Rice has become an important staple food for more than half of the world’s population. Research on grain quality traits has received much attention in recent years as cereals high in amylose content (AC) and resistant starch (RS) offer potential benefits to human health (Sun et al. 2017). Most grain quality traits are quantitatively inherited. For this reason, various quantitative trait locus (QTL) mapping strategies and populations have been performed to identify and assess QTLs related to grain quality (Bao et al. 2000, 2004; Aluko et al. 2004; Takeuchi et al. 2007, 2008; Kwon et al. 2011; Sabouri et al. 2012; Zhang et al. 2013; Yun et al. 2016). Comprehensive breeding efforts utilizing landrace strains or weedy rice to improve grain quality are underway since they have useful variations in AC and eating quality (Mo et al. 2014; Cho et al. 2019; Park et al. 2019).

Starch mainly consists of two major polysaccharides, amylose and amylopectin, and is considered the main component of many food crops in the human diet. Amylose is essentially a linear, relatively short polymer in which D-glucose units are linked by α-1,4 glycosidic bonds, while amylopectin is a longer, branched polymer that contains both α-1,4 and α-1,6 bonds (Praveen et al. 2018). Variations in the ratio of amylose and amylopectin in starch mainly influence the physicochemical and metabolic properties of rice. Hence, amylose content is considered the most important determinant affecting the eating and cooking quality of rice (Tian et al. 2009). Based on the variation in the amylose:amylopectin ratio in rice grains, different rice varieties are classified as waxy (0-5%), very low AC (5-12%), low AC (12-20%), intermediate AC (20-25%), and high AC (25-33%) (Biselli et al. 2014).

In plants, starch biosynthesis is a complex process that requires the concerted activities of various classes of enzymes including ADP-glucose pyrophosphorylase (AGPase), granule-bound starch synthase (GBSS), starch synthase (SS), starch branching enzyme (SBE), and starch debranching enzyme (DBE) (Qu et al. 2018). ADP-glucose is the precursor for starch biosynthesis and is generated from glucose-1-phosphate by the action of AGPase (Li et al. 2017). Granule-bound starch synthase is mainly involved in amylose synthesis; starch synthases catalyze chain elon-gation and a few branch points can be introduced by the action of branching enzymes. Amylopectin is synthesized by the coordinated actions of SS, SBE, and DBE (Wang et al. 2014). In the rice genome, AC is controlled by one major locus located on chromosome 6 and many minor QTLs (Zhang et al. 2019). The Waxy (Wx) gene encoding granule-bound starch synthase 1 (GBSS1) is mainly responsible for the synthesis of amylose in the endosperm and pollen sac (Wang et al. 1995; Zhang et al. 2019). In Asian cultivated rice, Wx^a (mainly found in indica varieties) and Wx^b (widely found in japonica varieties) are the two major Wx alleles (Dobo et al. 2010). A previous study reported that the differences among mature Wx mRNAs in the five Wx allele types (Wx^a, Wxⁱⁿ, Wx^b, Wx-mq, and wx) were primarily due to an intron 1 G>T splice-site mutation in Wx as well as exon 4 G>A, exon 5 T>C, and exon 6 G>T mutations causing conformational changes at the binding site of GBSS1 (Zhou et al. 2015). Consequently, the amount of GBSS1 and its joint activity are the major factors involved in amylose synthesis resulting in significant variations in AC. Starch branching enzymes, such as the one encoded by starch branching enzyme 3 (SBE3), also introduce α-1,6 links into starch and are critical for amylopectin formation (Rahman et al. 2007). During starch biosynthesis, SBE3 is responsible for structure in rice development and mainly participates in the formation of short chains within the amylopectin cluster (Nishi et al. 2001; Yang et al. 2016).

In the past, due to genetic relatedness among closely related cultivars, DNA polymorphisms had been detected at low frequency, thus creating a bottleneck for the genetic analysis of variations in agronomic traits (Hori et al. 2015). With recent advances in genome sequencing technologies, the detection of genetic variants using sequence reads aligned to a reference genome has made such analyses possible and straightforward. Since the discovery of single nucleotide polymorphisms (SNPs), SNPs have played a significant role as DNA markers. Recently, Kompetitive Allele-Specific PCR (KASP), a detection method that is useful in revealing important allelic variations among cultivars by typing SNPs and insertions and deletions (InDels) at specific sites (Yang et al. 2019), has been utilized for QTL mapping and identification of genes associated with target traits (Cheon et al. 2018).

In the present study, QTLs on starch-related traits were identified using recombinant inbred lines (RILs) derived from two closely related japonica cultivars Dodamssal and Hwayeong by KASP analysis. The KASP markers were successfully used for QTL identification of starch-related traits between japonica cultivars. Since SBE3 and GBSS1 encode two important enzymes for starch biosynthesis, understanding the possible interaction between these two genes was carried out using F₂ populations.

Population development and field trials

A total of 92 RILs developed from a cross between two closely related japonica cultivars Dodamssal and Hwayeong (a non-glutinous cultivar), with large differences in their AC and RS content, were evaluated for these two starch-related traits. Dodamssal was developed from a cross between Goamybyeo (developed from a Milyang95//Gimcheonaengmi/Ilpum cross) and Goami2 (an N-nitroso-N-methylurea-induced mutant) (Cho et al. 2019). For QTL detection and mapping, the 92 RILs were grown in the experimental field at Chungnam National University, Daejeon, South Korea. Plants were grown during the same season in 2017 and 2018. To determine the likelihood of association between amylose synthesis and A/G SNPs creating premature start codons in the 5ʹ-untranslated region (UTR) of GBSS1, 117 rice accessions from Kongju National University, South Korea were subjected to haplotype analysis (Kim et al. 2016, Supplementary Table S1). To assess the possible interaction between the SBE3 and GBSS1 genes, the F₂ population was developed from a cross between Hwayeong and two selected RILs (CR2121 and CR2138) with high AC and RS content. In 2018, the F₂ population [n = 156 (HY × CR2138); n = 54 (CR2121 × HY)] was grown in the Department of Southern Area Crop Science, National Institute of Crop Science, Rural Development Administration, Miryang, South Korea, while 210 F₃ lines were grown in the experimental field at Chungnam National University, Daejeon, South Korea in 2019.

DNA extraction and marker analysis

Total genomic DNA was extracted from fresh leaves of the 92 RILs using the cetyltrimethylammonium bromide (CTAB) method as described by Murray and Thompson (1980) with slight modifications. Leaves were randomly collected from ten individual plants per line and then pooled into a single sample per line. KASP markers, cleaved amplified polymorphic sequence (CAPS) marker, and simple-sequence repeat (SSR) markers were used to analyze genotypes (Supplementary Table S2). The two parental lines were genotyped with 513 KASP markers at the Seed Industry Promotion Center, Foundation of Agri. Tech. Commercialization & Transfer, Korea. The KASP analysis was carried out based on the method described by Yang et al. (2019). Seventy polymorphic KASP markers were used to screen the 92 RILs on 12 chromosomes (Cheon et al. 2018). In addition, total genomic DNA from individual plants of the F2 population was extracted using the abovementioned protocol.

QTL detection and characterization

QTLs were detected and identified by the method of inclusive composite interval mapping (ICIM) by QTL IciMapping version 4.1 (Li et al. 2008). ICIM analysis was performed using forward-backward stepwise regression model 6 with a 20-cM window size. The LOD threshold significance level (P < 0.05) was determined (LOD = 2.8) by computing 1000 permutations. The QTL positions were assigned to the points of maximum LOD score in the target regions.

Whole-genome sequencing

Thirty seedlings of each parental line were sent to Macrogen, Inc. (Beotkkot-ro, Geumcheon-gu, Seoul, South Korea) for whole genome resequencing. Shotgun DNA libraries were prepared from high molecular weight genomic DNA of the parental lines using the TruSeq Nano DNA Kit (San Diego, California 92122, USA). These libraries were used for cluster generation and were sequenced using the HiSeq X Ten.

Interaction between SBE3 and GBSS1 genes using cleaved amplified polymorphic sequences (CAPS) markers

To assess possible interactions between the SBE3 and GBSS1 genes, 210 F₂ plants were used. To detect the T/C polymorphism in exon 16 of the SBE3 gene by using the SpeI restriction enzyme, PCR analysis was carried out with the following reaction mixture: 5.0 mL of 10× buffer, 4.0 mL of 2.5 μM dNTP, 2.0 mL of each forward and primer reverse primer (10 pmol/μL each), 0.5 mL of Taq DNA polymerase (2.5 U/μL), 2.0 mL of genomic DNA (50 ng/μL), and 34.5 mL of water. Using C1000^TM Thermal Cycler (BioRad), amplification was achieved with the following PCR conditions: initial denaturation at 95℃ for 5 minutes; 35 cycles of denaturation at 95℃ for 30 seconds, annealing at 58℃ for 40 seconds, and extension at 72℃ for 90 seconds; and finally, extension at 72℃ for 10 minutes. Post-PCR restriction enzyme digestion was carried out based on the protocol described by Yang et al. (2012).

To detect the G/T polymorphism at the 5ʹ splice site of intron 1 of GBSS1 by using the AccI restriction enzyme, the digestion mixture was as follows: 8.75 mL of PCR product, 1.0 mL of 10× restriction buffer, and 0.25 mL of AccI (10000 U/mL). The digestion mixture was incubated at 37℃ for 3 hours, electrophoresed on a 2% agarose gel, and visualized with a UV transilluminator.

Estimation of AC and RS content

Analysis of AC and RS content was performed over the two-year period. Thirty seeds were collected from each of five plants per line and pooled into a single sample per line, followed by dehulling and polishing using the KETT “Pearlest” Polisher (Tokyo, Japan). The polished rice grains were a mixture of both whole and broken grains. The broken grains were discarded and only the whole grains were selected for further grinding. One hundred whole polished grains were selected, ground into a fine powder, and used for the analyses. The AC and RS content were analyzed in duplicate using the Megazyme assay kit (Megazyme Ltd., Wicklow, Ireland), following the manufacturer’s instructions.

Statistical analysis

All data were statistically analyzed based on one-way ANOVA and Tukey’s test using MINITAB 16.2.4 software at a significance level of 0.05. Tukey’s test was used for multiple comparisons. Pearson’s correlation coefficients were used to establish correlations between AC and RS content.

Seed characterization and phenotypic variations in starch-related traits

The grain appearance, AC, and RS content were compared between the japonica rice cultivars Dodamssal and Hwayeong. The grain appearance of the two parental lines was morphologically distinct with a large difference in AC and RS content. Dodamssal possessed a chalky endosperm with a white core in the center region of the seed, while Hwayeong had a translucent endosperm (Fig. 1). The AC and RS content were consistently higher in Dodamssal than in Hwayeong in the two-year field experiment (Fig. 2). The AC of Dodamssal was 39.25% (2017) and 36.23% (2018), while that of Hwayeong was 11.02% (2017) and 13.36% (2018). The RS content of Dodamssal was 9.48% (2017) and 9.17% (2018), while that of Hwayeong was 0.49% (2017) and 0.30% (2018).

Figure 1. Comparison of seed morphology between parental lines.

Figure 2. Comparison of starch-related traits of parental lines in milled rice grains in the two-year field experiment. Estimated (A) amylose content (AC) and (B) resistant starch (RS) content between Dodamssal and cultivar Hwayeong. Values are means ± standard deviation from two replications. *, **, and *** indicate significant differences of P < 0.05, P < 0.01, and P < 0.001 based on ANOVA, respectively.

Similarly, phenotyping for AC and RS content was carried out on 92 RILs in the two-year field experiment. The AC of the RILs varied from 11.51-40.59% and 12.05-37.42% in 2017 and 2018, respectively (Fig. 3A). We found that the AC of the 92 RILs had a right-skewed distribution in both years (skewness = 1.73 (2017) and 0.52 (2018)). In 2017, high kurtosis (5.59) was observed indicating that the data were tailed heavily relative to the normal distribution. However, in 2018, low kurtosis (0.80) was observed. In contrast, the RS content of the RILs varied from 0.41-10.71% and 0.27-10.19% in 2017 and 2018, respectively (Fig. 3B). The RS content showed a right-skewed distribution in the two years (skewness = 4.81 (2017) and 4.87 (2018)); high kurtosis was observed in both years (kurtosis = 23.30 (2017) and 23.79 (2018)). It is possible that the three RILs with high RS content affected the normal distribution of the data. High phenotypic variation was consistently observed for the two starch traits in the two-year field experiment signifying positive correlations (Table 1). The highly significant correlations indicate that AC and RS content might be interrelated and could influence each other.

Table 1 . Correlation analysis between amylose content (AC) and resistant starch (RS) content in the 92 RILs in the two-year field experiment.

Year	Trait	2017			2018
		AC	RS		AC
2017	RS	0.695***
2018	AC	0.789***	0.566***
	RS	0.700***	0.977***		0.588***

Data are represented as Pearson’s correlation coefficient. ***indicates a significant difference of P < 0.001.

Figure 3. Frequency distribution of (A) amylose content (AC) and (B) resistant starch (RS) content in the 92 RILs in the two-year field experiment. DD: Dodamssal; HY: Hwayeong.

QTL detection for starch-related traits

Based on one-way ANOVA, SMA led to the identification of one QTL associated with RS content on chromosome 2 and two QTLs associated with AC on chromosomes 2 and 6 (Table 2; Fig. 4). The QTL associated with RS content and AC on chromosome 2 was allelic to SBE3. Moreover, the other QTL associated with AC on chromosome 6 was near the Wx gene locus encoding GBSS1. Based on the two-year field experiment, the Dodamssal allele at qRS2 largely increased the RS content. Similarly, the Dodamssal alleles at qAC2 and qAC6 largely elevated the AC. qAC2 had the higher LOD score of 8.40 in 2017 while qAC6 had the higher LOD score of 16.39 in 2018, explaining 34.1% and 55.4% of the phenotypic variation, respectively.

Table 2 . List of QTLs for starch-related traits in the two-year field experiment detected and identified by a single marker analysis combined with composite interval mapping.

Trait	Chr	QTL	Marker	Year	Physical position (bp)	P-value	R2 (%)	Parent contributing higher value allele	LOD
RS	2	qRS2	CS02_001	2017	19358818	0.000***	93.07	Dodamssal	37.04
				2018		0.000***	95.38		28.02
AC	2	qAC2	CS02_001	2017	19358818	0.000***	46.55		8.40
				2018		0.000***	31.56		3.79
	6	qAC6	KJ06_005	2017	1038324	0.000***	28.96		5.49
				2018		0.000***	64.95		16.39

***Indicates significant difference of P < 0.001 based on one-way ANOVA. CS and KJ-markers derived from Cleaved Amplified Polymorphic Sequences (CAPS) and Kompetitive allele specific PCR (KASP), respectively. LOD: Logarithm of the odds.

Figure 4. Locations of QTLs on physical map. Numbers on the left side indicate the physical position (Mbp) along each chromosome while the name of each marker was on the right side. qAC: QTL for amylose content (AC), qRS: QTL for resistant starch (RS) content.

Candidate genomic regions for starch-related traits

Whole-genome sequencing of parental lines was carried out. Sequencing led to the identification of SNPs/InDels linked to starch-related candidate genes (Table 3). A non-synonymous SNP associated with SBE3 that resulted in a leucine (Hwayeong) to proline (Dodamssal) mutation was identified. This conservative amino acid change in Dodamssal is a missense mutation that corresponds to the T/C SNP located at exon 16 (Yang et al. 2012; Yang et al. 2016). Two SNPs, one located at the 5ʹ splice site of intron 1 and one generating a premature start codon in the 5ʹ-UTR, and an InDel located in the 5ʹ-UTR were identified on chromosome 6 associated with GBSS1. The G/T SNP located at the 5ʹ splice site of intron 1 and the 5ʹ-UTR InDel are well-known amylose-linked mutations. However, the function of the A/G SNP creating a premature start codon in the 5ʹ-UTR of GBSS1 remains unknown.

Table 3 . List of candidate genes with the SNPs/InDel detected by whole-genome sequencing between parental lines.

Trait	Chr.	Marker	Candidate gene	Locus ID	SNP/InDel Position (bp)	Ref (Nipponbare)	Dodamssal	Hwayeong	Region
RS	2	CS02_001	Starch Branching enzyme 3 (SBE3)	LOC_Os02g32660	19358818	T	C	T	Exon
AC	2	CS02_001	Starch Branching enzyme 3 (SBE3)	LOC_Os02g32660	19358818	T	C	T	Exon
	6	KJ06_005	Granule-Bound Starch Synthase I (GBSS1)	LOC_Os06g04200	1765668	gtctctctctctctctctctctctctctctctctctct (ct = 18)	gtctctctctctctctctctctctctctctctctct/gtctctctctctctctctctctctctctctctct (ct = 17/16)	gtctctctctctctctctctctctctctctctctct (ct = 17)	5ʹ UTR
					1765761	T	G	T	Intron variant
					1765799	A	G	A	5ʹ UTR

SNPs associated with GBSS1

To determine the possible association of the A/G SNP with amylose content, haplotype analysis was carried out using 117 rice accessions (KRICE_CORE). These rice accessions were categorized into three haplotypes based on the two SNPs that were identified (one located in the 5ʹ splice site of intron 1 and the other generating a premature start codon in the 5ʹ-UTR, Fig. 5). Wx^a harboring the G allele at the 5ʹ splice site of intron 1 predominantly occurred in indica, while Wx^b harboring the T allele was mostly present in japonica. All eight Tongil-type rice accessions belonged to HAP2-1 with the T allele at the 5ʹ splice site of intron 1, similar to HAP2-2. This result suggests that the T allele of japonica was selected in the Tongil-type rice breeding program to reduce AC because Tongil-type rice was derived from an intersubspecific cross between indica and japonica. HAP1-2 and HAP3 both had G alleles at the 5ʹ splice site of intron 1. Mostly, japonica rice has the T allele at this location. Nonetheless, the AC of the three haplotypes (HAP1-2, HAP2-1, and HAP3) was not significantly different. Among the haplotypes, HAP1-1 (composed of indica rice accessions harboring the G allele at the 5ʹ splice site of intron 1 and the A allele creating the premature start codon in the 5ʹ-UTR) had an AC of 26.13%, which was not significantly different from that of HAP3 (25.61%, composed of japonica rice accessions harboring the G allele at the 5ʹ splice site of intron 1 and the G allele creating the premature start codon in the 5ʹ-UTR). For HAP1-2, the existence of the G allele creating the premature start codon might be involved in increasing AC. The AC was significantly different between HAP1-1 and HAP1-2 (composed of japonica rice accessions harboring the G allele at the 5ʹ splice site of intron 1 and the A allele generating the premature start codon in the 5ʹ-UTR). However, the AC of HAP3 (25.61%) and HAP1-2 (22.06%), which are both composed of japonica rice accessions, was not significantly different. This initial finding suggests that the G allele creating the premature start codon has a minor effect on amylose synthesis. However, it is also possible that the sample size of HAP3 was too small and insufficient to elucidate an association between the 5ʹ-UTR A/G SNP and amylose synthesis.

Figure 5. Haplotype analysis of SNPs associated with granule-bound starch synthase 1 (GBSS1) using the 117 rice accessions from the KRICE_CORE set and the estimated amylose content (AC). Haplotypes are grouped on the basis of G/T at the 5ʹ splice site of intron 1 and A/G SNP generating a premature start codon in the 5ʹ-UTR of GBSS1. Values are means ± standard deviation. Means that do not share a letter are significantly different at P = 0.05 based on Tukey’s test. IND: indica; IND (Tongil-type): indica (Tongil-type); TEJ: temperate japonica; TRJ: tropical japonica.

Interaction between SBE3 and GBSS1

Since high phenotypic variation in AC was observed in the RIL population and was affected by the activities of SBE3 and GBSS1, we further analyzed the possible interaction of these two genes in the F₂ population. To detect the G/T polymorphism at the 5ʹ splice site of intron 1 of GBSS1 among the parental lines and F₂ plants, the CAPS marker CS06_001 was PCR amplified followed by digestion with AccI. Amplified fragments containing the G allele are cleaved by AccI, yielding 56 bp and 128 bp fragments, while amplified fragments containing the T allele are not cleaved, yielding the intact fragment size of 184 bp. The CAPS marker from the cultivar Hwayeong was not digested by AccI, while that from Dodamssal and two selected RILs used as parents for F₂ population development (CR2121 and CR2138) was digested. Genotyping revealed that Dodamssal and its ancestor, Gimcheonaengmi, which is a weedy rice variety, carry the same G allele at the 5ʹ splice site of GBSS1 intron 1, with an AC of 17.16% (Supplementary Fig. S1A). This result provides evidence that the G allele was inherited by Dodamssal from its ancestor, Gimcheonaengmi. Hence, the ancestral allele from Gimcheonaengmi might be beneficial for crop improvement of cultivated varieties. Out of 210 F₂ plants, 66 had the G allele, 42 had the T allele, and 102 were heterozygous (G/T allele).

To detect the T/C polymorphism at exon 16 of SBE3 among the parental lines and F₂ plants, the CAPS marker CS02_001 was PCR amplified followed by digestion with SpeI. Amplified fragments containing the T allele are cleaved by SpeI, yielding 246 bp and 371 bp fragments, while amplified fragments containing the C allele are not cleaved, yielding the intact 617 bp fragment. The CAPS marker from Dodamssal and the two selected RILs was not digested by SpeI, but that from Hwayeong was cut (Supplementary Fig. S1B). Out of 210 F₂ plants, 30 of them had the C allele, 64 had the T allele, and 116 were heterozygous (T/C allele). Thereafter, the F₂ population was subjected to interaction analysis using one-way ANOVA and regression analysis and was classified based on the T/C SNP at exon 16 of SBE3 and the G/T SNP at the 5ʹ splice site of GBSS1 intron 1. The classification was as follows: HHHH (Hwayeong homozygous (HH) at qAC2 and qAC6, respectively); HHHD (HH at qAC2 and heterozygous (HD) at qAC6); HHDD (HH at qAC2 and Dodamssal homozygous (DD) at qAC6); HDHH (HD at qAC2 and HH at qAC6); HDHD (HD at qAC2 and qAC6, respectively); HDDD (HD at qAC2 and DD at qAC6); DDHH (DD at qAC2 and HH at qAC6); DDHD (DD at qAC2 and HD at qAC6); and DDDD (DD at qAC2 and qAC6, respectively) (Fig. 6). The AC between DDHH (20.91%) and HHDD (20.66%) was not significantly different. However, the grain appearance varied. Translucent endosperm was observed in HHHH and HHDD, which is similar to the seed characteristic of Hwayeong. In contrast, opaque and chalky endosperm was observed in DDHH and DDDD, which is similar to the seed cha-racteristic of Dodamssal and the two selected RILs (Supplementary Fig. S2). The AC of HHDD and DDHH was comparatively elevated by the activity of the G allele at the 5ʹ splice site of GBSS1 intron 1 and the C allele SNP at exon 16 of SBE3, respectively. These initial findings were manifested in HHHH with low AC (14.51%) and DDDD with high AC (34.76%). The gene interaction between the two QTLs was significant (P < 0.0001). The additive effect of the Dodamssal alleles at qAC2 and qAC6 was 6.51 and 6.08, respectively, while the dominance deviation at qAC2 and qAC6 was 0.23 and 0.25, respectively, indicating that the action of the genes at both loci is additive in regulating AC. The results suggest that SBE3 and GBSS1 act additively and functionally complemented each other. Hence, our initial findings suggest the existence of interaction between SBE3 and GBSS1 in the F₂ population.

Figure 6. Interaction effect on amylose content of T/C SNP at exon 16 of SBE3 on chromosome 2 and G/T SNP at the 5ʹ splice site of intron 1 of GBSS1 on chromosome 6. Error bars indicate mean ± standard deviation. HH: Hwayeong homozygous; HD: Heterozygous; DD: Dodamssal homozygous. Additive effect: (DD-HH)/2; Dominance effect: HD ‒ (DD + HH)/2. Degree of dominance: dominance effect/additive effect, ^aR²: Coefficient of determination, ^bInteraction between qAC2 and qAC6, ^cTotal phenotypic variance was determined by regression analysis.

Starch-related traits such as AC and RS content are characteristic of endosperm having complex genetic and environmental controlling mechanisms since the development and formation of seed endosperm is primarily controlled by triploid interactive alleles (Ni et al. 2011). In our present study, an RIL mapping population was used to identify QTLs controlling AC and RS content. Dodamssal alleles contributed to the increased RS content and AC.

RS is a carbohydrate that is resistant to enzymatic hydrolysis; hence, it resists digestion in the small intestine and is fermented by microorganisms in the large intestine to produce short-chain fatty acids (Englyst et al. 1982; Yang et al. 2016). In the present study, the QTL associated with RS content was identified on chromosome 2, allelic to SBE3, and was responsible for the high phenotypic variation in the RIL population in the two-year field experiment. A single nucleotide substitution occurred within the coding region of this allele (T in Hwayeong to C in Dodamssal). This missense mutation in codon 599 caused an amino acid substitution of leucine to proline in Dodamssal. This amino acid change probably affects the local conformation and three-dimensional structure of the SBE3 protein (Yang et al. 2012). Since SBE3 is involved in the formation of short carbohydrate chains within amylopectin, this missense mutation may lead to extra-long branched chains as a result of reduced branching frequency; thus, RS content will increase. Our result is supported by a previous study where a missense mutation in SBE3 in the F₂ population derived from a cross between Jiangtangdao 1 and Milyang 23 explained 60.4% of the RS content variation (Yang et al. 2016).

Amylose content affects the eating and cooking quality of rice grains. While the Wx locus encoding GBSS1 is the major locus controlling AC synthesis, previous studies have indicated that amylose is under the control of other loci; however, the majority of these loci have not yet been identified (He et al. 1999; Aluko et al. 2004; Wambugu et al. 2018). Therefore, the discovery and identification of other candidate genes will be of great importance as they will likely offer new targets for AC modification. Two QTLs associated with AC were identified on chromosomes 2 and 6. The Dodamssal alleles at qAC2 and qAC6 predominantly contributed to the increased level of AC. However, the Dodamssal allele at qAC2 was not as competitive as the Hwayeong allele that was clearly observed in the RIL and F₂ populations. Most likely, the Dodamssal allele at this locus behaved in such a way that locked alleles into favorable genotypes that allowed “allele selection,” thus affecting the heritability and distribution of genotypes in the population (Neher and Shraiman 2009). The LOD scores of qAC2 and qAC6 were relatively high, providing more evidence to support the presence of a QTL. qAC6, which appears to be allelic to GBSS1, had a phenotypic variation of 55.4% in 2018, while qAC2, which appears to be allelic to SBE3, demonstrated 34.1% phenotypic variation in 2017. These results suggest that AC variation in the RIL population was not only controlled by the Wx locus encoding GBSS1, which is mainly involved in amylose synthesis, but also other candidate genes whose function in amylose synthesis is still unknown.

Whole genome sequencing of parental lines led to the identification of two SNPs in GBSS1. Specifically, two SNPs on chromosome 6 at positions 1765761 and 1765799 were detected between the parental lines, one lying at the 5ʹ splice site of intron 1 and the other creating a premature start codon in the 5ʹ-UTR of GBSS1, respectively. In the 5ʹ splice site of GBSS1 intron 1, the Wx^a allele, which is predominantly seen in indica rice, possessed the G SNP, while the Wx^b allele, which mostly occurs in japonica rice, had the G to T mutation. The two Wx alleles encode different levels of GBSS1, which is involved in AC synthesis. It is noteworthy that HAP1-2 (composed of five tropical japonica and 10 Korean weedy and landrace rice accessions) and HAP3 (composed of temperate japonica weedy accessions) had the G allele in the 5ʹ splice site of intron 1 of GBSS1. Recent studies have indicated that Korean weedy rice was formed from the hybridization of modern indica/indica or japonica/japonica and not from wild rice, and it possesses many valuable traits or haplotypes that may be helpful to improve cultivated rice (Cho et al. 1995; He et al. 2017). Altogether, these results provide support that the G allele may have been prevalent and maintained in temperate japonica weedy rice. It has been difficult and time consuming to introduce desirable indica traits into japonica due to reproductive barriers (Chung and Heu 1980). These obstacles could be overcome by em-ploying japonica weedy rice or landrace rice harboring the Wx^a allele (G allele) in breeding programs to manipulate the AC in cultivated japonica rice (Mo et al. 2014; Park et al. 2019). In contrast, HAP2-2 (composed of temperate japonica and tropical japonica) possessed the T allele. A previous study indicated that the Wx^a allele in indica had differentiated from the Wx^b allele in japonica by the G to T mutation. This mutation producing Wx^b may have increased rapidly in japonica by artificial selection (Hirano et al. 1998). Selection for the Wx splice-site mutation, which played a key role in the evolution of the non-glutinous temperate japonica group and glutinous rice varieties, is associated with a selective sweep. A previous study indicated that patterns consistent with a selective sweep in temperate japonica associated with the intron 1 splice-donor site mutation were observed across a 500-kb genomic region centered on Wx (Olsen et al. 2006). It is notable that HAP2-1 had the T SNP in the 5ʹ splice site of GBSS1 intron 1. Tongil-type rice in the HAP2-1 group derived from a cross between indica and japonica most likely inherited the T allele from japonica and was developed based on consumer preference for a soft and sticky texture of japonica cooked rice without sacrificing the high-yield potential of indica (Kim et al. 2014; Mo et al. 2014). In contrast, haplotype analysis revealed that the G allele of the premature start codon in the 5ʹ-UTR of GBSS1 might be involved in amylose synthesis, as manifested in comparable AC between HAP1-1 and HAP3. In addition, the AC between HAP1-1 and HAP1-2 differed significantly even if the G allele at the 5ʹ splice site of intron 1 and the A allele of the premature start codon were both present. This might be due to the effect of other starch synthesis-related genes and environments controlling AC. In contrast, the difference in AC values between HAP1-2 and HAP3 was insignificant, showing opposing effects of the A and G alleles on HAP1-2 and HAP3, respectively. This initial finding is not yet conclusive since the sample size of HAP3 was small. Further studies, including increasing the sample size, should be conducted to elucidate and verify the likelihood of association of the A/G SNP on amylose synthesis.

To determine possible interactions between SBE3 and GBSS1, genotyping using CAPS markers was carried out on 210 F₂ plants. The CAPS marker for GBSS1 in cultivar Hwayeong was not digested by AccI due to loss of the restriction enzyme site, while Dodamssal was cut, resulting in high AC. Dodamssal and its ancestor, Gimcheonaengmi, carry the same G allele in the 5ʹ splice site of GBSS1 intron 1. This supports the finding that the splice-donor mutation is prevalent in japonica cultivars, but rare or absent in indica, landrace, or weedy rice cultivars (Olsen et al. 2006). In contrast, the T allele in Hwayeong resulted in low AC. A previous study showed that Wx^b (T allele) at the splice donor site of the first intron reduces the efficiency of GBSS1 mRNA processing and causes low levels of the mature GBSS1 transcript and AC (Hoai et al. 2014). The SBE3 CAPS marker in Dodamssal was not digested by SpeI because a point mutation resulted in the loss of the re-striction site, while that of Hwayeong was cut. Since SBE3 plays a leading role in amylopectin formation, the C SNP in Dodamssal may have reduced SBE3 activity, leading to elevated AC. A previous study indicated that the proportion of amylopectin short chains (DP 6-12) of Dodamssal was significantly lower (20.31) than that of other cultivars (Sim et al. 2015). In rice, it has been reported that apparent AC increases 29-35% when SBE3 is deficient (Mizuno et al. 1993; Yang et al. 2012). The inactivation of SBE3 in rice is traditionally associated with elevated apparent AC, increased gelatinization temperature, and a decreased proportion of short amylopectin branches (Nishi et al. 2001; Burtado et al. 2011). In the present study, the AC of the RILs and F₂ plants revealed that AC variation is not only determined by the action of GBSS1 but also by the relative activity of SBE3. In the F₂ population, the lowest AC of HHHH was associated with the T allele at exon 16 of SBE3 and the T allele at the 5ʹ splice site of GBSS1 intron 1. The AC value between HHDD and DDHH was not significantly different, but they had distinct grain appearances. These results suggest that the G allele at the 5ʹ splice site of intron 1 of GBSS1 (HHDD) and the C allele at exon 16 of SBE3 (DDHH) act in an additive manner, functionally complementing each other. However, the difference in grain appearance revealed that the G allele at the 5ʹ splice site did not affect the starch granule morphology of HHDD while the C SNP at exon 16 of SBE3 affected the seed morphology of DDHH. The nucleotide substitution in SBE3 causing the missense mutation probably altered the grain appearance, resulting in the opaque and chalky endosperm of DDHH. Previous studies indicated that the amylose- extender (ae) mutant lacking SBE3 generates a white-core endosperm with a modified fine structure of amylopectin (Nishi et al. 2001; Zhang et al. 2018). Among the genotype groups, DDDD had the highest AC of 34.76%. The G allele at the 5ʹ splice site of intron 1 increased the efficiency of GBSS1 mRNA processing while the C allele at exon 16 of SBE3 probably freed more space for the synthesis of amylose by reducing the synthesis of amylopectin and/or the degree of branching (Wang et al. 2018). The results demonstrated that the two QTLs have different mechanisms in regulating AC and their effects work together to influence AC variation. Thus, significant interaction was observed between the two QTLs. This could be possibly attributed to the effect of Hwayeong allele at qAC2. This result may indicate the existence of gene interaction and may serve as a genetic basis for the fine-tuning and modification of AC.

The QTLs identified in the present study provide information on genes that are candidate targets for AC modification. In addition, the present study suggests that AC is determined not only by the activity of GBSS1 but also by the relative activity of the amylopectin synthesizing enzyme SBE3. Furthermore, our results provide preliminary evidence of interactions between SBE3 and GBSS1 for amylose synthesis. Further studies on the interactions between SBE3 and GBSS1 should be carried out to increase understanding on the mechanism(s) involved at both the molecular and biochemical level.

This work was carried out with the support of the Basic Science Research Program through the National Research Foundation (NRF) of Korea funded by the Ministry of Education (NRF-2017R1A2B2007554). Data for amylose content were in part from Korea genebank database (https://genebank.rda.go.kr:2360).