1. Introduction
Eating quality is generally affected by cohesiveness, tenderness and gross of cooked rice, and could be mainly represented by amylose content (AC), gel consistency (GC), gelatinization temperature (GT) and protein content in rice (Oryza sativa L.) (Juliano et al., Reference Juliano, Onate and Mundo1965; Bett-Garber et al., Reference Bett-Garber, Champagne, McClung, Moldenhauer, Linscombe and McKenzie2001). AC is related to the appearance and texture of rice and it affects cooking and eating quality. GC is responsible for softness in rice, whereas GT is a physical characteristic responsible for cooking time and the capacity of absorbing water when cooking (Sun et al., Reference Sun, Hao and Lin2006). Rice starch viscosity profile, also known as the rapid viscosity analysis (RVA) profile, could greatly reflect the physiochemical properties, palatability and cooking and eating quality by simulating the cooking process (Cai et al., Reference Cai, Liu, Wang, Zhang, Zhang, Yang and Tang2011), and was demonstrated to be closely related to AC and GC (Hu et al., Reference Hu, Zhai, Tang and Wan2004). AC and RVA profiles are important physical and chemical indices that were used to reflect the rice-eating quality (Shu et al., Reference Shu, Wu, Xia, Gao and McClung1998; Bao et al., Reference Bao, He, Xia, Chen and Zhu1999, Reference Bao, Shen and Xie2004; Roferos et al., Reference Roferos, Butardo, Fitzgerald and Juliano2008).
Among the four main elements of eating quality, the former three (AC, GC and GT) are closely related to starch properties. Thus, one may reasonably propose that starch synthesis-related genes (SSRGs) play an important role in determining rice-eating quality. From the primary carbon accumulation in the Calvin cycle to the final products, many enzymes and regulators contribute to this intact pathway. Among them, ADP-glucose pyrophosphorylase (AGPase; EC 2.7.7.21) is a critical rate-limiting enzyme with an allosteric regulatory function (Kato et al., Reference Kato, Shinmura and Taniguchi2007). Upstream enzymes mainly affect rice yield, whereas those downstream greatly influence rice quality (Bao et al., Reference Bao, Lu, Yang, Zhang, Shao, Corke and Sun2012). Starch synthases (SSs) and starch debranching enzymes also affect starch components and structure, which further affect eating quality. Most enzymes in the starch pathway have many isoforms, which perform similar functions in a temporal or spatial way (Ohdan et al., Reference Ohdan, Francisco, Sawada, Hirose, Terao, Satoh and Nakamura2005). Different gene regulation mechanisms, different gene expression levels or different enzyme activities could lead to different eating quality. Good eating quality is a product of collaborative reactions regulated by multiple enzymes and regulators.
Currently, increasing attention is being paid to breeding strategies of rice with great eating quality (Tan et al., Reference Tan, Li, Yu, Xing, Xu and Zhang1999; Sun et al., Reference Sun, Hao and Lin2006; Lestari et al., Reference Lestari, Ham, Lee, Woo, Jiang, Chu, Kwon, Ma, Lee, Cho and Koh2009). Due to natural mutation and selection, abundant sequence differences exist among multiple accessions from diverse resources (Jangsutthivorawat & Volkaert, Reference Jangsutthivorawat and Volkaert2009) that were characterized by molecular markers, such as single-nucleotide polymorphisms (SNPs), insertion-deletions (InDels) and simple sequence repeats (SSRs). Natural variation mining in these genes or regulator sequences will assist us to locate important alleles for eating quality using association mapping (AM) (Cardon & Bell, Reference Cardon and Bell2001). Tran et al. (Reference Tran, Daygon, Resurreccion, Cuevas, Corpuz and Fitzgerald2011) reported that a C/T transition in the tenth exon of the Wx gene showed a strong association with GC using a recombinant inbred line population. Sun et al. (Reference Sun, Abdula, Lee, Cho, Han, Koh and Cho2011) confirmed that several SNPs in starch branching enzyme 1 (SBE1), soluble SS 1 (SSS1) and SSS2a were significantly associated with AC, alkali digestion value, setback viscosity (SBV), consistency viscosity and pasting temperature (PT). These SNPs provide a possibility to enrich key alleles in one cultivar using a crossing strategy.
To find out those sequence variations contributing to good eating quality, AM is a good candidate approach to detect significant correlations between genotypes and phenotypes in a specified or unrelated mapping population on the basis of linkage disequilibrium (Remington et al., Reference Remington, Ungerer and Purugganan2001; Zhang et al., Reference Zhang, Hao, Ren, Chang, Liu and Jing2011; Zhao et al., Reference Zhao, Tung, Eizenga, Wright, Ali, Price, Norton, Islam, Reynolds, Mezey, McClung, Bustamante and McCouch2011). The most attractive aspect of AM lies in its ease when establishing mapping populations (Jin et al. Reference Jin, Lu, Xiao, Sun, Corke and Bao2010; Li et al., Reference Li, Yan, Agrama, Jia, Shen, Jackson, Moldenhauer, Yeater, McClung and Wu2011). The rice core set, also known as core collection, represents the largest genetic diversity within the smallest number of accessions (Brown, Reference Brown1989). It has proved to be one of the most valuable resources for genetic research and molecular breeding, especially for allele mining (Zhang et al., Reference Zhang, Li, Xue, Han and Deng2008; de Oliveira Borba et al., Reference de Oliveira Borba, Brondani, Breseghello, Coelho, Mendonça, Rangel and Brondani2010). Therefore, more sequence variations are expected to be present in worldwide diverse varieties. The discovery of those sequence variations or alleles with a partial contribution to phenotypes should be helpful in rice breeding using the marker-assisted strategy. In this study, we investigated sequence variations among 10 SSRGs, and evaluated the AC and RVA profiles in a heuristic core set of 166 accessions, aimed at finding some effective nucleotide sequence variations closely related to AC as well as RVA profiles, which would provide basic guidance for breeding good eating-quality rice.
2. Materials and methods
(i) Plant material
Morphological and molecular genetic diversities of rice (O. sativa L.) were evaluated in a heuristic core set of 166 accessions (Chung & Park, Reference Chung and Park2009). This core set collection was selected by the advanced maximization strategy implemented through a modified heuristic algorithm using Powercore program (Kim et al., Reference Kim, Chung, Cho, Ma, Chandrabalan, Gwag, Kim, Cho and Park2007) from 4406 worldwide rice accessions, which were obtained from Genebank of the Rural Development Administration (RDA), Republic of Korea. Plants were grown in the Kongju National University farm field in a natural environment during 2009. Young plant leaves were collected at the grain-filling stage for each accession and stored at −80°C until genomic DNA extraction. The mature seeds for each accession were harvested in bulk for phenotypic data analysis. By removing those accessions with lots of missing genotype and phenotypic data (>25%), 104 accessions originated from 28 countries over the world in the heuristic core set were finally chosen for association analysis in this study (see Supplementary Table S1 available at http://journals.cambridge.org/GRH).
(ii) Phenotypic performance
AC determination was performed using a method based on colorimetric analysis described by Cho et al. (Reference Cho, Lee, Kim, Yu, Lee, Hong, Kim, Ma, Kwon, Kang, Lee, Gwag, Kim and Park2008). Starch viscosity profiles were determined on a Rapid Visco Analyzer (serial-3; Newport Scientific, Warriewood, NSW, Australia). A total of 3-g starch from each accession was dispersed in 25 ml of distilled water in the viscometer test canister. The time–temperature regimen used was set up as follows: idle temperature 50°C for 1 min, heated from 50 to 95°C in 4·7 min, then held at 95°C for 2·5 min. The sample was subsequently cooled down to 50°C in 3·7 min, followed by 2 min at 50°C. The profiles of RVA parameters, including peak viscosity (PV), trough viscosity (TV), breakdown viscosity (BDV=PV−TV), final viscosity (FV), setback viscosity (SBV=FV−PV) and PT were investigated for the core set of 166 accessions. For all traits, the Kolmogorov–Smirnov test was applied to determine a normal distribution at a significance of 0·05.
(iii) Genotype data
Genomic DNA was extracted using the cetyltrimethylammonium bromide (CTAB) method. A total of ten genes (six AGPase units, GBSSI, pullulanase, SSSIIa and one SSSIIa-like gene) were selected based on their important roles in the starch synthesis in rice endosperm. The amplicons were chosen by searching NCBI nucleotide database using blastn (http://blast.ncbi.nlm.nih.gov), and the primers were designed using the corresponding genomic sequences retrieved from NCBI (Table 1). PCR was performed using H-Taq DNA polymerase (Solgent, Daejeon, Republic of Korea) on a GenePro Thermal Cycler (Bioler, Hangzhou, China). Reaction components were as follows: 2·5 μl of H-Taq buffer (10×), 0·5 μl of dNTP (10 mM), 1·0/1·0 μl of forward/reverse primers (10 pmol), 0·25 μl of H-Taq DNA polymerase, 1·0 μl of genomic DNA (50 ng/μl) and 18·75 μl of ddH2O. Cycling profiles were set as follows: denaturation at 95°C for 15 min; 40 cycles of denaturation at 95°C for 20 s, specific annealing temperature (Table 1) for 40 s and extension at 72°C for 1 min; and a final extension at 72°C for 3 min. PCR products were purified using the PCR Purification Kit (Solgent, Daejeon, Republic of Korea), and then sent to Solgent Co. Ltd (Daejeon, Republic of Korea) for sequencing based on a ABI 3730 sequencing platform (Applied Biosystems, Foster City, CA, USA). Rare alleles were re-sequenced with additional independent PCR products.
T m, melting temperature.
Gene locus: from rice genome annotation project (Pseudomolecules version 6.1, http://rice.plantbiology.msu.edu/).
(iv) Candidate sequence variation discovery and genetic analysis
To discover the sequence variations, Sequence Analysis software (v. 5.4; Applied Biosystems) was used for base calling and sequence masking and Variant Reporter (v. 1.1; Applied Biosystems) was employed to align the sequences and find variations. Basic statistics were calculated using the PowerMarker genetic analysis package (v. 3.25; Liu & Muse, Reference Liu and Muse2005) to measure the diversity at each locus, including variation frequency, gene diversity and polymorphism information content (PIC). The observed heterozygosity, expected heterozygosity and Shannon's information index were calculated using the POPGENE software (v. 1.32; Yeh & Boyle, Reference Yeh and Boyle1997). Previously, we characterized this heuristic core set using 170 genome-wide SSR markers (see Supplementary Table S1). To reduce spurious deviations, a dendrogram was constructed from a distance matrix based on the SSR genotypes using the unweighted pair group method and the arithmetic averages (UPGMA) algorithm embedded in the PowerMarker program, and edited using the MEGA5 program (Tamura et al., Reference Tamura, Peterson, Peterson, Stecher, Nei and Kumar2011).
(v) AM
The hypothesis behind associating those sequence variations with AC and RVA profiles in the presence of population structure was tested using a general linear model (GLM) and a mixed linear model (MLM), as described by Yu et al. (Reference Yu, Pressoir, Briggs, Vroh Bi, Yamasaki, Doebley, McMullen, Gaut, Nielsen, Holland, Kresovich and Buckler2006), in the TASSEL program (v. 2.1; Bradbury et al., Reference Bradbury, Zhang, Kroon, Casstevens, Ramdoss and Buckler2007). To reduce possible spurious associations, the population structure matrix (Q matrix) and the relative kinship matrix (K matrix) were calculated based on the 170 genome-wide SSR data (Supplementary Table S1). The Q matrix at k=4 was identified by running the Structure program (v. 2.3.3; Hubisz et al., Reference Hubisz, Falush, Stephens and Pritchard2009) following a method described previously (Falush et al. Reference Falush, Stephens and Pritchard2003), whereas the Loiselle kinship coefficients (Loiselle et al., Reference Loiselle, Sork, Nason and Graham1995) between accessions were estimated using the SPAGeDi program (v. 1.2 g; Hardy & Vekemans, Reference Hardy and Vekemans2002). Only markers with a variation frequency of 5% or more were included in the association analysis. For GLM, to account for type I error bias, the P-value was adjusted for multiple tests with 1000 permutations using a procedure proposed by Whitt & Buckler (Reference Whitt, Buckler and Grotewold2003). The P-value determined whether a locus was associated with a marker, and the Rsq_marker evaluated the magnitude of the locus effects. For MLM, a correction for multiple testing was applied using the false discovery rate (FDR) method to confirm the association significance (Storey & Tibshirani, Reference Storey and Tibshirani2003).
3. Results
(i) Statistics and distribution of phenotype data
Detailed information of phenotypic data means, ranges and correlations for amylose and RVA parameters (PV, TV, BDV, FV, SBV and PT) are summarized in Table 2. Among the correlations between each other, more than half of them were nominally significant at P<0·01. Notably, PV showed significant positive correlations with TV (0·879) and FV (0·759). The correlation (0·911) between FV and TV confirmed the most significantly positive, whereas BDV showed the most significant negative correlation (−0·715) with SBV. The distribution of all traits except for PT conformed to a classical Gaussian distribution at a significance cut-off of 0·05 in the Kolmogorov–Smirnov test, which was proved to be suitable for AM analysis (Fig. 1 and Table 2).
AC, amylose content; PV, peak viscosity; TV, trough viscosity; BDV, breakdown viscosity; FV, final viscosity; SBV, setback viscosity; PT, pasting temperature; StDev, standard deviation.
** Correlation is significant at the 0·01 level (two-tailed).
* Correlation is significant at the 0·05 level (two-tailed).
(ii) Sequence variation discovery in SSRGs
Among the 10 amplicons sequenced, the sequenced length ranged from 377 (OsAGPS2) to 834 bp (OsPUL3). Overall, 86-sequence variations were discovered from nine amplicons of the ten genes sequenced, including 79 SNPs, six InDels and one SSR. No SNP loci were found in the OsAGPL1 gene amplicon. Specifically, 25 and 61 variations were separately found in the intronic and exonic parts of the nine amplicons. Among the 61 exonic variations, 41 (47·67% in 86) in exon regions should have resulted in amino acid changes by analysing the genetic code changes. Most SNPs were A/G (39·53%) and C/T (34·88%) transition substitutions (Table 3). Notably, 47 of the 86 total sequence variations were discovered from the OsSSllaChr7 gene on chromosome 7. InDels were found in the first exon and first intron of OsAGPL4, 5-untranslated region of OsAGPS2 and the first exon of OsSSIIchr6, in addition to two InDels in the OsGBSSI gene. Notably, the two InDels in the OsGBSSI were found separately in first exon (23 bp insertion) and first intron (16 bp deletion), both of which were confirmed to be rare variations (variation frequency <0·05). In particular, the 23 bp InDel was expected to result in a frame shift during translation process. The polymorphic SSR appeared in the second exon of OsSSIIchr6.
Intron Var, number of variation in introns; Exon Var, number of variations in exons; AAc Var, number of variations presumed to result in amino acid changes in the exon parts; InDel, insertion-deletion; SSR, simple sequence repeat.
(iii) Sequence polymorphisms
Each sequence variation locus revealed two variation types through a total of 86 variation loci among the 104 rice accessions (see Supplementary Table S2 available at http://journals.cambridge.org/GRH). According to the distribution of variation loci, rare variations (frequency <0·05) accounted for 21·51%, while the common (0·05<frequency <0·2) and abundant variations (frequency >0·2) made up 6·39 and 72·10%, respectively. These results indicated the presence of a relatively large proportion of rare and abundant variations, and most variations were of a low frequency among the rice studied (Fig. 2). The frequency of heterozygous alleles in one individual was rare, but it existed. The averaged observed heterozygosity was obviously lower than the averaged expected heterozygosity, indicating the presence of heterozygosis deficit probably due to forces such as inbreeding. The genetic diversity for all variation loci varied from 0·0190 to 0·4936, with an average of 0·2256. PIC values ranged from 0·0189 to 0·3718, and averaged 0·1848 (Fig. 3). The broad ranges of genetic diversity and Shannon's information index represented the multiple-level inheritable variations in the selected rice accessions.
(iv) Population structure
A model-based clustering method was performed using our previous 170 genome-wide SSR genotyping data in 104 accessions (Supplementary Table S1) to infer the exact population number using Structure program. Estimated likelihood [LnP(D)] values for a given k in five independent runs showed consistent results, but the LnP(D) distribution did not indicate a clear plateau for the true k. Therefore, another ad hoc quantity Δk was employed to overcome the difficulties in interpreting the real k value (Evanno et al., Reference Evanno, Regnaut and Goudet2005). The ΔK showed a clear peak at the true value of k=4 (Fig. 4a), which was in accordance with previous data. A sum of 93 accessions was assigned to four subpopulations (Fig. 4b), of which each shared at least 75% ancestry, leaving 11 accessions as admixtures. The size of Pop2 and Pop3 were relatively small, including only 8 and 19 individuals, respectively. This was supported by the alpha parameter (α=0·0464), indicating that most accessions in each subpopulation originated from one primary ancestor, except the 11 admixed individuals (Falush et al., Reference Falush, Stephens and Pritchard2003; Ostrowski et al., Reference Ostrowski, David, Santoni, McKhann, Reboud, Le Corre, Camilleri, Brunel, Bouchez, Faure and Bataillon2006). The genetic distance-based UPGMA tree supported the model-based population structure assignment (Fig. 4c).
(v) Association analysis
Marker–trait associations (P<0·001) for traits were evaluated in 50 variation loci after removing 36 loci with rare frequency using the program TASSEL 2.1. A total of four associations were detected between three loci and three trait indices in GLM and/or MLM (Table 4). Notably, one 12-bp InDel (TCCGCCGCCGCC) in the first exon of the OsALPL4 gene were significantly associated with BDV at P<0·001 in both GLM and MLM, with an interpretation of up to 16·6% of the trait variations. Moreover, one SNP with an A/G transition, which should result in the encoded amino acid from asparagine to aspartic acid, from the 33rd nucleotide of OsAGPS1's third exon was significantly associated with two trait indices (AC at P<0·001 for both models, and FV at P<0·001 for MLM model only), explaining up to 15·6% of the phenotype variation. Some sequence variations in introns were also detected to be significant in association with some traits. A 8-bp InDel (cgcgtgcg) in the first intron of OsAGPL4 gene was closely associated with BDV at P<0·001 in GLM and MLM, mostly because this InDel showed exactly the same insertion- and deletion-types with the 12-bp InDel in the first exon described above in each investigated accession.
AAc, types of amino acid changes; P_GLM, adjusted P-values with 1000 permutations; P_MLM, P-values that were significant in FDR test; amino acid code: S, serine; A, alanine; N, asparagine; D, aspartic acid.
(vi) Allelic effects
Based on the population structure revealed by the 170 genome-wide SSR genotyping data, those 104 accessions investigated were assigned into four subgroups, which was consistent with our previous data using 130 core set accessions. With regard to the variations with highly significant associations (P<0·001), the allelic effect showed a subpopulation-specific pattern to some extent in Pop2 and Pop3, mostly because of the small population sizes. In OsAGPL4, a 12-bp deletion between the 106th and 117th nucleotide (TCCGCCGCCGCC) that encodes four amino acids (Ser-Ala-Ala-Ala) at the first exon was strongly associated with a dramatic decrease in BDV, and those accessions with the deletion were all assigned only to Pop1 (Fig. 5a). In OsAGPS1, a transition from A to G at the 133rd nucleotide of the third exon that changed the amino acid from asparagine to aspartic acid was closely related to a slight AC increase (Fig. 5b) and a significant FV decrease (Fig. 5c). According to the differences observed in AC and FV, those accessions with A and G seemed to be present only in Pop 3 and Pop 1 plus Pop 2, respectively, and then mixed in Pop 4.
4. Discussion
Eating quality represents an important index reflecting the rice quality, which was greatly controlled or regulated by those genes involved in starch and protein metabolism network. Each gene may play a partial role in determining eating quality by affecting carbohydrate partitioning, whereas some gene(s) may exert main-effect forces on the eating quality, such as the Wx gene, but the contributions from other enzymes could not be excluded. In the present study, we characterized 86 sequence-based variations, including SSRs, SNPs and InDels, from the 10 selected SSRG amplicons in the heuristic core set, and then evaluated their association relationship to eating quality indices (i.e. AC and RVA properties).
A large proportion (47·67%) of sequence variations detected in the 10 amplicons investigated was expected to cause amino acid changes, which should have potentially contributed to the relative phenotypes among these core set accessions. However, a large proportion of the total detected variations notably confirmed to be of low frequency, accounting for rare alleles. The effect of rare alleles in the association population mean would be minimal, whereas it could be detected in a biparental population. This portion of rare variations with modest effects could be difficult to detect because they explain only a trivial fraction of the variance in a trait (Hirschhorn & Daly, Reference Hirschhorn and Daly2005). Therefore, to reduce spurious associations, this portion of rare alleles in the population was treated as missing data, and excluded from association tests (Breseghello & Sorrells, Reference Breseghello and Sorrells2006; Liu et al., Reference Liu, Wang, Yao, Zheng and Zhao2010; Lu et al., Reference Lu, Zhang, Shah, Xie, Hao, Li, Farkhari, Ribaut, Cao, Rong and Xu2010). Furthermore, those variations discovered from OsSSIIchr7 comprised over half of the total 86 variations in a 764-bp amplicon, indicating that it is most probably a non-functional gene, despite the fact that it was annotated as a soluble SS-like protein (protein ID: BAC16084.1).
Most previous studies found that AC is greatly determined by Wx, SSI, SSII-3, SSIII-2 and PUL (Roferos et al., Reference Roferos, Butardo, Fitzgerald and Juliano2008; Tian et al., Reference Tian, Qian, Liu, Yan, Liu, Yan, Liu, Gao, Tang, Zeng, Wang, Yu, Gu and Li2009), while RVA profiles were mainly controlled by the Wx gene, encoding GBSSI enzyme on rice chromosome 6 (Bao et al., Reference Bao, He, Xia, Chen and Zhu1999, Reference Bao, Corke, He and Zhu2003). However, in the present study, two InDels from OsGBSSI gene was declared to be rare variations, and should be excluded during AM. Huang et al. (Reference Huang, Wei, Sang, Zhao, Feng, Zhao, Li, Zhu, Lu, Zhang, Li, Fan, Guo, Wang, Wang, Deng, Li, Lu, Weng, Liu, Huang, Zhou, Jing, Li, Lin, Buckler, Qian, Zhang, Li and Han2010) reported a transition SNP (T/C) around 1 kb away from OsGBSSI gene was significantly associated with AC. Based on their resequencing data of 950 rice accessions (http://202.127.18.221/RiceHap2/), only three sequence variations with major frequency were found in the fifth, eighth and ninth exons, indicating that the coding region of OsGBSSI gene was rather conservative. It is reasonable for the difficulties in finding some sequence variations with major frequency in the OsGBSSI coding region.
Instead, an SNP from OsAGPS1 and an InDel from OsAGPL4 were identified as significantly associated with AC and RVA profiles. It is partly supported by the important role of AGPase in starch synthesis in the previous reports. AC could be significantly affected by AGPlar at the step from Glc-1-P to ADP-Glc catalysed by AGPase (Tian et al., Reference Tian, Qian, Liu, Yan, Liu, Yan, Liu, Gao, Tang, Zeng, Wang, Yu, Gu and Li2009). Kato et al. (Reference Kato, Taniguchi and Horibata2010) investigated the nucleotide variations in OsAGPS2 and OsAGPL2, and found that alleles at OsAGPS2 may associate with the grain-filling degree and AGPase activity in the extra-heavy panicle type of rice. Bao et al. (Reference Bao, Lu, Yang, Zhang, Shao, Corke and Sun2012) also reported an InDel in OsAGPL4 and an SNP in OsAGPL2 were associated with grain weight in two environments and in one environment, respectively, by investigating the nucleotide variations in the OsAGP genes.
This paper was dedicated to unveil those desired natural sequence variations in SSRGs significantly associated with eating quality indicators for rice breeding using AM strategy. It does not conflict with those previous studies determining the main-effect gene(s) for eating quality using genome-wide mapping strategy. Undoubtedly, some SSRGs, especially the OsGBSSI gene, are necessary to be conserved through the long evolution history to ensure the plant progenies’ survival (Mason-Gamer et al., Reference Mason-Gamer, Weil and Kellogg1998). The significant associations between gene-based variations in AGPase subunit genes and eating quality do not exclude the importance of other genes including the OsGBSSI (Cui et al., Reference Cui, Wu, Shi, Littell and Wu2006).
This work was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education, Science and Technology (NRF-2010-0022616).
5. Declaration of interest
None.
Supplementary material
The supplementary Tables S1 and S2 are available online at http://journals.cambridge.org/GRH.