The rediscovery of Mendel’s laws of heredity in 1900, followed by the establishment of Johansen’s “pure lineage theory” in 1903, marked the beginning of modern plant breeding, in which mutation breeding was attempted in various crops. While mutation breeding with treatments that increase mutation rates, such as gamma rays, heavy ion beams, and ethyl methanesulfonate (EMS) (Jankowicz-Cieslak et al., 2016), has become common in the modern era, mutation breeding at that time depended on spontaneous mutation under natural conditions. Although the extremely low frequency of spontaneous mutation was a major problem for mutation breeding, public experimental institutions in various countries, including Japan, actively tried to find mutant lines with desirable traits and isolate pure lines from them, and several lines have been preserved to the present day.

With the recent remarkable development of next-generation sequencing (NGS) technology, it has become possible to obtain whole-genome information of the various rice lines at relatively low cost, enabling genetic and molecular biological analysis (Guo et al., 2014; Nguyen et al., 2019; Kalendar et al., 2022). In such a situation, the low frequency of spontaneous mutation could be an advantage in finding causative genes involved in traits that differ from the parental varieties. Based on this assumption, we propose an in silico approach to identify the causative gene by comparing the whole genome sequence of a spontaneous mutant line with that of closely related varieties, with almost identical genomes. In this study, using several spontaneous mutant lines that have been isolated in Japan, we succeeded in identifying four causative genes, including a novel gene involved in awn formation. Thus, this method offers the possibility of finding genes involved in traits that breeders are looking for with virtually no effort other than NGS analysis, and can be applied not only for rice but also for other crops.

First, we estimated the number of mutations that occur within 373 Mb of the rice entire genome per generation ( Figure 1 ). For estimating the number of fixed spontaneous mutations, we assume that mutations occur heterozygously and that the causative gene is recessive. The number of heterozygous mutations that newly occur per generation (N) is the spontaneous mutation rate x genome size. Heterozygous mutations are the sum of newly occurring mutations and mutations inherited from heterozygous mutations in the previous generation (Li et al., 2016). As shown in the gray shaded area of Figure 1 , the number of heterozygous mutations can be expressed as the sum of a geometric progression with an initial value of N and a tolerance (r) of 1/2. Spontaneous mutation is a constantly occurring phenomenon, and heterozygous mutation should be considered to have reached saturation in an individual plant. Thus, the number of heterozygous mutations per individual plant is estimated to be 2N. The causative mutation in the mutant line is derived from heterozygous mutations present in the premutation line. The fixed homozygous mutations per generation are estimated to be one-quarter of the heterozygous mutations in the previous generation, i.e., N/2, according to Mendel’s laws. Based on the reports of spontaneous mutation rate in rice (Yang et al., 2015; Ichikawa et al., 2023), the number of fixed mutations per generation is estimated to be 1.38-2.25, and when 50 generations have passed, the number of mutations is estimated to be 69.0-112.5 in the whole genome, and 10.4-16.9 in the coding region based on Itoh et al. (2007) ( Figure 1 ). Furthermore, by eliminating mutations that are not involved in changes in gene function, including non-synonymous substitutions, the number of candidate causative polymorphisms can be reduced to ~10. Based on this speculation, we calculated that approximately 10 to 17 genes would be mutated, and by examining the surrounding information on the effects of these mutations on gene function and tissue-specific expression patterns, etc., we thought it highly likely that we would be able to identify the causative gene.

For this study, we used data from NARO genebank (https://www.gene.affrc.go.jp/databases-plant_search_char.php?type=428), which has registered varieties selected through a mutation selection process from previous breeding programs in Japan. Our first study focused on Ginbouzu-Miidashi, which was isolated in 1919 as a dwarf mutant of Ginbouzu (Terada, 1955). Its entire genome is almost identical to the other three Ginbouzu varieties ( Figure 2A ), but Ginbouzu-Miidashi has a shorter culm length compared to the others ( Figure 2B ), as shown in the old literature. By comparing the whole genome sequence between Ginbouzu-Miidashi and the other Ginbouzu varieties, we identified polymorphisms specific to Ginbouzu-Miidashi and scored these as HIGH (frameshift, stop codon gain/lost) or MODERATE (amino acid substitution, inframe insertion/deletion) by snpEff (Cingolani et al., 2012). As a result, 22 polymorphisms in nine genes were identified as candidate causative polymorphisms ( Supplementary Table 1 ). Based on information on gene function, gene expression, and the alignment of amino acid substitution (details shown in Supplementary Table 1 ), the amino acid substitution R238S in Os06g0570100/LOC_Os06g37364 was determined to be the causal polymorphism ( Figure 2C ). The residue R238 is highly conserved in land plants ( Figure 2D , Supplementary Figure 1 ). As this polymorphism has already been reported as the cause of dwarfism in Tan-Ginbouzu (D35/KO2; Itoh et al., 2004), this supports the validity of this approach.

In a second study, we focused on Kamenoo-Daikoku. According to its name, we assumed that Kamenoo-Daikoku was selected for the “Daikoku (short grain in Japanese)” phenotype (Hayashi and Takamure, 1999). We compared the whole genome sequence and grain length between Kamenoo-Daikoku and five Kamenoo and related varieties, including Rikutou-Rikuu 2 and Rikuu 2 (almost identical genome to Kamenoo) ( Figures 2A, B ). Seventeen polymorphisms in eight genes were identified as candidate causative polymorphisms by the above prediction method ( Supplementary Table 2 ). By comprehensive evaluation, the amino acid substitution Q58L in Os05g0333200/LOC_Os05g26890 was judged to be the causative polymorphism ( Figure 3C ). Because the residue Q58 is highly conserved in land plants, this amino acid substitution was predicted to be deleterious ( Figure 2D , Supplementary Figure 2 ). This gene has been reported as the causative gene for five different mutants showing the same short grain phenotypes (Fujisawa et al., 1999). The five mutants analyzed in this paper by Fujisawa et al. (1999) all had different mutations occurring independently in the same gene, but none of them contained the Kamenoo-Daikoku mutation found here. The short-grain rice produced by the Daikoku mutation is not considered to be an agriculturally useful mutation because of its smaller grain size and lower yield, but if the mutant lines are maintained, the gene responsible for it can be easily identified by the method we propose.

Okada (1919) reported that mutation breeding had been carried out in Japan to remove the awn from Aikoku and its awnless mutants had been isolated. Based on this old document, we compared four long awned Aikoku with the three awnless varieties, which have almost identical genomes ( Figure 4 ). There were no polymorphisms common to the three awnless varieties, but there are nine polymorphisms shared by the two mutants, Tokyo-Mubo-Aikoku and Mubo-Aikoku ( Supplementary Table 3 ), while eight are specific to another mutant, Bozu-Aikoku ( Supplementary Table 4 ), and we predicted that these are candidates for causative polymorphisms. Based on a comprehensive evaluation, DNA insertions expected to cause frameshift in Os06g0695900/LOC_Os06g48065 annotated as E3 ubiquitin ligase (53-bp insertion for Tokyo-Mubo-Aikoku and 85-bp insertion for Mubo-Aikoku) and a 4-bp DNA insertion expected to cause frameshift in Os03g0115700/LOC_Os03g02460, Short-chain dehydrogenase/reductase SDR family protein, for Bozu-Aikoku were selected ( Figure 4C , Supplementary Figure 3 ). When the expression patterns of these genes were compared with known awn-forming genes (An-1/RAE1, Luo et al., 2013 and An-2/LABA1, Gu et al., 2015), they were found to share a common feature in that their expression is strongly repressed by ABA and jasmonic acid ( Supplementary Figures 4 , 5 ). We also found a similarity in their expression is observed in the lemma/palea, pistil, and inflorescence ( Supplementary Figures 4 , 5 ), supporting our findings.

To confirm that these genes are indeed involved in awn formation, we expressed the whole genomic regions of both genes or cDNAs under the control of the maize ubiquitin promoter in Tokyo-Mubo-Aikoku/Mubo-Aikoku or Bozu-Aikoku plants, respectively. As a result, all these transformants formed a long awn ( Figures 4D, E ). These results confirmed that both genes have functions essential for awn formation.

In this study, we proposed an in silico approach that compares the whole genome sequence of a spontaneous mutant line with that of closely related lines and successfully identified four causative genes. These results indicate that this approach is effective in identifying genes that confer useful traits that can be used for breeding. The causative gene for dwarfism found in Ginbouzu-Miidashi and the gene for short grain found in Kamenoo-Daikoku were previously reported genes, respectively (Fujisawa et al., 1999; Itoh et al., 2004). On the other hand, of the two genes identified in this study, although Os06g0695900/LOC_Os06g48065 found in Tokyo-Mubo-Aikoku and Mubo-Aikoku is the same gene as RAE3, very recently identified in African rice (Bessho-Uehara et al., 2023), Os03g0115700/LOC_Os03g02460 found in Bozu-Aikoku is a novel gene involved in awn formation.

Previously, a method called “MutMap” was proposed as a way to map important agronomic traits using rice whole genome sequence information (Abe et al., 2012). This method can identify causative mutations by whole-genome sequencing of pooled DNA from segregants to reveal key genetic loci associated with important agronomic traits in rice, providing valuable insights for rice breeding and genetic improvement. In fact, the MutMap method has been successfully used to isolate many causative genes in several crop mutants. In addition, its improved methods have been proposed and their effectiveness has been reported in several cases (Fekih et al., 2013; Nordstrom et al., 2013; Sahu et al., 2020). However, these methods are essentially based on a genetic approach in which the mutant of interest is crossed with its parental line, the F2 generations are distinguished on the basis of phenotype, and the whole genome is compared between the two to find the mutation that causes the difference in trait. This requires at least two generations of cultivation and a reasonable number of individuals for the F2 segregation generation, which is time consuming and labor intensive. Such problems may not be a major obstacle in the case of model plants/crops such as Arabidopsis and rice, but they can be a major obstacle for plants that take a long time to update a generation, such as trees, perennial crops, some grasses that flower only once every few decades. In addition, many of these plants are often difficult to grow in large populations.

The method we propose in this study does not involve a genetic approach and therefore does not have any of the limitations of MutMap and its modified approaches. Instead, it may be difficult to apply this method widely and generally to many plants and crops. In fact, this method can only be applied to mutants produced by natural mutation. There are two sides to this feature - a good side and a bad side. On the positive side, as we have already mentioned, the natural mutation rate per generation is not high, and therefore it is possible to identify the gene responsible for the mutation by directly comparing the whole genome sequence without resorting to genetic methods, as shown in this paper. On the other hand, this fact clearly shows a disadvantage of this method: the spontaneous mutation rate is low and the resulting mutants are rare. However, in the case of rice, for example, this plant has been cultivated for a very long time as a very important crop, and during this time many mutants have accumulated due to natural mutation. For example, the mutant that produces Daikoku-type seeds was repeatedly introduced in various literature published in the 17th and 19th centuries (Sase, 1684, Iwasaki, 1828), and in fact many different mutant alleles have been found as Daikoku mutants ( Figure 3 , Fujisawa et al., 1999). In the case of plant species that have a long history of cultivation and many lines with abnormal phenotypes have been preserved (e.g. fruit trees and orchids), the approach proposed here can be expected to lead to the causal gene of the mutation with only a very low-cost and low-task whole genome analysis.

Theoretically, it is most likely to succeed by directly comparing the sequences between the plant with the mutant phenotype and its siblings without the mutant phenotype. However, in this study, since those siblings were not available, we used several varieties whose genomes were almost identical to the mutants. In this case, the number of mutations detected was higher than theoretical ( Figure 1 ), but still the causal polymorphisms were successfully identified. Therefore, this method is applicable when genomically close varieties are available. It should be noted, however, that this approach will not always be successful, for example, when the mutation alters gene expression or has epigenetic effects. However, even in such cases, it is possible to predict candidate genes by searching for causative polymorphisms in the promoter regions and comparing their level and specificity of expression.

In terms of mutation rates, a large study of de novo mutations in Arabidopsis has recently been reported (Monroe et al., 2022). According to this study, contrary to the common theory (mutagenesis occurs randomly across the genome), genes under strong selective pressure have a low mutation frequency. If this is true for crops, the mutation rate could vary depending on the trait of interest. In any case, the mutation rate at the whole genome level in plants, especially in crops that are always under artificial control, needs to be analyzed in the future.

Although there are many lines that have been produced by spontaneous mutation in the past, only a few of them have been used in breeding programs, leaving a large amount of unutilized genetic resources. The in silico approach proposed in this study will enable the identification of useful genes that can be used for crop breeding with minimal effort required for NGS analysis. Effective use of this approach will accelerate the molecular breeding through such as the generation of molecular markers (Hasan et al., 2021) and the breeding of new varieties through pinpoint improvement using the latest genome editing technologies (Mishra et al., 2018).

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found in the article/ Supplementary Material .

MS: Data curation, Formal Analysis, Investigation, Methodology, Software, Validation, Visualization, Writing – original draft, Writing – review & editing. SK: Data curation, Formal Analysis, Investigation, Validation, Visualization, Writing – original draft, Writing – review & editing. HY: Data curation, Formal Analysis, Investigation, Methodology, Resources, Software, Validation, Visualization, Writing – original draft, Writing – review & editing. MMo: Data curation, Formal Analysis, Investigation, Resources, Writing – original draft. MK: Data curation, Formal Analysis, Investigation, Resources, Writing – original draft. EK: Data curation, Formal Analysis, Investigation, Resources, Writing – original draft. MMa: Conceptualization, Data curation, Formal Analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Supervision, Validation, Visualization, Writing – original draft, Writing – review & editing.