Previously, an introgression line population was developed from a cross between “Hwayeong” (O. sativa ssp. japonica), a Korean elite rice cultivar, and W1944 (O. rufipogon Griff.), a wild relative of cultivated rice from China (Lee et al., 2005). A DTE phenotype was observed in a line, CR6078 (BC₃F₃), which contained eight W1944 chromosome segments in the Hwayeong (HY) genetic background (Supplementary Figures S2, S3). CR6078 was backcrossed to HY, and the BC₁ plants were self-pollinated to produce 204 BC₄F_2:3 lines. Preliminary mapping with eight simple sequence repeat (SSR) markers, one per W1944 segment, indicated that the gene for the DTE trait was located on chromosome 9 (designated DTE9) between RM553 and RM24637. Six BC₄F₃ plants that were heterozygous for the target region at RM553 and RM24637 were self-pollinated to produce a BC₄F₄ population for fine mapping DTE9. In total, 4,500 BC₄F₄ plants were genotyped with RM553 and RM24637, and 50 recombinants between the two markers were detected. These 50 BC₄F_4:5 recombinant lines were used to narrow down the location of the gene for DTE9 by substitution mapping.

A total of 204 BC₄F₃, 4,500 BC₄F₄ plants, 50 BC₄F₄ recombinant lines, and the parents (HY and CR6078) were grown at the experimental field of Chungnam National University, Daejeon during the summers of 2008, 2009, and 2010, respectively. The 204 BC₄F₃ and 4,500 BC₄F₄ plants were planted at 15 cm intervals within and 30 cm intervals between the rows. The 50 BC₄F₄ lines were represented by a single row of 30-day-old seedlings planted at an interval of 15 cm within and 30 cm between the rows. The BC₄F₃ population was evaluated for ripening ratio (RR), 1,000-grain weight (TGW), and grain weight per plant (GW). RR, a measure of fertility, was calculated as a percentage of the number of filled spikelets divided by the number of spikelets per panicle. The number of spikelets per panicle was measured by averaging the values of three major panicles per plant. The grains were dried naturally after harvesting, and the partially filled or unfilled seeds were removed with water (i.e., density separation). The filled seeds were re-dried in an oven at 30°C for 24 h. The TGW was evaluated by measuring the weight of 100 randomly selected filled grains averaged over two replications. The grain yield per plant (GW) was evaluated by averaging the grain yield (g) at 13% moisture content of five plants.

The germination rate, measured as a percentage, was compared among the seeds of HY, W1944, CR6078, and OsMADS8 transgenic plants. Germination tests were conducted as described by Nguyen et al. (2012) and Shim et al. (2019) with some modifications. Seeds were collected at 45 days after flowering. After breaking the seed dormancy at 50°C for 5 days, two sets of 30 seeds per line were germinated at 30°C under dark condition. Germination tests were replicated three times. Three panicles of HY and CR6078 were sampled at 1 or 2 days before flowering to determine pollen viability. Pollen collection and staining were carried out as described by Jiang et al. (2018). The anthers were crushed into powder and used to observe fertility with a light microscope. Pollen grains that were round and stained black were considered to be viable.

DNA extraction from the leaf tissues, PCR amplification of SSR markers, size separation using polyacrylamide gel electrophoresis, and marker detection by silver staining were performed as described by Shim et al. (2019). Silver staining kits were purchased from Bioneer Co., Daejeon, Korea.¹ The linkage and orientation of the SSR markers were based on SSR maps (McCouch et al., 2002). Using the publicly available sequences for Nipponbare² and 93-11³ reference genomes, insertion and deletion markers were developed using the PRIMER 3.0 program (Supplementary Table S1). The quantitative trait loci (QTL) were fine mapped by comparing the phenotypic means of the genotypic classes of recombinants within the target region using the analysis of variance (ANOVA) feature in the Data Desk 4.0 and based on the interval analysis (p ≤ 0.01 and/or LOD ≥ 3.0) for the markers within the target region using QGENE (Nelson, 1997).

RNA was isolated from the seeds of the three lines (HY, W1944, and CR6078) and 7-day-old seedlings of transgenic plants using a NucleoSpin RNA kit (Macherey Nagel, Duren, Germany) according to the manufacturer’s instructions. Following reverse transcription into the first-strand cDNA with SmartGene Mixed cDNA synthesis kit (SJ Bioscience, Daejeon, Korea), real-time PCR was performed using a CFX Connect Real-Time System (Bio-Rad, Hercules, CA, United States). The real-time PCR protocol was as follows: 15 min at 95°C denaturation and enzyme activation, followed by 40 cycles at 95°C for 20 s, 60–55°C for 40 s (depending on the primer annealing temperature), and 72°C for 30 s. The rice tumor protein homolog (LOC_Os11g43900, OsTMP) was used for normalization, and the relative expression levels were compared using Student’s t-test and Tukey’s test. The primers used in this study are listed in Supplementary Table S1.

To clone the full-length cDNA of OsMADS8 gene from W1944, a pair of gene-specific primer (Forward: 5'-ATGGGGAGAGGGAGGGTGG-3' and Reverse: 5'-GGGGTAG CCATGTCGGCATG-3'), was synthesized based on the sequence of Nipponbare OsMADS8 gene from the RAP-DB.⁴ The PCR product was cloned into the pGEM-T Easy vector (Promega, Madison, WI, United States) and was sequenced to confirm fidelity. To construct the plant expression vectors, we used the Gateway cloning system (Invitrogen, Carlsbad, CA, United States). The PCR products were used for the second PCR reaction with primers containing attB sites (5'-GTACAAAAAAGCAGGCTATGGGGAGAGGGAGGGTGG-3') and (5'-ACCACTTTGTACAAGAAAGC TGGGTCGGGGGTAGCCATGTCGGCATG-3'). The linear fragments flanked by the attB sequences were subjected to site-specific recombination with the entry vector, pDONR207 (Invitrogen), which contains the ccdB gene flanked by attP sites; the reactions catalyzed using BP Clonase yielded entry clones that were used to transform competent DH5a (Escherichia coli) cells. Using LR Clonase enzyme mix (Invitrogen), the pDONR207- OsMADS8 clones were subjected to site-specific recombination with the plant expression vector pGWB5, which contains the cauliflower mosaic virus 35S promoter and the C-terminal GFP epitope. The resulting expression constructs (OsMADS8-OX) were transformed into the Agrobacterium tumefaciens strain GV3101 using the freeze-thaw method (Hofgen and Willmitzer, 1988).

To construct the OsMADS8 RNAi vectors, the OsMADS8-specific primers were designed from the sequence of OsMADS8, containing the following EcoRI and XhoI restriction sites: 5' -GGATCCCAGGCAGGTGACGTTCGCG-3' and 5'-GAGCTCCTTTTATGCCAAGTGTCCC-3'. The PCR product was digested with EcoRI and XhoI and ligated into the pENTR11 vector (Invitrogen). The pENTR11-OsMADS8 clones were then subjected to site-specific recombination with a plant RNAi expression vector, pH7GWIWG2(I), together with the cauliflower mosaic virus 35S promoter (Kim et al., 2012) using LR Clonase enzyme mix (Invitrogen). The resulting expression constructs (OsMADS8-RNAi) were transformed into the A. tumefaciens strain GV3101 using the freeze-thaw method.

The overexpression (OX) and RNA interference (RNAi) constructs were constructed and transformed into HY (hereafter OsMADS8-OX) and CR6078 (hereafter OsMADS8-RNAi), respectively, using the piercing and vacuum infiltration method (Lin et al., 2009).

To obtain embryo sections, seeds of the three lines (HY, W1944, and CR6078) and transgenic plants were incubated in distilled water for 4 h at 4°C. Embryos were manually thin-sectioned in a petri dish using a razor blade (DORCO, Seoul, Korea). The sections (9–12 μm thick) were stained with Perls’ Prussian blue (2% hydrochloric acid mixed with 2% potassium ferrocyanide) for 10 min followed by gently washing with distilled water for 2 min. Imaging was performed with an Olympus BX40 microscope and a Carl Zeiss AxioCam MRc 5 digital camera operated by the Zeiss AxioVision AC system.

The sequence of DTE9 was compared among 137 Asian cultivated rice accessions (O. sativa L.) from the KRICE_CORE (Kim et al., 2016) and sequence variant data were provided by Kongju National University. This mini core collection is comprised of 81 japonica, 43 indica, seven aus, four admixture, and two aromatic accessions (Supplementary Table S2). To compare the OsMADS8 sequence between W1944 and W1943, a putative ancestor of japonica rice (Huang et al., 2012), the 9-kb upstream and downstream of OsMADS8 genomic sequence of W1943 was downloaded from the EnsemblPlants database.⁵

The ABA content was measured using a Phytodetek^® ABA Kit for ABA (Agdia, Elkhart, IN, United States) according to the manufacturer’s instructions. For this ELISA-based assay, seeds were ground in liquid nitrogen, and ABA was extracted in 80% acetone for 24 h at −20°C under dark condition. The supernatant was collected after centrifugation at 13,475 g for 10 min and diluted 10-fold with 1× TBS (Trizma Base 3.03 g/L, NaCl 5.84 g/L, magnesium chloride hexahydrate 0.20 g/L, NaN₃ 0.20 g/L) for use with the Phytodetek^® kit. The absorbance was detected at 405 nm using a Multiskan SkyHigh microplate spectrophotometer (Thermo Fisher Scientific, Waltham, MA, United States). Each assay was performed in duplicate.

The data were analyzed using a one-way ANOVA. Tukey’s test was used to compare the mean values. All the statistical analyses were performed using the Statistical Package for the Social Sciences (SPSS 12; SPSS Inc., Chicago, IL, United States), and statistical significance was set at either p < 0.05, p < 0.01, or p < 0.001.

Dark tip embryos were observed in the seeds of a BC₃F₃ introgression line CR6078, which originated from a cross between HY and the O. rufipogon accession W1944 (Figure 1A). CR6078 showed a significantly higher occurrence of DTE (approximately 42%) compared to the parental lines HY (approximately 1%) and W1944 (approximately 0%; Figure 1B). CR6078 seeds were intermediate in shape and size compared to HY and W1944 and had protruding embryos.

Differences in the number and morphology of the floral organs among the three accessions were also observed. CR6078 had one carpel with three or four stigmas, whereas HY and W1944 had normal pistils with one carpel and two stigmas (Figure 1C). We investigated the stigma characteristics of these three accessions. W1944 had longer stigmas than HY and CR6078, which is consistent with the finding that eight accessions of O. rufipogon and four of O. nivara had CC alleles at OsSYL2 and OsSYL3, contributing to an increase in the stigma length in these wild Oryza species (Dang et al., 2020). Histological analysis of thin sections from these CR6078 embryos showed a cleavage between the coleoptile and epidermis without any tissue covering the coleoptile (Figure 1E). Cleavages were not observed and the coleoptiles were covered by the surrounding tissues in both HY and W1944 (Figures 1D,F). Another cleavage was observed between the coleoptile and scutellum in CR6078, and the scutellum on the radicle side was swollen in CR6078 (Figure 1E). These cleavages suggest that CR6078 seeds initiate a precocious germination process with coleoptile protrusion, but germination is arrested perhaps because of environmental conditions, such as water deficit in the panicle.

For genetic analysis of the DTE trait, CR6078 was crossed with HY, and the F₁ plants (BC₄F₁) were self-pollinated to produce BC₄F₂ seeds. All the BC₄F₂ seeds had the HY phenotype, suggesting that DTE was present in the maternal tissue. In the BC₄F₃ population (204 plants), a segregation of 3 (normal): 1 (DTE) was observed, suggesting that DTE is controlled by a single recessive gene in this population from the HY/CR6078 cross (Table 1). However, the results did not explain the finding that DTE was only observed in CR6078 but not in the two parents.

To better understand the genetics of DTE, a second cross was made between an indica rice cultivar Milyang23 and CR6078. A total of 370 F₂ plants were grown, and F₃ seeds were obtained from each F₂ plant. All the F₂ seeds were Milyang23 type, and the F₃ population segregated 347 normal and 23 DTE plants, suggesting that DTE was controlled by two recessive complementary genes (Table 1). When more than 300 F₃ plants from the cross between HY and Milyang23 were screened, no F₃ line with the DTE phenotype higher than 1% was detected. The results suggested that two or more non-allelic genes regulate the DTE trait, which is not detected in the parents, and that the W1944 allele is required to display DTE.

The pollen viability of HY and CR6078 was also evaluated. The pollen grains of HY and CR6078 were uniform in size and shape and approximately 90% were viable, indicating that the low fertility was not associated with pollen viability of CR6078 (Supplementary Figure S4).

Fifteen SSR markers on the chromosomes 1, 2, 5, 7, 8, 9, and 11 detected W1944 segments in CR6078 (Supplementary Figure S2). A BC₄F₃ population of 204 individuals derived from a cross between HY and CR6078 was evaluated for DTE and some agronomic traits, such as TGW (g), RR (%), and GW (g) per plant. The frequency distribution of the DTE trait in the BC₄F₃ population followed a Mendelian ratio of 3 normal:1 DTE (χ² = 0.53, 0.1 < p < 0.5), indicating that this trait is controlled by a single recessive gene. The gene was mapped to the interval between RM553 and RM24637 on chromosome 9 and designated DTE9 (Figure 2A). The frequency distribution of TGW, RR, and GW per plant, also followed a 3:1 segregation ratio (Table 2). The single-point analysis showed that the QTL for all four traits were linked to DTE9. These results suggest that DTE9^OR (O. rufipogon W1944 allele) has a pleiotropic effect (Table 2).

For fine mapping of DTE9, six heterozygous plants at RM553 and RM24637 were selected from the BC₄F₃ population to produce a BC₄F₄ population, which was screened with the SSR markers RM553 and RM24637 to identify recombinants. Substitution mapping using 50 BC₄F₄ recombinants indicated that DTE9^OR was located between the RM7343 and RM3249. This 230-kb interval has 38 genes based on the Nipponbare reference genome (Figure 2A). Sixty insertion/deletion (InDel) markers were designed by comparing the Nipponbare sequences of 38 genes with that of 93-11 sequences. Eight markers showed polymorphisms between HY and W1944 (Supplementary Table S1). Genotyping of 15 recombinants with eight InDel markers delimited DTE9^OR to a 27.7 kb interval flanked by markers CRDI48 and CRDI37. This interval contains two predicted genes, LOC_Os09g32944 and LOC_Os09g32948 (Figure 2B). LOC_Os09g32944 encodes the SQUAMOSA promoter-binding-like protein 18 (OsSPL18), which controls ligule development, grain weight, and grain number in rice (Lee et al., 2007b; Yuan et al., 2019). LOC_Os09g32948 encodes OsMADS8/24, a member of the MADS-box transcription factor gene family in rice, which is involved in floral organ specification, plant growth, and development (Arora et al., 2007).

The expression of the two candidate genes was compared among HY, W1944, and CR6078 (Figure 2C). No differences in OsSPL18 expression were observed among the three lines. In contrast, OsMADS8 was expressed at higher levels in both CR6078 and W1944 compared to HY suggesting that OsMADS8 is more likely the candidate gene for DTE9^OR (Figure 2C). Thirty single nucleotide polymorphisms (SNPs; 26 and four in the promoter and intron regions, respectively) and 10 InDels (seven and three in the promoter and intron regions, respectively) were detected in the OsMADS8 sequences between HY and W1944 (Supplementary Table S2). The lack of sequence differences in the eight exon regions suggests that the variation in the OsMADS8 promoter region underlies the differential expression of this gene in HY and W1944/CR6078.

Forty OsMADS8 sequence variants between HY and W1944 were detected in the promoter and intron regions. Two hotspots were observed in the promoter region between the nucleotides 19,651,477 and 19,651,511 (SNP1-2 and InDel1-3) and between nucleotides 19,653,010 and 19,654,306 (SNP24-26 and InDel6-8; Figure 2D). Considering that the promoter is responsible for regulating the expression of OsMADS8, the difference between the two variable promoter regions might be associated with the variation of DTE trait in CR6078.

When the comparison was expanded to 137 accessions from the KRICE_CORE collection, a total of 81 SNPs and 35 InDels were detected in the promoter, 5' UTR, and intron regions of OsMADS8 (Supplementary Tables S3–S6). Eighteen out of 40 SNPs in the promoter and intergenic regions were indica or japonica-specific (Supplementary Table S5). Among these subspecies-specific SNPs, W1944 had different SNPs from those of W1943, a putative ancestor of japonica rice (Huang et al., 2012). For SNP5, W1944 with an A-allele was different from all 137 accessions with a G-allele. In addition, 14 and 21 InDels were observed in the promoter and intron regions, respectively (Supplementary Table S4). Ten InDels were observed in the 137 accessions and half of the 10 InDels were species-specific. Six InDels were W1944-specific compared to W1943 (Supplementary Table S6).

To confirm the relationship between OsMADS8 and the abnormal embryo phenotype (DTE), the W1944 full-length OsMADS8 open reading frame was expressed under the control of the cauliflower mosaic virus 35S promoter. Overexpression (OX) and knockdown (RNAi) constructs were developed and transformed into HY (hereafter OsMADS8-OX) and CR6078 (hereafter OsMADS8-RNAi), respectively. Six independent OsMADS8-OX lines and 46 independent OsMADS8-RNAi lines were generated (Figures 3A,B). The T₂ plants of OsMADS8-OX lines #77 and #88 showed significantly higher expression of OsMADS8, whereas T₂ plants of OsMADS8-RNAi lines #38-1 and #40-4 showed significantly reduced transcript levels of OsMADS8 (Figures 3C,D). Significant differences between HY and OsMADS8-OX and CR6078 and OsMADS8-RNAi were observed for grain weight and ripening ratio (i.e., fertility), but not for the other traits (Supplementary Table S7). When the seeds of HY and OsMADS8-OX were compared, OX lines showed a higher rate of occurrence of DTE (about 22–33%) than HY (~3%). As expected, the two RNAi lines had lower DTE rates (~0.2%) than CR6078 (~42%; Figure 3D). Taken together, these results indicate that OsMADS8 is associated with the DTE phenotype (Figures 3A–D). The histological analysis showed that the coleoptile, plumule, and radicle of OsMADS8-OX embryos were protruded, showing clear cleavage from the scutellum tissue (Figures 3E–G). In contrast, the embryos of the HY and the OsMADS8-RNAi lines had coleoptiles embedded in the endosperm and appeared normal (Figures 3H–J). These results indicated that DTE9 is an allele of OsMADS8.

CR6078 has more styles and stigmas than HY and W1944, indicating that the floral organ identity genes were differentially regulated in this line (Figure 1C). Therefore, the expression of genes possibly interacting with OsMADS8, an E-class floral identity group gene, was analyzed. Among five other E-class genes, the expression of OsMADS5, OsMADS7, and OsMADS34 was the highest in W1944, followed by CR6078 and HY (Figure 4A; Supplementary Figure S1). The higher expression of these three genes in CR6078 compared to HY suggests a background effect from W1944 in CR6078. Interestingly, CR6078 showed higher expression of OsMADS1 and OsMADS17 than HY and W1944, indicating that OsMADS8 might regulate OsMADS1 and OsMADS17 expression. This finding is consistent with the DTE trait in CR6078 resulting from the combined effect of the two complementary genes, the W1944 allele of OsMADS8 and an unknown gene from HY (Figure 4A).

To examine potential interactions with other floral homeotic genes, we analyzed the expression of the ABCD group genes. OsSNB, OsMADS4, and OsFZP were expressed at higher levels in W1944 than in CR6078 and HY (Figures 4B,C). It is noteworthy that OsMADS16 was expressed at higher levels in CR6078 than in HY and W1944, whereas OsMADS29 was expressed at a lower level in CR6078 than in HY and W1944 (Figure 4C). The expression of the C-class gene OsMADS3 was remarkably higher in CR6078 than in HY and W1944 (Figure 4D).

The floral organ identity genes that exhibited higher or lower expression in CR6078 compared to the parental lines HY and W1944 were then investigated in the transgenic plants (Figure 5). The expression of these four genes, OsMADS1, 3, 16, and 17, was higher in OsMADS8-OX than in HY. The OsMADS8-RNAi lines also showed lower expression of OsMADS1 and OsMADS16 than in CR6078. Unexpectedly, the OsMADS8-RNAi lines showed higher expression of OsMADS3 and OsMADS17 than CR6078 (Figures 5B,D). These results indicate that OsMADS8 directly or indirectly regulates the expression of these genes, which may underlie the formation of the abnormal floral structure including pistil in CR6078.

Expression of the other ABCDE functional floral identity group genes was also examined in the transgenic lines. Among the A- and B-class genes, OsSNB and OsFZP exhibited higher expression in the OsMADS8-OX lines than in HY. The expression of OsSNB and OsFZP was at higher level in the OsMADS8-RNAi lines compared to CR6078, however, unexpectedly OsMADS4 and OsMADS29 showed lower expression levels in CR6078 than in the OsMADS8-RNAi lines (Supplementary Figures S5A–D). Among the C-class genes, OsMADS58 was expressed at higher levels both in the OX and RNAi transgenic lines, whereas the expression of OsMADS13 was remarkably higher in the OsMADS8-OX lines than in HY (Supplementary Figures S5E,F). Among the E-class genes, OsMADS5 and OsMADS34 expression was remarkably higher in the OsMADS8-OX lines than in HY (Supplementary Figures S5G,H).

No differences in the number of stigmas and styles were observed between HY and OsMADS8-OX lines (Supplementary Figure S6). In addition, the OsMADS8-RNAi lines did not show any difference in the number of floral organs compared to CR6078 plants (Supplementary Table S8; Supplementary Figure S7). These results suggest that the OsMADS8 expression in the OsMADS8-OX and OsMADS8-RNAi lines was not sufficient to induce phenotypic changes in the floral organs.

Previous studies have shown that other genes are also involved in determining the carpel and style or stigma identity (Cui et al., 2010; Li et al., 2011; Conrad et al., 2014; Deng et al., 2017; Zhang et al., 2018). Interestingly, the transcription levels of Open Beak (OPB) and Drooping Leaf (DL) in CR6078 were higher than those in HY and W1944 (Figure 4F). In contrast, OsEMF2b was expressed at lower levels in CR6078 than in HY and W1944. In the OsMADS8-OX plants, the transcript levels of OPB and DL were higher than those in HY, whereas the expression of OsEMF2b was not significantly different (Figures 5E–G). As expected, the transcript levels of OPB and DL in the OsMADS8-RNAi plants were lower than those in CR6078, whereas the OsEMF2b expression was higher (Figures 5E–G). Thus, OsMADS8 in CR6078 altered the expression of OPB, DL, and OsEMF2b in addition to the homeotic genes of the ABCDE model of flower development.

To explain the DTE trait exhibited by CR6078, OsEMF2b was investigated based on the previous finding that this gene directly regulates the expression of OsVP1/ABI3 and the ABA signaling genes regulating seed germination (Chen et al., 2017). CR6078 displayed reduced expression levels of OsVP1/ABI3, ABI4, and ABI5 compared to that of HY and W1944 (Figures 6A–C). Among the three genes regulating ABA catabolism, the OsABA8’ox1 expression levels were higher in CR6078 than in HY and W1944, while no clear pattern was observed in the expression of OsABA8’ox2 and OsABA8’ox3 (Figures 6D–F), suggesting that DTE might be associated with altered expression of OsEMF2b and OsABA8’ox1, which regulates seed dormancy and germination.

To determine whether OsEMF2b is associated with changes in embryonic traits, the expression of genes regulating the ABA signaling pathway and catabolism in OsMADS8-OX and RNAi plants was examined. The expression of ABI3 and ABI4 was significantly reduced in OsMADS8-OX compared to that of HY (Figure 6). However, the expression of ABI5 was significantly reduced in HY compared to that in OsMADS8-OX (Supplementary Figures S8A–C). All three genes (OsABA8’ox1, OsABA8’ox2, and OsABA8’ox3) related to ABA catabolism showed higher expression in the OsMADS8-OX plants than in HY (Supplementary Figures S8D–F). The expression of ABI3, ABI4, and ABI5 was significantly increased in OsMADS8-RNAi compared to that of CR6078 (Supplementary Figures S8A–C). By contrast, the expression of OsABA8’ox1 was lower in the OsMADS8-RNAi plants than in CR6078, whereas OsABA8’ox3 was expressed at a higher level in the OsMADS8-RNAi than in CR6078 (Figures 6D–F). Taken together, these results suggest that OsMADS8 negatively regulates the expression of OsEMF2b, a positive regulator of seed dormancy, by overexpressing OsABA8’ox1.

To determine the effect of OsMADS8 on germination, the germination ratio of HY, CR6078, W1944, and the transgenic lines were compared. CR6078 and the OsMADS8-OX lines started germination at 24 h after sowing with a germination ratio of 82% while the germination ratios of HY and OsMADS8-RNAi lines were less than 3% (Figures 7A,B). The ABA content in the seeds of HY, CR6078, and transgenic lines was analyzed to determine whether the decreased germination ratio was related to the ABA content. The ABA level of OsMADS8-OX plants (385–377 pmoles ml⁻¹) was significantly lower than that of HY (919 pmoles ml⁻¹; Figure 7C). The two OsMADS8-RNAi lines had significantly higher ABA contents than CR6078 (Figure 7C). These results support the idea that in the CR6078 line, OsMADS8 regulates the expression of genes involved in ABA catabolism (Figure 7).

Hybrid weakness, the poor development of hybrids compared to their parents, hinders gene exchange between different species at the postzygotic stage. Hybrid weakness has been reported in rice at both the interspecific or intraspecific crosses, and most studies have reported that this phenomenon is mainly controlled by dominant or recessive complementary interactions of unlinked loci (Fukuoka et al., 1998, 2005; Kubo and Yoshimura, 2002; Matsubara et al., 2007; Yamamoto et al., 2007; Jiang et al., 2008). Characterization of the genetic basis of hybrid weakness contributes to our knowledge of the underlying mechanisms of reproductive isolation and has practical significance with regard to the maximal utilization of invaluable genes from the interspecific or intraspecific crosses in plant breeding (Chen et al., 2014). This study characterized a newly discovered gene for hybrid weakness in rice, DTE9. The DTE trait was observed in CR6078, an introgression line derived from an interspecific cross between O. sativa spp. japonica “Hwayeong” and the wild relative O. rufipogon.

To understand the genetic basis of DTE, a population derived from a cross between HY and CR6078 was developed. In the BC₄F₃ population, a segregation of 3 (normal): 1 (DTE) was observed, suggesting that DTE is controlled by a single recessive gene in this cross. However, a single recessive gene mode of inheritance for DTE is not compatible with the finding that the two parental lines do not display DTE. The F₃ population from a second cross between an indica rice cultivar Milyang23 and CR6078 displayed a segregation ratio of 15:1 (normal:DTE). These results support the involvement of two complementary recessive genes, which is consistent with the previous findings that two complementary recessive genes regulate hybrid weakness in rice (Sato, 1997; Fukuoka et al., 1998; Kubo and Yoshimura, 2002; Ichitani et al., 2007; Matsubara et al., 2007; Kuboyama et al., 2009). Based on our genetic analysis of the progeny from two crosses, we hypothesize that two or more genes (QTLs), OsMADS8 and other unknown genetic factors control DTE in rice (Table 2). Another interesting feature of DTE was that all F₂ seeds from the F₁ plants of the two crosses showed normal Milyang23 and HY embryo types with segregation observed in the F₃ population. This suggests that the DTE trait is of maternal tissue origin.

Dark tip embryo was significantly associated with several deleterious traits including reduced GW (Table 2). Considering that there are no reports or annotations for traits of agronomic importance in or near this region, DTE9 is likely to be associated with the deleterious phenotypes. Based on the DTE phenotype, it is possible that the CR6078 seeds, which have significantly lower ABA content than that of HY, began preharvest germination with coleoptile protrusion. However, germination was arrested perhaps due to environmental conditions, such as water deficit in the panicle. During the germination process, CR6078 consumed carbohydrates in grains, leading to a decrease in the GW. Given the association of DTE9 with several deleterious traits, the interaction between OsMADS8 and the other gene would likely not be selected for during breeding and domestication.

CR6078 displayed an abnormal floral organ phenotype with more styles and stigmas than those of HY and W1944. Gene expression analyses showed that the alteration of OsMADS8 expression in CR6078 and transgenic plants not only changed the expression of the various ABCDE model genes, but also that of genes, such as OPB and OsEMF2b (Figure 4). It is noteworthy that among the ABCDE model genes, the expression of OsMADS1, 3, 16, and 17 was higher in CR6078 than in HY and W1944. Additionally, OPB and DL displayed similar expression patterns, whereas CR6078 showed lower expression of OsEMF2b than HY and W1944. A similar expression pattern was also observed for the genes associated with the ABA signaling pathway and catabolism in CR6078, supporting the idea that the two complementary genes, OsMADS8, and another unknown gene in HY, control the DTE in this population (Figure 6). However, the OsMADS8-RNAi lines showed higher expression of OsMADS3 and OsMADS17 than CR6078 (Figure 5). The expression patterns of several genes were not the same in the transgenic plants as in CR6078. For example, OsMADS3 and OsMADS17 showed higher expression in the OsMADS8-RNAi lines than in CR6078, and ABI5 and OsABA8’ox3 displayed higher expression in the OsMADS8-OX lines than in HY. This is probably due to interactions between transgenes and genes associated with the floral organ phenotype in the background of HY and CR6078. Moreover, CR6078 harbors eight O. rufipogon segments in the HY background.

OsMADS8-OX plants failed to show phenotypic changes compared to HY, possibly owing to the low expression of OsEMF2b in the OsMADS8-OX lines. In a previous study, the loss of function of EMF2b resulted in abnormal changes in the internal floral organ structure. When the phenotype of the emf2b mutant was compared with that of CR6078, the emf2b mutant showed more carpels and stigmas and had lower expression of genes including OsMADS1, OPB, and DL than the wild type (Conrad et al., 2014; Deng et al., 2017). In contrast, CR6078 showed higher expression of OsMADS1, OPB, and DL and lower expression of OsEMF2b than HY, leading to phenotypic changes in CR6078. These results raise the possibility that OsMADS8 directly or indirectly regulates the expression of the B-, E-, and C-class genes as well as other genes, such as OPB and OsEMF2b, and this regulation is responsible for the changes in the inner floral organ phenotype observed in CR6078. It is rational to infer that OsMADS8 alone induces a weak phenotypic change (Cui et al., 2010).

OsEMF2b regulates seed dormancy and seedling growth by activating and inhibiting the expression of the ABA signaling response genes (Chen et al., 2017). ABA promotes seed dormancy and suppresses seed germination and seedling growth. Downregulation of OsEMF2b leads to vivipary and decreased expression levels of OsVP1, an ortholog of ABI3, which is a component of the ABA signaling pathway in Arabidopsis. CR6078 displayed reduced expression levels of OsEMF2b, OsVP1/ABI3, ABI4, and ABI5 compared to that of HY and W1944 (Figures 4F, 6). Additionally, the expression of genes related to ABA catabolism was differentially regulated in CR6078 (Figures 6D–F). Expression of these genes was examined in the OsMADS8-OX and RNAi transgenic plants (Supplementary Figure S8), revealing that OsMADS8 negatively regulates the expression of OsEMF2b, which is related to the activation and repression of expression of the ABA signal responsive genes. Downregulation of OsEMF2b expression might be associated with vivipary and decreased expression levels of OsVP1, which is related to the seed ABA signaling pathway and seed dormancy. Taken together, our results indicate that OsMADS8 plays an important role in regulating the expression of OsEMF2b associated with floral organ determination and seed germination via the ABA signaling pathway and ABA catabolism.

Several studies have reported the role of plant hormones in hybrid weakness (Reiber and Neuman, 1999; Yamada et al., 2001; Yamada and Marubashi, 2003; Hannah et al., 2007; Kobori et al., 2007). Our finding that HY had significantly higher seed ABA content than CR6078 is consistent with the role of OsMADS8 in controlling DTE in CR6078 by directly influencing the expression of genes regulating ABA catabolism. Thus, the hybrid weakness exhibited by CR6078 underscores the importance of plant hormones in this phenomenon (Yamada et al., 2001; Kobori et al., 2007).

In this study, a novel interspecific hybrid weakness gene controlling DTE phenotype was discovered in an introgression line derived from a cross between the japonica variety “Hwayeong” and the common wild rice W1944 (O. rufipogon). We characterized the phenotypic and physiological properties of this newly discovered expression of interspecific hybrid weakness and isolated the causal gene, DTE9, responsible for the DTE and several other traits of hybrid weakness. Based on the finding that two recessive complementary genes are responsible for full expression of the DTE trait, further studies to identify this second gene are in progress. These results contribute to a better understanding of the molecular mechanism causing hybrid weakness and will benefit rice breeding programs by assisting in strategies to maximize exploitation of invaluable genes from the interspecific or intraspecific crosses.