulk. If fewer than 100 seeds germinated, additional seeds were germinated to add to the bulk. Thus, each shoot bulk originated from 100–200 individuals, with the exception that only 22 ‘ILW1’ plants were used. DNA was isolated with a Plant DNeasy mini kit (Qiagen, Hilden, Germany). For ‘QM16’, ‘QM19’ and ‘Niigatakei’, three DNA samples per accession were prepared as biological replicates. A GRAS-Di library was prepared for all 98 pooled-DNA solutions by Bioengineering Lab. Co., Ltd. Inc. (Kanagawa, Japan), under license. The library was constructed by two sequential PCR steps as described in Ito et al. (2020). For ‘Waseaoba’, four libraries from one DNA solution were prepared as technical replicates. Sequence data of paired reads (150 bases × 2) were obtained on a NextSeq 500 sequencer (Illumina) by Bioengineering Lab. Co. A list of unique sequences and read numbers was generated in GRAS-Di software (v. 1.0.3, Toyota, Aichi, Japan). The sequences were mapped to L. multiflorum 130 k scaffolds (Knorst et al., 2019) as a reference genome using BWA (Li & Durbin, 2010). Sequences with mapping quality ≥20 were filtered using SAMtools (Li et al., 2009). The output SAM file was split by the mapped scaffold, and the unique sequences mapped to overlapping genomic regions were assigned to the same genomic loci by our own “awk” script. Loci with read depth (DP) ≥ 15 for >75% of samples (n = 74) were selected, and allele frequency (read number of unique sequences ÷ DP at each locus) was calculated for each sample, except for samples with DP < 15. Unique sequences with an average minimal allele frequency of >0.03 at polymorphic (SNP, indel or combinations of both, i.e., haplotypes) loci were selected, and allele frequency and number of alleles per locus were recalculated.

2.3 Analysis of genetic diversity among or within accessions

Nei's standard genetic distance (D_ST; Nei, 1972) as a measure of genetic diversity between two populations (e.g., cultivars) and expected heterozygosity (H_E; Nei, 1973) as a measure of genetic diversity within a population were calculated from all allele frequency data by using our own Python scripts. The values of replicates were averaged; analysis of variance was performed for pairwise D_ST within and between four types of accessions—Japanese cultivars (n = 38), Japanese breeding lines (n = 29), Japanese landraces and ecotypes (n = 10) and other countries' accessions (n = 12); and then, the means were compared. Hierarchical cluster analysis based on D_ST was conducted by the unweighted pair-group method with arithmetic mean (UPGMA) with the hclust function in R software (R Core Team, 2020). Using allele frequency data of loci without missing data, we conducted principal component analysis (PCA) using the prcomp function in R except for ‘ILW1’. k-means clustering, a nonhierarchical cluster analysis, was conducted using the scikit-learn library (Pedregosa et al., 2011) in Python with the same allele frequency data as PCA. k-means was performed 100 times, with k = 2 following the results of UPGMA and PCA.

2.4 Evaluation of heading date

Young seedlings grown in a greenhouse for 30–40 days were transplanted into the field nursery of the Nasushiobara Research Station, NARO (36°55′N, 139°56′E), on October 24, 2019. Seven or eight seedlings per accession were planted in line at a spacing of 80 cm × 50 cm with two replicates in a randomized block design. Fertilizer was applied at 8 g N/m², 3.5 g P/m², 6.6 g K/m² before transplanting and at 2 g N/m², 0.87 g P/m², 1.7 g K/m² on February 14, 2020. Dates when the tips of ≥3 inflorescences per plant appeared from the flag leaf sheath were recorded. ‘BMS1’, ‘BMS2’ and ‘Tomokei 32’ were not field tested, so their maturity types were judged from data of previous field tests (data not shown).

2.5 Statistical analysis

All other statistical analyses were performed in JSPS software (JASP Team, 2020).

3.1 GRAS-Di analysis in the Italian ryegrass population

GRAS-Di analysis of the pooled DNA obtained 47,976,782 raw reads (average 489,559 per sample). Out of 67,058 unique sequences, 36,819 were mapped at 10,412 loci on the reference genome. Finally, 2629 unique sequences (alleles) at 456 reliable polymorphic loci were screened. The average sequence length was 141.3 bases (min. 50 to max. 200). Out of the 456 loci, 427 (93.6%) were each mapped on a unique scaffold, but in 13 cases (5.7%), two loci were mapped on the same scaffold, and in one case (0.7%), three loci were mapped on the same scaffold. The average number of loci for the genetic diversity analysis in each sample ranged from 196 (‘ILW1’) to 455 (‘Tachimasari’), with a mean of 412. Figure 1 shows the distribution of the number of alleles per locus in this population. The average number was 5.8 in the population and 4.4 in each sample.

3.2 Genetic diversity among Italian ryegrass accessions

We calculated the genetic diversity between all pairs of accessions from the allele frequency in the pooled DNA. The effect of accession type on pairwise D_ST was significant (P < 0.001): The mean D_ST within Japanese breeding lines was significantly higher than those within the other three types, and those between Japanese breeding lines and other types were also significantly higher than those between the other types (P < 0.05, Table S1). On the other hand, there was no significant difference in the mean D_ST within or between Japanese cultivars, Japanese landraces and ecotypes and other countries' accessions, except between that within Japanese landraces and ecotypes and that between Japanese cultivars and other countries' accessions. Figure 2 shows a UPGMA dendrogram based on D_ST calculated from allele frequencies at the 456 loci with maturity type judged from field tests and kinship based on breeding history (Table 1). Accessions presumed to be closely related on the basis of breeding history tended to show closer affinities than others (Figure 2). All samples within each biological or technical replicate (four in all) clustered together (data not shown). Fifteen accessions including ‘Gulf’ formed a subgroup: It included cultivars, landraces and ecotypes from both Japan and overseas with mainly medium maturity (Figure 2). PCA was conducted based on the frequencies of 404 alleles at 83 loci without missing values (except for ‘ILW1’). The first and second PCs explained 14.7% and 7.6% of the variance (Figure 3). The first PC seemed to split the population into two groups as suggested by the hierarchical clustering: accessions with >0.5 of the first PC score were consistent with the 15 accessions that formed a subgroup in the hierarchical clustering, except for ‘Yamaiku 130’ (Figures 2 and 3). Following the results of hierarchical clustering and PCA, k-means clustering was conducted at k = 2 using the same data as PCA. The clustering result obtained with highest frequency split the population into a large group with 73 accessions and a small group with 15 accessions: The small group corresponded to the subgroup in the hierarchical clustering (named group II here: the other is group I) with the exception that that ‘Yamaiku 130’ assigned to the small group by k-means was not included in the subgroup in the hierarchical clustering; and ‘Tachimusha’ was assigned conversely (Table 2 and Figure 2).

3.3 Genetic diversity within each Italian ryegrass accession

Mean expected heterozygosity (H_E) and mean number of alleles per locus (A) over all loci of each accession are shown in Table 2 and Figure 4. Coefficients of variation among replicates were 0.5% to 1.8% in H_E and 1.0% to 4.4% in A. H_E ranged from 0.290 (‘ILW1’) to 0.589 (‘Harutachi’), and A ranged from 2.54 (‘ILW1’) to 5.24 (‘Harukaze’). Among the Japanese cultivars, H_E ranged from 0.489 (‘Sachiaoba’) to 0.589 (‘Harutachi’) and A from 3.87 (‘Nasuhikari’) to 5.24 (‘Harukaze’). H_E and A were significantly correlated (Pearson's correlation coefficient; r = 0.955, P < 0.001). There was no significant correlation between heading date and H_E or A. A significant negative correlation (P < 0.05) was observed between the year of application and H_E of the 29 cultivars registered in the Japanese PVP (r = −0.373, P = 0.046) but not between A and the year of application (r = −0.271, P = 0.154); however, there was a more significant negative correlation for 14 early-maturing cultivars on both indices (r = −0.855, P < 0.001, Figure 5; r = −0.840, P < 0.001, respectively).

4.1 Validity of the genotyping method in this study

We used GRAS-Di analysis of pooled DNA to genotype heterogeneous and heterozygous populations and analyzed the genetic diversity among and within populations (=accessions) on the basis of the allele frequencies. For bulk analysis, ≥48 individuals are recommended to represent a ryegrass cultivar for exploration of relationships and genetic diversity using a pooled-DNA approach (Pembleton et al., 2016); we use pooled DNA from 100 to 200 individuals for genotyping. The genetic distances among replications used for four accessions were small enough to form an isolated cluster, and the variation of within-population diversity was also small (0.8% to 4.4%). These results support the validity of our sampling methods and the stability of GRAS-Di. The fact that the genetic distances of cultivars that were presumed to be closely related on the basis of breeding history were small (Figure 2) also supports the validity of the methods. However, analysis of closer genetic relationships may require more replications of DNA pools.

We used published scaffold-level genome sequences for reference. Accession ‘ILW1’, derived from six rounds of self-pollination of a single heterozygous individual and used as a check here, had the lowest values of H_E and A, but they were larger than expected (theoretical expected H_E is 0.008 even if it is derived from a plant completely heterozygous at all investigated loci). This discrepancy may reflect the low accuracy of the mapping. In addition, only 55% of the unique sequences obtained by GRAS-Di could be mapped. The use of a high-quality reference genome sequences will be important for further analysis of Italian ryegrass diversity.

4.2 Genetic diversity and structure related to breeding history in Japan

Cluster analyses and PCA divided Japanese diploid Italian ryegrass cultivars and breeding materials into two major groups as indicated in Figures 2 and 3 and Table 2: group I, a large population that includes most of the Japanese cultivars, and group II, a small group with mainly medium maturity. The grouping results of ‘Yamaiku 130’ and ‘Tachimusha’ differed depending on the analysis method; it may be due to the difference in the data sets used. We discuss the relationship between the results and the breeding history in each group.

Full-scale breeding in Japan began with landraces derived from cultivars introduced from overseas and multiplied at public stud farms (Table 1). Five such landraces (‘Ibaraki Zairai’, ‘Tottori Zairai’, ‘Miyazaki Zairai’, ‘Kuroishi Zairai’ and ‘Kochi Zairai’) show very close kinship (Figure 2). This suggests that they were derived from a common ancestor, which may be closely related to other foreign cultivars in group I (Figure 2). From the 1970s to the 1980s, ‘Waseaoba’ (Yoshioka et al., 1971), ‘Waseyutaka’ (Kinoshita et al., 1973), ‘Tachiwase’ and ‘Tachimasari’ were bred with the goals of early maturity, erect stature with lodging tolerance and high yield. They were placed in the same cluster as their parent landraces (Figure 2). These early-maturing cultivars still have a high market share in Japan today.

Since 1977, when ‘Minamiwase’ (Kinoshita, 1977) was bred from landraces and foreign cultivars, extremely early-maturing cultivars have been bred to enable double cropping, allowing time for summer crop in warm regions. The hierarchical cluster analysis confirms the close relationship of ‘Minamiwase’ and its selections, ‘Minamiaoba’ and ‘Siwasuaoba’. Extremely early-maturing cultivars can be harvested before the end of the year, but they need to be sown in late summer, which requires blast resistance (Arakawa, 2021). Such cultivars have been bred from ‘Sachiaoba’ (Mizuno et al., 2003). They formed a cluster separate from another extremely early cluster (Figure 2), suggesting different genetic backgrounds.

Since the 2000s, to reduce the risk of nitrate poisoning in cattle, breeding to reduce the nitrate-N content in Italian ryegrass has been conducted in Japan (Harada et al., 2003). As a parental line, ‘Parental Line Nou 3’, with low-nitrate content, was bred from early-maturing cultivars in group I, and then, three cultivars were bred by crossing a breeding line selected from ‘Parental Line Nou 3’ with high-yield breeding lines or cultivars. They and ‘LNG9’, further selected from ‘Parental Line Nou 3’ for low nitrate-N, formed a cluster and are distinguished from each other by the second PC. This structure suggests that the influence of the low-nitrate parental breeding line is significant in the genetic background, and these accessions have a unique allele frequency structure.

In recent years, public breeding of diploid Italian ryegrass has been carried out at the NARO Institute of Livestock and Grassland Science (NILGS) and the Kyushu Okinawa Agricultural Research Center (KARC). Breeding materials in these institutions were derived from very diverse breeding lines and cultivars (Table 1), and they were closely related to each other in group I by clustering analysis, as ‘Nakei 33’, ‘Nakei 34’, ‘Nakei 35’ and ‘Nakei 38’ in NARO/NILGS and ‘QM16’, ‘QE19’, ‘QM19’ and ‘QML19’ in NARO/KARC (Figure 2).

One of the group II accessions, ‘Gulf’, was introduced from Uruguay into the United States and registered with the OECD in 1958. ‘Gulf’ is grown in Japan and accounts for one third of the total Italian ryegrass seed distribution in Japan (Japan Livestock and Seed Association, unpublished data), mainly on account of its low seed price. Because of its widespread use in Japan, it is likely that the ecotypes and landraces in Japan assigned to group II are derived from ‘Gulf’ or closely related foreign cultivars. ‘Dryann’ and ‘Tachimusha’ (assigned to group II by hierarchical clustering only) were domestically bred cultivars having ‘Gulf’ as one of their parental material, which is consistent with the fact that they were assigned in this group. ‘Yamaiku 130’ was classified as group II by nonhierarchical clustering, but it was distantly related to the group II cluster in the hierarchical clustering, maybe because ‘Yamaiku 130’ is a ‘Gulf’-derived line, but was obtained by selfing one or a few genotypes (personal communication from the breeder).

Many group II accessions, with the exception of ‘Dryann’ and ‘Tachimusha’, have an open-book morphology, unlike the erect morphology of Japanese cultivars in group I (Tase & Kobayashi, 1994; our unpublished data). However, many landraces and foreign cultivars in group I also have the open-book form. In addition, group I includes accessions with medium to late maturity. These differences suggest that the genetic structure of the Japanese diploid Italian ryegrass population revealed here cannot be determined from morphological characteristics or earliness but can be determined only by using molecular tools. Guan et al. (2017) classified 122 L. multiflorum genotypes from 41 cultivars collected worldwide into three subgroups on the basis of SSR marker polymorphisms. Among them, ‘Gulf’, ‘Tur’ and ‘Midmar’ overlapped with our materials; ‘Gulf’ and ‘Tur’ belonged to different subgroups, consistent with our results, while ‘Midmar’ belonged to different subgroups according to genotype.

Four accessions—‘ILW1’, ‘Parental Line Nou 2’, ‘Hm4PC’ and ‘Yamaiku 179’—were distantly related to other accessions (most of accessions in group I and group II) and to each other by hierarchical clustering (Figure 2). ‘ILW1’, ‘Parental Line Nou 2’ and ‘Hm4PC’ were derived from only one to three individuals and showed low genetic diversity (Table 2). They have unique alleles, which is the reason for the large genetic distance between them and other accessions. We think that the greater genetic diversity within Japanese breeding lines than within other types is due to the inclusion of diverse accessions, including these experimental lines. On the other hand, there was little significant difference in genetic diversity within or between the other accession types, which suggests that these types are genetically related to each other: That is, the Japanese cultivars are derived from foreign cultivars and Japanese landraces, which in turn originated from foreign cultivars.

4.3 Genetic diversity related to use in breeding

In the breeding of allogamous Italian ryegrass, it is important to maintain genetic diversity while ensuring a degree of homogeneity within the cultivar. We have found a significant decrease in genetic diversity within early-maturing cultivars over the past 60 years of breeding in Japan (Figure 5). This decrease may be attributed to the reliance on a limited gene pool and the increased selection pressure in recent years by targeting specific traits such as low-nitrate content. In general, yield is most important, and new cultivars yield the same as or more than existing cultivars. Therefore, early-maturing cultivars in Japan have maintained yields with decreasing genetic diversity. However, any further reduction in genetic diversity may inhibit future efforts to improve yield and wide environmental adaptability. To maintain or increase genetic diversity within a cultivar, the genetic diversity profile of breeding materials obtained here will be an important tool. In particular, accessions in group II, which were suggested to be genetically distant from the major breeding material population in group I, can be used to maintain and expand genetic diversity. As a practical example, in the recent diploid Italian ryegrass breeding at KARC/NARO, accessions in group II, such as ‘08-KK-08’, have been introduced into breeding lines ‘QE19’, ‘QM19’ and ‘QML19’ (Table 1). In fact, among breeding lines with medium maturity, H_E and A of ‘QM19’, bred in 2019, were significantly greater than those of ‘QM16’ bred in 2016, which corresponds to the previous generation of ‘QM19’ (Table 2; P < 0.001 and P < 0.05, respectively, t test). The construction of more detailed diversity profiles of more diverse breeding materials and cultivars will contribute to the advancement of Italian ryegrass breeding.