The backcrossed progenies carrying chromosomal segments derived from O. glumaepatula accession IRGC105668 at S22A or S22B in the O. sativa L. cultivar T65 genetic background were developed in a previous study (Sakata et al., 2014). In this study, plants with the genotypes S22A-T65⁺/S22A-glum^s|S22B-T65⁺/S22B-T65⁺, S22A-T65⁺/S22A-T65⁺|S22B-T65⁺/S22B-glum^s, or S22A-T65⁺/S22A-glum^s|S22B-T65⁺/S22B-glum^s were designated S22A_SS, S22B_SS, or S22A+B_SS plants, respectively. These lines were maintained by marker-assisted selection using the simple sequence repeat (SSR) markers RM12317 and RM7451 for S22A_SS plants, RM7033 and RM279 for S22B_SS plants, and RM12317 and RM279 for S22A+B_SS plants. Phenotypes of S22A_SS, S22A+B_SS, and S22B_SS plants were discriminated according to the morphology of the sterile pollen grains stained with a 1% iodine-potassium iodide solution. Semi-sterile plants, in which almost all of the sterile pollen grains show no staining, stain slightly, and stain dark brown in color and are smaller than normal grains, were classified as S22A+B_SS, S22A_SS, and S22B_SS (Supplementary Figure S1). Transmission frequency of allele was estimated by maximum likelihood method (Supplementary Methods). The homozygous plants for S22B-glum^s and S22A-T65⁺ obtained in the previous study (Sakata et al., 2014) were used in this study.

Total genomic DNA was extracted according to the method described by Dellaporta et al. (1983), with minor modifications. Primer sequences of the polymerase chain reaction (PCR)-based DNA markers used in this study are listed in Supplementary Table S1. Each 15-μl reaction mixture consisted of 50 mM KCl, 10 mM Tris (pH 9.0), 1.5 mM MgCl₂, 200 mM dNTPs, 0.2 mM primers, 0.75 units of Taq polymerase (Takara, Otsu, Japan), and 10 ng genomic DNA template. PCR was performed in a GeneAmp PCR System 9,700 (Applied Biosystems, Foster City, CA, United States). The cycling profile was an initial denaturation step at 95°C for 5 min; 35 cycles of 95°C for 30 s, 55°C for 30 s, and 72°C for 40 s; and a final elongation step at 72°C for 7 min. Amplified products were electrophoresed on a 4% agarose gel in 0.5× TBE buffer. Agrobacterium-mediated transformation was conducted as described by Yamagata et al. (2010). In brief, genomic fragments digested by restriction enzymes were cloned into the Ti-plasmid binary vector pPZP2H-lac (Fuse et al., 2001), and were then transformed into S22A_SS and S22B_SS plants. The copy numbers of transgenes were analyzed by quantitative PCR (qPCR) in an MX3000P QPCR system (Agilent Technologies, Santa Clara, CA, United States) using QuantiTect SYBR Green PCR Kits (Qiagen, Venlo, The Netherlands) according to our previous analysis (Nguyen et al., 2017). The gene models in the mapping region were obtained from MSU Rice Genome Annotation Project Database Release 7 (MSU7; http://rice.plantbiology.msu.edu/) in O. sativa cv. Nipponbare Os-Nipponbare-Reference-IRGSP-1.0 and O. glumaepatula W1183 accession ALNU02000000 (Jacquemin et al., 2013). The homologous sequences were searched in bl2seq program with cut off score at 1e-50 (Altschul et al., 1990).

For temporal expression analysis of S22B, the developmental stages of male gametophytes in T65, S22B-glum^s heterozygotes, and S22B-glum^s homozygotes were observed under a light microscope without staining. Fifty anthers were sampled from the unicellular, bicellular, and mature stages, respectively, which were placed into 1.5-ml tubes and then frozen in liquid nitrogen. Total RNA was extracted using Trizol reagent (Life Technologies Japan, Tokyo, Japan) from the ground anther using a Multibeads shocker (Yasui Kikai, Osaka, Japan). Protein fractions of the Trizol extract were reserved for subsequent western blotting analysis. Extracted RNA was treated with DNase I (Takara, Otsu, Japan) to degrade contaminated genomic DNA. The first-strand cDNA was synthesized from approximately 100 ng of the extracted RNA and then reverse-transcribed using Revertra Ace (Toyobo, Otsu, Japan). Reverse transcription-quantitative PCR (RT-qPCR) was conducted in a MX3000P QPCR system (Agilent Technologies) using QuantiTect SYBR Green PCR Kits (Qiagen). For RT-qPCR of S22B, a pair of primers, 5'-CTC TGC CAA CTT CTG CAT CGC CAG G-3'/5'-GCT GAT AAG CTT GTA CAT CTC CGA C-3' and 5'-TGG AGG ATC CAT CTT GGC ATC AT-3'/5'-ACA GCT CCT CTT GGC TTA GCA-3' were used for amplification of S22B and the actin 1 gene (Os03g0718100) as an internal control, respectively.

For the promoter-β-glucuronidase (GUS) assay, genomic sequences of the 1,572-bp region upstream of the initiation codon (ATG) of S22B-T65⁺ were cloned into the Gateway-entry vector pENTR/TOPO using pENTR Directional TOPO Cloning Kits (Life Technologies Japan). The cloned insert was transferred into the destination vector pGWB3 (Nakagawa et al., 2007) using LR clonase (Life Technologies Japan) to fuse the promoter and GUS gene derived from pGWB3. The construct was transformed to T65 by Agrobacterium-mediated transformation.

The coding sequence at S22B was amplified from the Nipponbare full-length cDNA clone J023058D10 (AK070727) provided by the National Agriculture and Food Research Organization (NARO), Japan, and was cloned into the Gateway-entry vector pENTR/TOPO (Life Technologies Japan). The GFP-fused gene at N-terminal under a control of 35S CaMV promoter was constructed using LR clonase (Life Technologies Japan) into the destination binary vector pGWB6 (Nakagawa et al., 2007). The construct was transformed into T65 by Agrobacterium-mediated transformation. The root of the obtained T₀ plants was observed in a fluorescence microscope (Biozero, BZ-8000, Keyence, Osaka, Japan) after 500 nM of mitochondrial fluorescent dye MitoTracker Red CMXRos staining.

For western blotting of the S22B product, we prepared a rabbit polyclonal antibody against two synthesized 14-amino acid peptide sequences: N-KLATPLDAGAHDG-C and N-ISGGRKPEQHSLLP-C (Eurofins Genomics, Tokyo, Japan). The antibody specificity was confirmed by western blotting of a 6×His + S22B-T65 fused recombinant protein expressed in Escherichia coli strain BL21. The protein fractions extracted by Trizol were mixed with 2× Laemmli sample buffer and 2-mercaptoethanol, and then run on a 12% TGX gel (Biorad, Hercules, CA, United States). The proteins were transferred onto a polyvinylidene difluoride membrane with a 0.2-μm pore size (ATTO, Tokyo, Japan). Goat anti-rabbit IgG (H + L) antibody and horseradish peroxidase-conjugated antibody (Bio-Rad) were used for the secondary antibody reactions, and detection was performed using chemiluminescence detection with Western BLoT Hyper HRP Substrate (Takara Bio, Shiga, Japan).

All protein sequence data deduced in Setaria viridis, Setaria italica, Panicum virgatum, Botryococcus distachyon, Sorghum bicolor, Zea mays, Botryococcus stacei, and Oryza sativa were downloaded from the Phytozome 12 database (Goodstein et al., 2012). The protein sequences harboring the DUF1668 domain were detected by hidden Markov model searches using hmmsearch software¹ with a cutoff score of 1e−8. After the amino acid sequences of the DUF1668 domains were aligned using MUSCLE software with default parameters (Edgar, 2004), a phylogenetic tree based on maximum-likelihood inference was constructed in RAxML v. 8.2.8 software (Stamatakis, 2014) and drawn in FigTree v. 1.4.2 software.²

To narrow down the candidate region of S22B, we conducted high-resolution mapping of S22B in the BC₄F₆ population in which both S22A and S22B segregated (n = 7,424; Figure 1). Since S22B was previously mapped between the SSR markers RM12329 and RM279 (Sakata et al., 2014), 308 recombinants obtained between the SSR markers RM12329 and RM279 were screened by genotyping at the seedling stage. Linkage analysis of the 308 recombinants demonstrated that five recombinants (9-7, 23-2, 28-1, 38-3, and 39-7) were the most informative plants to map S22B within 18.4 kb of the genomic region in the reference sequence Nipponbare between DNA markers M48 and M46 (Supplementary Figure S2; Figure 1B). For the complementation test of S22B, 13.1-kb Eco72I, 13.5-kb XmnI (XmnI_a), 13.7-kb SalI, and 10.7-kb XmnI (XmnI_b) fragments of the Nipponbare genomic fragment derived from BAC clone OSJNBa008C13 were introduced into S22B_SS plants by Agrobacterium-mediated transformation (Figure 1B). In the non-transgenic progeny of S22B_SS, a reduced transmission efficiency (k) of S22-glum^s alleles (k = 0.11) via pollen from the theoretical transmission (k = 0.5) was observed, which was attributed to the sterility of pollen grains carrying S22-glum^s (Table 1). Similarly, the T₁ generation derived from the T₀ plants transformed by the Eco72I, XmnI_a, and XmnI_b fragments showed segregation distortion of genotypes at RM7033 linking to S22B because of reduced transmission of S22-glum^s (Table 1). Meanwhile, transformation of the SalI fragment significantly altered the segregation of the genotype at RM7033 in the T₁ generation (Table 1), suggesting that pollen grains harboring S22-glum^s recovered pollen fertility by the transformation of the SalI fragment. Pollen fertility of the T₁ plants derived from the T₀ plant with the SalI fragment or XmnI_a fragment was observed (Supplementary Figure S3). The two T₁ lines, 6 and 7, were derived from the two independent T₀ plants carrying a single copy and more than two copies of the SalI fragment, respectively. The heterozygotes at S22B harboring SalI fragment showed recovered pollen fertility as compared with null segregants of heterozygotes or the heterozygotes harboring XmnI_a fragment (Supplementary Figure S3). In the T₁ line 6, transgene segregated at one locus and the heterozygous plants at S22B carrying two, one, and zero copies of the SalI fragment showed more than 90%, approximately 75, and 50% of pollen fertility, respectively, whereas the transformation of the XmnI_a fragment did not recover pollen fertility in T₁ as a negative control (Supplementary Figure S4). LOC_Os02g04420 was predicted to be located on the SalI fragment but not on the other genomic fragments in the MSU7. These data demonstrated that the causal gene of S22B is LOC_Os02g04420, which encodes a protein containing DUF1668, designated S22B_j (Figures 1C,D). The genomic sequences of S22B including 2,500 bp of upstream region from the transcription initiation site are identical to those of Nipponbare. Sequencing of the O. glumaepatula (Acc. IRGC105668) BAC clone GL47D11 showed that the O. glumaepatula allele S22B-g, allelic to S22B_j, possesses three non-synonymous substitutions CCG(P) > GCG(A), TTC(F) > GTC (V), and CCC(P) > GCC(A) and a 2-bp deletion, leading to a frameshifted translation at the S22B C-terminal region (Supplementary Data 1). Many nucleotide substitutions at the promoter and first exon regions were also observed (Figures 1C,D; Supplementary Data 1). The S22B sequences of T65 and IRGC105668 were deposited to DNA databank of Japan (DDBJ; LC596092 and LC596094).

Expression of S22B from the anthers at the unicellular, bicellular, and mature stages was investigated by RT-qPCR among S22B-T65⁺ homozygotes, S22B-glum^s heterozygotes, and S22B-glum^s homozygotes. Transcript levels gradually decreased during the progression of post-meiotic male gametogenesis (Figure 1E). At the unicellular, bicellular, and mature stages, the S22B expression level was elevated in the order S22B-T65⁺ homozygotes, heterozygotes, and S22B-glum^s homozygotes, suggesting that transcription of S22B-glum^s was not as active as that of S22B-T65⁺. The promoter-GUS assay was conducted in T₀ plants with 1.5 kb of endogenous genomic sequences upstream from the initial codon of S22B-T65 in the unicellular, bicellular, and mature pollen stages (Figure 1F). Half of the pollen grains displayed GUS signals in T₀ transgenic plants carrying a single copy of the transgene, but no GUS signals were detected in the anther tissues, demonstrating that S22B is gametophytically expressed in the haploid generation. Together, the results from these expression analyses suggested that the level of gametophytic transcripts of S22B-T65⁺ or S22B-glum^s did not solely determine the fertility of pollen grains carrying S22B-T65⁺ or S22B-glum^s alleles.

Expression of S22B protein was investigated by western blotting using anti-S22B antibody in the anthers at the unicellular, bicellular, and mature stages (Figure 1G). The deduced molecular mass of S22B-T65⁺ and S22B-glum^s proteins was expected to beapproximately 39 kDa based on the coding nucleotide sequences. The density of approximately 37 kDa band estimated in western blotting (37 kDa) gradually increased during the pollen stage. Another major band at a molecular mass probably 46 kDa (46 kDa band) appeared only at the mature stage, and was reduced in the S22B-glum^s heterozygotes as compared with those of homozygotes for S22B-T65⁺ or S22B-glum^s alleles. Although the levels of S22B transcripts were reduced in homozygotes for the S22B-glum^s allele (Figure 1E), protein levels between homozygotes for the S22B-T65⁺ or S22B-glum^s allele were comparable (Figure 1G), implying that the transcript level was sufficient for expression of S22B protein and could be adjusted via feedback regulation.

The intracellular localization of S22B was investigated on root cells of the stable transformant (T₀) transformed by the construct of 35SCaMV prom::N-GFP-S22B protein (Figure 1H). Green fluorescent protein (GFP) signals were colocalized with the mitochondrial fluorescent dye MitoTracker Red CMXRos on root cells. The program for subcellular localization prediction in plant cells, TargetP-2.0 (Almagro Armenteros et al., 2019), MitoFate (Fukusawa et al., 2015), and Localizer (Sperschneider et al., 2017) did not find apparent prediction of localization to the mitochondria. WolfPSORT programs (Horton et al., 2007) weakly suggested that S22B possesses mitochondrial-targeting peptide sequences.

High-resolution mapping of S22A was conducted using the BC₄F₇ population (n = 3,072) derived from the BC₄F₆ plants heterozygous at the S22A genomic region between the SSR markers RM12317 and RM12350, and homozygous for T65 at S22B in the genetic background of T65 (S22A_SS plants; Figure 2). Our previous study revealed that S22A_SS plants showed approximately 50% pollen sterility due to the sterility of pollen grains harboring S22A-glum^s; consequently, homozygotes for S22A-T65⁺, heterozygotes, and homozygotes for S22A-glum^s segregated at a 1:1:0 ratio (Sakata et al., 2014). The genomic region responsible for S22A was delimited within a 151.4-kb region between the DNA markers M24 and SSR33 (Figure 2A). Homozygous plants for S22A-glum^s were not obtained in the high-resolution mapping population, suggesting that male gametophytes with S22A-glum^s in heterozygotes are completely sterile. Within the candidate genomic region, three gene models, LOC_Os02g01790 designated S22A_j1, LOC_Os02g01870 designated S22A_j2, and LOC_Os02g01900 designated S22A_j3, were found to harbor DUF1668 in the reference genomic sequence of Nipponbare based on a Pfam search (Figure 2B). We speculated that the S22A-T65⁺ allele has a function to gametophytically provide fertility to pollen grains carrying the S22A-T65⁺ allele in spite of the haploid genotype, as in the case of S22B. Three Acc65I-digested genomic fragments Acc65I_a, Acc65I_b, and Acc65I_c, containing S22A_j1, S22A_j2, or S22A_j3, respectively, were subcloned from the Nipponbare BAC clone OSNBb0096M07 and transformed into S22A_SS plants. The T₁ population for S22A_j1 showed segregation of homozygous genotypes for S22A-glum^s, demonstrating that pollen grains carrying the S22A-glum^s allele restored pollen fertility due to the effect of the S22A_j1 transgene (Table 2; Supplementary Table S2). The transgene in the T₁ population for S22A_j3 resulted in segregation of two homozygotes for S22A-glum^s, the frequency of which was lower than that in the T₁ population in S22A_j1. Moreover, the homozygous plants for S22A-glum^s were not observed in the T₁ population derived from the T₀ plants transformed by S22A_j2. If S22A_j2 has equivalent function to S22A_j1, non-segregation of the homozygous plants for S22A-glum^s means S22A_j2 does not function to restore pollen fertility with S22A-glum^s. Meanwhile, if S22A_j2 has a partial function to restore the sterility as shown in transformation of S22A_j3, the number of T₁ individuals in this population might not have been sufficient to detect complementation of gene function. Therefore, further study is necessary to elucidate the function of this gene. Eventually, it was concluded that transgenes encoding protein with the DUF1668 domain derived from S22A-T65⁺, at least S22A_j1 and S22A_j3, gametophytically restored the fertility of pollen grains carrying the transgene. The amino acid sequences of S22A_j1, S22A_j2, and S22A_j3 deduced from O. glumaepatula genomic sequences lacked DUF1668, suggesting that these three alleles at S22A-glum^s lost their function as a DUF1668-harboring protein (Figure 2C).

During the process of development of introgression lines of O. meridionalis accession W1625 (MER-ILs) in the genetic background of T65 (Yoshimura et al., 2010), pollen semi-sterile plants carrying W1625 chromosomal segments on the short arm of chromosome 2 with T65 cytoplasm were observed in the BC₄F₂ population (Supplementary Figure S5). Linkage mapping using the BC₄F₃ progeny demonstrated that semi-sterile and normal fertile pollen completely co-segregated to RM7033, and one single Mendelian factor close to RM7033 located in the genomic region between SSR markers RM7451 and RM5984 controls pollen sterility in this population. This genetic factor was designated as S22-mer. Further genetic dissection of the Mendelian factor S22-mer using 753 plants of the mapping population indicated that S22-mer is located between SSR24 and RM5984 containing the S22B locus. The O. meridionalis allele at LOC_Os02g04420 corresponding to S22B_j in O. sativa was designated S22B_m. Sequencing of the O. meridionalis BAC clone OMERIa-82O19 containing S22B_j allelic region showed that O. meridionalis has lost the upstream gene region, including the promoter, transcription initiation site, and N-terminal region of the DUF1668-containing protein (Supplementary Data 1). However, S22B_m does not possess the non-synonymous substitutions and a 2-bp deletion found in O. glumaepatula (Supplementary Data 1) except the CCG (P) > GCG (A) mutation. These results suggest that S22B_m is a loss-of-function allele and evolved from the different lineage of O. glumaepatula. The S22B sequence of W1625 was deposited to DDBJ (LC596093).

A BLASTP search using the amino acid sequences of S22B_j as a query in the proteome of angiosperm species in the Phytozome 12 database (Goodstein et al., 2012) found homologous sequences of proteins only in Poaceae species, including the PACMAD clade species S. viridis, S. italica, P. virgatum, S. bicolor, and Z. mays, and the BOP clade species B. distachyon, B. stacei, and O. sativa (Supplementary Figure S6). Phylogenetic relationships based on DUF1668-containing sequences in Poaceae species revealed that each genome showed more than 40 copies of DUF1668-containing genes except for Z. mays (Supplementary Figure S6). The phylogenetic tree of DUF1668 domain sequences did not exhibit apparent monophyletic branches by species (Figure 3A), suggesting that active gene duplication and diversification of multiple copies of DUF1668-containing genes likely started before the diversification of Poaceae. However, S22B_j, S22A_j1, S22A_j2, and S22A_j3 were included in a single clade (Clade 1) harboring other members from Setaria, Panicum, Botryococcus, and Oryza species (Figure 3, yellow-shaded area). Reconstruction of the phylogenetic tree of DUF1668 members in Clade 1 revealed that DUF1668 sequences deduced from S22A_j1, S22A_j2, and S22A_j3 formed a monophyletic clade, but S22B_j was found to have an independent origin from the clade containing S22A_j1, S22A_j2, and S22A_j3 (Figure 3B).

S22B transcript levels were reduced in the order of homozygotes for S22B-T65⁺, heterozygotes, and homozygotes for S22B-glum^s in the anthers at each of the unicellular, bicellular, and mature stages using forward and reverse primers targeting the same sequences at S22B_j and S22B_g (Figure 1E). These data suggest that the expression level of S22B_g is lower than that of S22B_j. The expression level of S22B also decreased as male gametogenesis progressed. The S22B transcripts are likely mainly contributed from male gametophytes rather than from sporophytic tissues (anthers) based on the results of the promoter-GUS assay (Figure 1F). On the other hand, S22B broadly expressed in vegetative tissues in RiceXpro expression database (Supplementary Figure S6; Sato et al., 2013). Comparison of the genomic regions of S22B_j and S22B_g revealed many nucleotide substitutions in the promoter region and the 5' untranslated region near the transcriptional start site. These substitutions may be involved in regulating interspecific differences in transcription between the two alleles.

Western blotting showed that the level of S22B protein increased with the progression of male gametogenesis, in contrast to the decrease observed at the transcription level (Figure 1G). No obvious difference in S22B protein accumulation was observed between the two homozygotes for S22B-T65⁺ and S22B-glum^s. These data suggest that the level of S22B protein is under control by a post-translational regulation mechanism or that S22B-glum^s is sufficient for male gametogenesis. By contrast, the level of S22B protein was reduced in heterozygotes (Figure 1G). As one example of reduced expression only in heterozygotes in HI systems, genetic variants of a single gene between two diverged alleles at the Sc locus were reported to constitute the HI system (Shen et al., 2017). The japonica allele Sc-j contains a pollen-essential gene, and the indica allele Sc-i contains two or three tandem duplicates of an Sc-j homolog with a distinct promoter. In Sc-j/Sc-i hybrids, the high expression level of Sc-i in sporophytic cells causes suppression of Sc-j expression in pollen and selective abortion of Sc-j-pollen (Shen et al., 2017). Their study further revealed that feedback-mediated regulation of genes or proteins may result in the misregulation of gene expression between differentiated alleles in heterozygotes. Similar to the Sc system, the reduced S22B protein level in the anthers of heterozygotes observed in the present study could be due to allelic suppression of S22B protein via incompatible feedback modulation between the two alleles.

Our western blotting analysis of S22B also revealed that the 46 kDa band appeared specifically at the mature stage. In S22B_SS, development of pollen grains carrying the S22B-glum^s allele starts to delay from the late bicellular stages, and this delay is particularly apparent at the mature stage as compared with normal genotypes (Sakata et al., 2014). Almost all of the pollen grains carrying the S22B-glum^s allele could reach the tricellular stage but did not complete the formation of the male germ unit and failed to produce the pollen tube. We speculate that the reduction of the 46 kDa band in heterozygotes is involved in pollen semi-sterility. It is possible that the 46 kDa band resulted from alternative splicing of S22B_j and/or S22B_g, or from post-translational modification of the 37 kDa band, such as phosphorylation, lipidation, or glycosylation. The genetic effects of nucleotide substitutions and the 2-bp deletion on the S22B coding sequence for protein modification, and the mitochondrial localization of S22B have not yet been elucidated. Thus, these biochemical properties require further study.

In the linkage analysis of S22A (n = 3,072), homozygotes for S22A-glum^s were not obtained (Supplementary Figure S2). The total seed set of self-pollinated S22A_SS plants was fertile, and self-pollinated seeds of S22A_SS showed normal seed germination. These results demonstrate that the S22A-glum^s allele is insufficient for male gametogenesis in heterozygotes. Alternatively, the S22B-glum^s allele is transmitted via male gametophytes in heterozygotes and homozygotes, which segregates at low frequency. The S22A-glum^s allele encodes a truncated protein with loss of the complete DUF1668 domain (Figure 2C). In contrast, the S22B-glum^s allele has a few single nucleotide substitutions and a 2-bp frameshift mutation at the C-terminal, but the DUF1668 domain was predicted to exist in the Pfam search (Finn et al., 2014). S22B_m is likely a loss-of-function allele, and a homozygous plant for the S22-mer allele segregated and showed normal pollen fertility (Supplementary Figure S4). Therefore, an unidentified causative mutation in O. glumaepatula may inactivate gene function, although the DUF1668 domain was predicted in silico.

Diverged haplotypes, including multiple tightly linked genes, are known to induce HI in intra-specific and interspecific hybrids of rice, such as Sa, S5, and S1. The incompatible gene complex includes sporophytic genetic factors and gametophytic genetic factors that determine pollen fertility of its own gametophytes in heterozygotes. The BLAST similarity search using the cloned genes at the known HI as a query did not find homologous sequences within the S22A and S22B mapping regions. If HI systems conferred by S22A and S22B also evolve incompatible haplotypes to induce HI in hybrids, the genes cloned in this study only represent a portion of gametophytic members acting as a protector and were not sufficient to induce HI. To further identify other possible genes including a killer factor involved in HI at S22A or S22B, it is necessary to conduct defective mutant experiments using genome-editing methods such as CRISPR/Cas9 in heterozygotes showing HI. As suggested in the BDM model, nucleotide variants causing HI are heterozygous only at birth, which is necessary to escape from negative selection due to their own maladaptive phenotypes owing to this incompatibility. When S22A_j1, S22A_j2, and S22A_j3 are considered as the ancestral types and the truncated genes at S22A-glum^s are considered as the variant types, nucleotide variants of S22A-glum^s occurring only at birth need to escape from natural selection via BDM partners such as duplicated genes or interacting genes in other genomic regions of O. glumaepatula or ancestral species. Since near-isogenic lines of S22A and S22B in the T65 genetic background were used in this study, the genetic phenomena may appear as monogenic events, and other BDM partners have not yet been identified.

In contrast, a single gene can also cause an HI system. Allelic suppression of the japonica allele at the Sc locus results from the feedback-mediated regulation of gene expression between differentiated alleles in inter-subspecific hybrids between japonica and indica rice (Shen et al., 2017). If the allelic suppression occurred on the semi-sterile S22B_SS, transcription level on heterozygote would show half of the homozygotes for S22B-T65⁺ because transcription of S22B-glum^s is suppressed on heterozygote. However, expression level on the S22B_SS was closed to average of homozygotes of S22B-T65⁺ and S22B-glum^s, suggesting that allelic suppression has not occurred on transcription level. Instead, we suggest that alternative splicing of S22B_j and/or S22B_g or production of the 46 kDa band resulted from post-translational modification is suppressed in pollen grains harboring S22B-glum^s in allele-specific manner. The further studies may reveal this opinion. Another possibility may be that copy number variation of functional copies of DUF1668-containing genes in pollen grains results in competitive transmission efficiencies between fertile and sterile alleles. The allelic differences in development and starch absorption capacity may also result in the biased allocation of nutrient resources between gametophytes, leading to the distorted transmission efficiency of alleles.

In summary, HI at S22A and S22B could be caused by gene complexes within the candidate region, or by structural changes in a single gene. Further genetic analysis to identify other genetic loci interacting with S22A or S22B, and biochemical and genetic characterization of DUF1668-containing genes may reveal the HI mechanism via DUF1668.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.