Figures
Abstract
Mining elite genes within rice landraces is of importance for the improvement of cultivated rice. An association mapping for 12 agronomic traits was carried out using a core collection of rice consisting of 150 landraces (Panel 1) with 274 simple sequence repeat (SSR) markers, and the mapping results were further verified using a Chinese national rice micro-core collection (Panel 2) and a collection from a global molecular breeding program (Panel 3). Our results showed that (1) 76 significant (P<0.05) trait-marker associations were detected using mixed linear model (MLM) within Panel 1 in two years, among which 32% were identical with previously mapped QTLs, and 11 significant associations had >10% explained ratio of genetic variation; (2) A total of seven aforementioned trait-marker associations were verified within Panel 2 and 3 when using a general linear model (GLM) and 55 SSR markers of the 76 significant trait-marker associations. However, no significant trait-marker association was found to be identical within three panels when using the MLM model; (3) several desirable alleles of the loci which showed significant trait-marker associations were identified. The research provided important information for further mining these elite genes within rice landraces and using them for rice breeding.
Citation: Zhang P, Liu X, Tong H, Lu Y, Li J (2014) Association Mapping for Important Agronomic Traits in Core Collection of Rice (Oryza sativa L.) with SSR Markers. PLoS ONE 9(10): e111508. https://doi.org/10.1371/journal.pone.0111508
Editor: Jauhar Ali, International Rice Research Institute, Philippines
Received: February 3, 2014; Accepted: September 30, 2014; Published: October 31, 2014
Copyright: © 2014 Zhang et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: This work was supported by Fund of the National Natural Science Foundation of China grant 30700494. The funder had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Introduction
As a staple cereal crop, rice (Oryza sativa L.) feeds more than 50% of the world's population [1] and is one of the most important components of human diet in many regions of the world. Thus, genetic improvement of rice for yield is important to the meet food demand of a growing global population. Rice landraces have a greater genetic diversity than elite cultivars (or commercial cultivars) and represent an intermediate stage in domestication between wild rice and elite cultivars [2], which make it easier to be used in rice breeding than wild rice and at the same time still keeping most of the diversity in rice germplasm resource. Therefore, mining elite genes within the germplasm of rice landraces is of importance for the improvement of cultivated rice.
Linkage mapping and association mapping based linkage disequilibrium (LD) are two main methods for locating genes or QTLs. The major limitations of linkage mapping are that only two alleles at any given locus can be studied in bi-parental crosses and a low mapping resolution [3], whereas association mapping promises to overcome the limitations of linkage mapping [4]. Moreover, association mapping identifies QTLs by examining the trait-marker associations and enables researchers to use modern genetic technologies to exploit natural diversity and locate valuable genes in the genome [5].
Association mapping has been widely used in plant research since it was firstly reported in maize [6], [7]. In recent years, association mapping has been applied in Arabidopsis, maize, barley, durum wheat, spring wheat, sorghum, sugarcane, sugar beet, soybean, grape, forest tree species and forage grasses [8] as well as rice [9], [10], [11], [12]. For example, an association mapping was performed with 60 simple sequence repeat (SSR) markers and 114 restriction fragment length polymorphism (RFLP) markers for 12 agronomic traits within 218 inbred lines of rice originating from United States of America (USA) and Asia [13]. An association mapping was performed for five agronomic traits in a population of 103 cultivars using 123 SSRs [14] as well as for grain shape using a collection of 293 accessions of Asian cultivated rice [15]. An association mapping for starch quality traits using both candidate gene-based association mapping and genome-wide association study (GWAS) strategies was performed [16]. More than 3.6 million SNPs were detected by sequencing 517 rice landraces and applied for GWAS for 14 agronomic traits [17]. However, to our knowledge, an association mapping with a high number of SSR markers was seldom performed in the previous studies. Moreover, no earlier research performed an association mapping in one population and at the same time verified the association mapping results in other populations.
The choice of appropriate germplasm to maximize the number of historical recombinations and mutation events (and thus reduce LD) within and around the gene of interest is critical for the success of association analysis [18]. One of the methods to obtain most of the phenotypes is to construct a core collection. A core collection is a subset chosen to represent most genetic diversity of an initial collection with a minimum of redundancies [19], [20], [21]. Core collections facilitate the users to access useful samples of small sizes while still keeping most of the genetic variability contained within the gene pool of a specific crop [22]. The construction of a core collection was widely applied in rice as well as other crops. Thus, a core collection might be an ideal mapping population for association mapping. Some rice core collections have been used as association mapping populations in previous studies [23], [24]. However, the mapping population in the studies mentioned above were two subsets consisting of 547 and 203 accessions chosen randomly from United States Department of Agriculture (USDA) rice core collection which consists of 1790 rice entries, which cannot effectively maintain the genetic diversity in the original collections. Moreover, the number of SSR markers for genotyping was low (72 and 155) in the studies. As far as we know, no earlier research on association mapping based on a core collection of rice landraces was available.
Population structure may cause false positives in association mapping. To overcome this problem, an approach using a mixed-model was proposed for association mapping, which take both population structure (Q) and kinship (K) into account for the reduction of false positives [25]. In recent years, comparisons of different statistical models e.g. Q, Q+K and P+K have been conducted for Arabidopsis [26], sweet sorghum [27], maize [28] and rice [23]. However, false positive might not be absolutely avoided through the aforementioned models. To avoid them, it required that the significant associations identified within one population should be verified in another population [29].
In our previous studies, a rice core collection (Ting's rice core collection) consisting of 150 accessions of rice landraces has been constructed based on 15 quantitative traits and 34 qualitative traits from 2262 accessions of rice landraces of the Ting's collection with an optimal sampling strategy [30]. Moreover, population structure and LD of the rice core collection had been examined in details [31]. In this study, an association mapping was performed for 12 agronomic traits in the Ting's core collection assessed with 274 SSR markers. Moreover, the significant trait-marker associations identified in the population were verified within a Chinese national rice micro-core collection and a collection from a global molecular breeding program. The study aimed to (1) perform association mapping for 12 important agronomic traits in the Ting's core collection and verify some of the mapping results in another two core collections, (2) compare the effectiveness of different statistical models and different significant thresholds for association mapping, and (3) identify desirable alleles of the loci which showed significant trait-marker associations for rice breeding.
Materials and Methods
Plant material
Three rice collections, i.e. Ting's core collection (Panel 1), the Chinese national micro-core collection (Panel 2), and a collection from the core collection of a global molecular breeding program (Panel 3) were used in this study. Panel 1 was collected by the researcher Ying Ting during 1920–1964 from all over China as well as from Korea, Japan, Philippines, Brazil, Celebes, Java, Oceania, and Vietnam. The original collection comprises 7128 rice landraces [32]. The core collection (Panel 1) with 150 accessions was constructed from 2262 accessions of 7128 based on a strategy of stepwise clustering and preferred sampling on adjusted Euclidean distances and weighted pair-group average method using integrated qualitative and quantitative traits [30]. Panel 2 with 197 accessions was provided by China Agricultural University, and Panel 3 with 122 accessions was offered by the International Rice Research Institute (IRRI). The information for each variety is shown in Table S1 in File S1.
Phenotyping
All of the three panels were cultivated at the farm of South China Agricultural University, Guangzhou (23°16N, 113°8E), during the late season (July-November) for two consecutive years (2008 and 2009). A randomized complete block design with three replications was used during each season. The space between rows and between plants was set to 20 and 16.5 cm, respectively. Thirty plants of each variety were grown in three rows with 10 plants per row. For each block, the five plants in the middle position of the second row of each variety were selected so that the marginal effect was avoided. 12 agronomic traits for these plants were investigated. Heading date (HD) was recorded as days from sowing to flowering time when 30% of the individuals of one variety started flowering. Plant height (PH), panicle length (PL), grain length (GL), grain width (GW), flag leaf length (FLL), and flag leaf width (FLW) were measured in centimeters. Seed set rate (SS, %) was the percentage of filled grains divided by the total grains per plant. For 1000-grain weight (1000GW), 100 grains were measured in grams with three replicates and then its average was multiplied by 10. For grain length (GL) and width (GW), ten grains were randomly selected and measured with a digital vernier caliper.
Genotyping
274 SSR markers evenly distributed across the 12 chromosomes of rice were selected to genotype all varieties in Panel 1 (Table S2 in File S1). A total of 23, 25, 24, 22, 21, 22, 21, 25, 23, 24, 23, and 21 of these markers were mapped to chromosomes 1 to 12, respectively. The average distance between the loci in chromosomes 1 to 12 is 7.5 cM, 8.2 cM, 9.4 cM, 7.4 cM, 7.1 cM, 6.3 cM, 5.8 cM, 5.4 cM, 5.2 cM, 4.7 cM, 5.6 cM and 5.3 cM, respectively. Markers which prefix RM were summarized in [33], [34], [35], [36] and those with prefix PSM were summarized in [37]. DNA was extracted using a modified SDS method [38]. The volume of the polymerase chain reaction (PCR) was 10 µl. The profile of the PCR program was as follows: 94°C for 5 mins followed by 29 cycles of 94°C for 1 min, 55°C for 1 min, 72°C for 1 min with a final extension of 5 minutes at 72°C. PCR products were separated in size by 6% polyacrylamide gel electrophoresis and detected by silver staining [39]. A standard marker (100–600 bp, produced by Shanghai Biocolor BioScience & Technolgy Company) was added on each gel as control during the gel run. The size of PCR products were detected by BIO Imagine System with software Genetools from SynGene and were manually re-checked twice [31]. The length of each allele was compared to the standard bands of the standard marker and scored.
Data analysis
Means and standard deviation (SD) for 12 traits were calculated using Excel software. The percentage of phenotypic variation explained by population structure was calculated using a General Linear Model (GLM) with software SPSS 17.0 for Windows (SPSS Inc. Chicago, IL, USA). The broad-sense heritability (H2) was calculated as H2 = /
, where
is the genetic variance,
is the environmental variance. They were calculated using software QGA Station 1.0 (Zhu Jun, Zhejiang University, China). Correlation coefficients between traits were calculated using the software SPSS.
Polymorphism information content (PIC) which measures the extent of polymorphism for marker gene(s) or marker sequence(s) was calculated using the program POWERMARKER V3.25. Software Structure V2.3.1 was used to infer population structure and get Q matrices [40], [41]. During the running, a range of genetic clusters from K = 1 to 15 with the admixture model was examined, and for each K it was replicated 5 times. Each run implemented with a burn-in period of 100,000 steps followed by 100,000 Monte Carlo Markov Chain replicates. Due to the distribution of L(K) did not show a clear cutoff point for the true K, an ad hoc measure ΔK was used to detect the numbers of subgroup. That run with the maximum likelihood was applied to subdivide the varieties into different subgroups based on the maximum membership probability. A Q-matrix was obtained from the membership probability of each variety. Our previous study indicated that there were two distinct subgroups in Panel 1, which were in accordance with the germplasm types of indica and japonica rice [31]. The Q-matrix was used for further association mapping. The Loiselle algorithm was chosen for calculating kinship matrix (K) by software SPAGeDi [42]. Rare alleles with frequency of less than 10% in population were filtered as missing data in association analysis. Quantile–quantile plots were generated for observed against expected −log10 (P) using software SAS version 9.0 (SAS Institute 2002), where observed P values were obtained from association mapping and expected P values from the assumption that no associations happened between marker and trait.
Association analysis was performed using the software TASSEL (www.maizegenetics.net/tassel). For the mixed linear model (MLM) method, both K and Q matrices were incorporated, whereas for the GLM method, only population structure information (Q-matrix) was used as a covariate. Significance of associations between loci and traits were determined by their P values (P<0.05) which were calculated by the statistical models, and the phenotypic variance explained by the significant loci was calculated through analysis of variance (ANOVA). Since MLM method performs better in controlling spurious associations than GLM method [43], we first ranked the significant (P<0.05) association from MLM and then compared the significance of these markers (P<0.05) in the permutation based on GLM association tests. For the comparison, we calculated and used other two significant thresholds (i.e. Minimum Bayes factor (BF) and Bonferroni threshold) besides the P value. BF was calculated using the following formula: BF = −e*P*ln(P) [44], [45]. The Bonferroni threshold [46] was 1/274 = 0.00365, where 274 is the number of association tests for each traits in this study. Duncan multiple comparisons was implemented in SPSS for comparisons of performance of agronomic traits relevant to different alleles of the significant trait-marker associations.
Results
Phenotypic variation
The rice landraces in Panel 1 revealed a wide range of phenotypic variation in 12 agronomic traits (Table 1). Heading date, plant height, 1000-grain weight, flag leaf length, flag leaf length/width, and panicle numbers per plant showed similar distributions in both two years (Figures S1–S6 in File S1). On average about 12.4% of phenotypic variation was influenced by population structure. The broad-sense heritability ranged from 74.8% (1000GW) to 99.8% (GW) for these traits.
Phenotypic correlation analysis
Extremely significant (P<0.01) positive correlations both in 2008 and 2009 were found between HD and PH, PH and PL, FLL and FLL/FLW, PL and FLL, PL and FLW, GL and GL/GW, GW and 1000GW, GL and 1000GW, HD and FLW, PH and FLL, SS and 1000GW, PH and FLW (Table 2). Extremely significant (P<0.01) negative correlations in both two years were found between HD and 1000GW, GW and GL/GW, FLW and FLL/FLW, FLW and PN.
Relative kinship among individuals in the three panels
In Panel 1, about 55% of pairwise kinship estimates were zero and only 4.73% of pairwise kinship coefficient were larger than 0.5, indicating that these varieties were unrelated (Figure 1). In Panel 2 and 3, 55.9% and 60.4% of pairwise kinship coefficient were larger than 0.5, respectively (Figure S7 in File S1), indicating that these varieties have certain kinship relationship.
The height of the black bar represents the percentage of varieties in different ranges of kinships.
The effect of controlling type I error using MLM
Observed versus expected P values for each trait-marker association were plotted to assess the control of type I errors. Uniform distributions between the observed and expected P values for all traits were observed, and were demonstrated by similar distributions in two years (Figures 2 and 3). As the deviations from the expectation demonstrated that the statistical analysis may cause spurious associations [28], our result indicated that the false positives were well controlled in the MLM method in this study.
The blue symbol the represents expected P-values, and the red symbol represents the observed P-values.
Blue symbol represents expected P-values, and red symbol represents observed P-values.
Trait-marker associations
152 significant (P<0.05) trait-marker associations were found using the GLM model for the 12 agronomic traits both in 2008 and 2009, and 15 (∼10%) of 152 trait-marker associations were detected in the previous studies (Table 3). Furthermore, 184 and 217 significant (P<0.05) trait-marker associations were identified using MLM in 2008 and 2009, respectively. Among them, 76 trait-marker associations were significant (P<0.05) both in 2008 and 2009. The number of significant loci associated with each agronomic trait in two years ranged from 0 (seed set rate) to 13 (plant height). Moreover, 24 (∼32%) of the 76 trait-marker associations were in the same or similar genomic regions where QTLs were detected in previous studies (http://www.gramene.org/), and the other 52 trait-marker associations were new associations which were not previously identified.
Eleven of the 76 trait-marker associations had 10% or more explained percentage of the total variation (R2), i.e. HD (PSM184), PH (RM530, RM590), PL (PSM184), GL/GW (RM447), FLL (RM287), FLW (RM235), 1000GW (RM7, RM538 and RM206), and PN (RM311) both in 2008 and 2009 (Table 4). When using BF and the Bonferroni threshold as significance thresholds, there were 15 and 3 trait-marker associations out of the 76 significant associations which still showed significant associations, respectively. Moreover, the three trait-marker significant associations shown by Bonferroni threshold were also significant when using BF as significant threshold. Furthermore, 59 of the 76 trait-marker associations were found to be significant when using the GLM model in two years.
Impact of allele frequency on the power to detect a QTL
We further investigated the relationship between the P values of significant trait-marker associations and the PIC values of related markers. For all trait-marker associations, only 3.5% of markers had a PIC value lower than 0.2 (Figure 4). Most of the markers which showed significant associations with related traits had a PIC value larger than 0.2, which meant that these markers showed a higher power to detect a QTL.
Green asterisk refers to the total markers used in traits in 2008. A red asterisk refers to the markers significantly associated with traits in 2008. A purple asterisk refers to the total markers used in traits in 2009. A green triangle refers to the markers significantly associated with traits in 2009.
Verification of association mapping results in Panel 2 and Panel 3
For the 76 significant trait-marker associations in Panel 1, because some SSR markers show more than one significant associations with related traits, the number of related SSR markers is less than 76, i.e. 55 SSR markers in this study. All these 55 SSR markers were further used to genotype Panel 2 and 3. Based on these genotyping data, the population structure of both Panel 2 and 3 indicated two distinct subgroups (Figure S8 in File S1).
Association analysis was performed within the two Panels using both MLM and GLM approaches with the 55 SSR markers. A total of 20 and 31 significant trait-marker associations were detected using MLM within Panel 2 and Panel 3, respectively. Seven significant trait-marker associations which were detected in Panel 1 using MLM model were identical with those in Panel 2 and Panel 3 using the GLM model, respectively. However, there was no identical trait-marker association within the three Panels when using the MLM model (Table 5). In Panel 2, RM219 [47], RM469 [48] and RM204 [49] showed significant associations with plant height and they were also reported by previous researches. Among them, the association for marker RM469 with plant height had the highest R2 (10.08%). Similarly, in Panel 3, the association for marker RM590 with plant height had the highest R2 (39.96%). RM339 which showed significant associations with heading days, were reported by previous researches [50] (Table 6).
Performance of traits relevant to different alleles of significant loci
Seven markers, i.e. PSM184, RM447, RM469, RM235, RM206, RM311, and RM277, were selected for analysis of trait performance relevant to different alleles of significant loci based on their high explained percentage of genetic variation and supported by several significant thresholds (Table 4). For PSM184, the individuals carrying the allele 222 bp (the size of PCR product for the SSR markers, the same as below) had a significantly (P<0.01) lower plant height and panicle length than those carrying other two alleles 205 bp and 215 bp (Table 7). For RM447, the individuals carrying the allele 109 bp had a significantly (P<0.01) higher grain width and significantly (P<0.01) lower grain length/width ratio than those carrying other two alleles 100 bp and 117 bp. For RM469, the individuals carrying the allele 94 bp had a significantly (P<0.01) lower flag leaf length than those carrying other two other alleles 83 bp and 88 bp. For RM206, the individuals carrying the allele 162 bp had a significantly (P<0.01) higher 1000-grain weight than those carrying the other four alleles 123 bp, 125 bp, 130 bp and 143 bp. For RM311, the individuals carrying the allele 143 bp, 143 bp and 153 bp showed a significantly (P<0.05) higher panicle number per plant than those carrying other two alleles 147 bp and 157 bp. For RM235, the individuals carrying the allele 108 bp showed a significantly (P<0.05) higher flag leaf width than those carrying the alleles 115 bp, 117 bp, 121 bp and 123 bp, whereas the individuals carrying the allele 123 bp had a significantly (P<0.05) lower flag leaf width than those carrying the alleles 91 bp, 108 bp, and 115 bp. For RM277, the individuals carrying the allele 117 bp had a higher grain length than those carrying the allele 111 bp (Duncan multiple comparisons was not been performed due to it had only two alleles).
Discussion
Comparison of different mapping populations for association mapping
An appropriate population with maximized phenotypic variation is critical for the success of an association analysis [18], [51]. Rice landraces represent an intermediate stage in domestication between wild and elite cultivars [2], which possess high genetic diversity and many exotic genes, and therewith provide useful germplasm resources for rice breeding. Moreover, association mapping based on a core collection of rice landraces would help to catch as much phenotypic variation as possible.
China is well known as one of the origin center of cultivated rice with abundant genetic resources for rice. As early as in 1920–1964s, Professor Ying Ting collected more than 7128 accessions of rice landraces from all over China as well as some countries which grow rice as a major crop. The collection is one of the earliest collections for rice germplasm resources and therefore was named Ting's rice germplasm collection [30]. Our previous results based on the core collection from it indicated that (1) the percentage of SSR loci pairs in significant (P<0.05) LD was 46.8%; (2) LD decayed rapidly to the threshold, i.e. the 95% quantile of r2 between unlinked loci pairs, at 1.03 cM in the entire collection; and (3) there were many LD blocks. These previous results indicated that Panel 1 was an appropriate population for association mapping. Therefore, our association mapping was performed based on Panel 1.
The populations in previous studies for association analysis in rice included populations from the USDA core collection [14], [16], [24], landraces [16], [17], elite cultivars [16], and mini-core collection [23]. The mapping populations in the researches of Agrama et al. [14], [24], [52] and Li et al. [23] were subsets chosen randomly from the USDA core collection, which consisted of 92, 547 and 203 accessions, respectively. Moreover, the number of SSR markers was 123, 72 and 155, which was rather low for association mapping. In the study of Zhao et al. [11], 416 rice accessions including only two landraces were randomly selected and only 100 SSR markers were used.
Our results indicated that there is a wide-range of phenotypic variation for 12 agronomic traits in Panel 1. For heading days, flag leaf length, flag leaf width, grain length, grain width, grain length/width and panicle length, there was less phenotypic variations than described in the research of Jin et al. [16], while for plant height and 1000 grains weight, more phenotypic variation was found than reported in the research of Jin et al. [16]. The comparison with the results of Li et al. [23] indicated that less phenotypic variation was found in this study for heading days, 1000-grain weight and panicle length, while more was found for plant height, panicle number per plant and seed set rate. More phenotypic variation was found than reported in the research of Agrama et al. [14] for grain length, grain width and 1000-grain weight.
Choice of statistical models and statistical parameters to control type I error
There are two frequently used models (i.e. MLM and GLM) which were implemented in the software TASSEL for association analysis [17], [23], [28]. In this study, we used the MLM (Q+K) [25] which accounted for population structure and kinship relationship to minimize spurious associations. For comparison, GLM was also used. In our study, 137 (∼90%) trait-marker associations were possibly new loci when using GLM model, whereas 52 (∼68%) trait-marker associations were possibly new loci when using MLM model. The ratio of possibly new significant loci detected using GLM model was much higher than that using MLM model. However, the new significant loci might be false positive because GLM model did not account for kinship.
Furthermore, the significance threshold (P value) must be set considerately in the association mapping. Using a smaller P value as threshold might lose more minor QTLs, while using a higher P value as threshold might get more false positive QTLs. To reliably interpret the MLM-derived significant associations in our study, we also used minimum BF estimation [44] for the MLM association results. Minimum BF estimates over P values of MLM approach may help to understand the overall impact of the associations [45]. We also used a Bonferroni threshold for identifying the associations derived from MLM analysis. The statistical parameters had been used successfully in association mapping of cotton [8]. Our results indicated that three significant trait-marker associations (i.e. plant height-RM530, grain length-RM156 and grain width-RM276) reached simultaneously the three thresholds (i.e. P<0.05, minimum BF, and the Bonferroni), which should be emphasized in future studies.
Moreover, molecular markers can be used to calculate the relative kinship between pairs of individuals in a study, which provides useful information for quantitative inheritance studies. Relative kinship reflects the approximate identity between two given individuals over an average probability of identity between two random individuals [25]. Our results indicated that most varieties had no or weak relationship with each other in the Ting's core collection, which might be due to the fact that these varieties were chosen from a diverse rice cultivating region including all over China, East Asia, and Southeast Asia. The quantile-quantile plot indicated that MLM (Q+K) performed well in association mapping on 12 agronomic traits, which could correct false positive trait-marker associations (Figure 2 and 3).
Association analysis within Ting's core collection
Using Ting's rice core collection genotyped with 274 SSR markers, we performed association mapping for 12 agronomics traits with two years data using the MLM and GLM models implemented in TASSEL. In this study, most (∼80%) of the significant associations found using the MLM approach were also supported by the GLM approach in both years. The percentage of associations identical to previous reported QTLs was about 32%, which was higher than those in the research of Li et al. [23], but lower than those in the research of Agrama et al. [14]. The 76 significant trait-marker associations which were detected in both years were potential markers for effective marker-assisted selection programs in rice. Moreover, 52 of the 76 significant associations which were not detected in previous studies might be some new potential loci. For instance, the trait-marker associations for heading days with PSM184, plant height with RM590, grain length/width with RM447, flag leaf length with RM287, flag leaf width with RM235, 1000-grain weight with RM538, and 1000-grain weight with RM206, explained more than 10% of genetic variations both in 2008 and 2009.
For heading days, two of the four significant trait-marker associations were identical to previous reported QTLs, i.e. RM341 and RM339, were identical to previous reported QTLs in the research of Mei et al. [48] and Kunihiro et al. [50], respectively. Moreover, RM339 was also significantly associated with heading days in Panel 2 and 3. For heading days, ten of 13 significant trait-marker associations were identical to previous reported QTLs, i.e. RM530 in the research of Mei et al. [53], RM138 in the research of Fang et al. [51], PSM130 in the research of Cao et al. [54], RM469 (which also showed significant association in Panel 2 and 3) and PSM184 in the research of Mei et al. [48], RM204 (which also showed significant association in Panel 2 and 3) and RM225 in the research of Yang et al. [49], RM219 (which also showed significant association in Panel 2 and 3) in the research of Xiao et al. [47], RM21 and RM147 in the research of Lanceras et al. [55]. For panicle length, the two significant trait-marker associations were also identical to previous reported QTLs, i.e. RM228 and PSM184 in the research of Mei et al. [53] and Jiang et al. [56], respectively. For grain length, three of ten significant trait-marker associations were identical to previous reported QTLs in the previous researches, i.e. RM127 in the research of Tan et al. [57], PSM158 in the research of Xing et al. [58], and PSM171 in the research of Yoshida et al. [59]. For grain length/width, two of nine significant trait-marker associations were identical to previous reported QTLs in the previous researches, i.e. RM276 and RM557 reported by Tan et al. [57]. For flag leaf width, one of nine significant trait-marker associations were identical to previous reported QTLs, i.e. RM571 in the research of Mei et al. [48]. For 1000-grain weight, there of eight significant trait-marker associations were identical to previous reported QTLs in the previous researches, i.e. RM7 in the research of Hittalmani et al. [60], RM239 in the research of Gao et al. [61], and RM206 in the research of Cho et al. (this reference cannot be found, but QTL ID can be found in GRAMENE website). For panicle number per plant, the only one significant trait-marker association was also identical to previous reported QTL, i.e. RM311 in the research of Kobayashi et al. [62].
Verification association mapping results within Panel 2 and Panel 3
It is worthwhile to further verify the significant associations identified within one population in a different population [29]. In this study, 55 SSR markers for the 76 trait-marker associations identified in Panel 1 were used to genotype two other populations, i.e. Panel 2 and Panel 3, and an association mapping was performed using both MLM and GLM approaches. When using the GLM approach, seven significant trait-marker associations were identical within Panel 1 and Panel 2 or Panel 3. Moreover, three of the seven identical significant trait-marker associations in the two panels were reported by previous studies. Although the GLM would bring more false positive results than the MLM when it was used alone, however, some significant trait-marker associations were first detected Panel 1 in our research and proved by several statistical thresholds as well as by previous mapping results. After that, we used the GLM to verify our mapping results in Panel 2 and 3. Therefore, it makes sense for verification of association mapping results by the fact that some common trait-marker associations were detected by the GLM approach.
We observed that there were no overlapping QTLs among the three panels with the GLM approach. The reasons might be (1) different compositions and origins of the varieties in three panels, where Panel 1 only consists of original rice landraces from China and some other rice growing countries which were collected during 1920–1964 before the emergence of hybrid rice, while Panel 2 consists of rice landraces as well as modern rice cultivars and maintainer lines in hybrid rice breeding from China, and Panel 3 is a worldwide collection and consists of modern rice cultivars including cytoplasmic sterile line, maintainer lines, and some landraces; (2) that different allelic frequencies might exist for the three panels which consist of different compositions and origins. The explanations were supported by our observations that (1) frequency of some alleles was different in the three panels and some alleles only exist in one panel (Table S3 in File S1), and (2) in our another experiment some alleles associated with aluminum tolerance were different for different germplasm types (data not shown).
When using the MLM approach, no identical significant trait-marker associations were found among the three panels. Previous studies on linkage mapping and association mapping also found that different mapping populations detected different QTL regions [14], [48], [63], [64], [65]. The reasons might be due to that (1) a much lower number of SSR markers (55 SSRs) was used in Panel 2 and Panel 3 than in Panel 1 (274 SSRs); (2) the 55 SSR markers are associated with relevant traits which were not randomly distributed across the genome, which might reduce the exactness of measurement for population structure and kinship; (3) the relative kinship calculated by 274 SSRs in Panel 1 was quite different than those calculated by the 55 SSRs in Panel 2 and 3, where in Panel 1 only 4.73% of pairwise kinship coefficient were larger than 0.5 and most of them were zero, whereas 55.9% and 60.4% of pairwise kinship coefficient in Panel 2 and 3 were larger than 0.5, respectively (Figure S8 in File S1); and (4) the degree of association might be reduced in MLM compared to those in GLM [50], which meant that when using much less SSR markers, the weak significant trait-marker associations in GLM might be not significant in MLM. As verification experiments were rarely performed in previous association studies, it is required to find an efficient solution for verification in future as well as to check the repeatability in different association mapping populations.
Prospects for association mapping based on core collections
Association mapping has become a promising approach to mine elite genes within germplasm populations compared to traditional linkage mapping. Association mapping based on a core collection would help to capture as much phenotypic variation as possible. Compared to a natural population or a breeding population with a broad genetic basis, the LD level in a core collection might be low due to its diverse origin. Therefore, more markers might be required for association mapping. However, due to the quick LD decay, fine mapping using association analysis might be possible with a core collection. As quick, automated, economic genotyping technologies (such as genotyping by sequencing) have been developed, genotyping large germplasm resources with high density markers and GWAS in such mapping populations has become possible. Because such an association could be further applied in rice breeding by molecular marker assisted selection, it would be promising to make use of the elite genes in the diverse germplasm resources by the current strategy.
Supporting Information
File S1.
Table S1, Accessions, variety names, origin, germplasm types of 150 rice varieties in Panel 1. Table S2, Summary statistics of the 274 SSR markers used in this study. Table S3, Allele frequency of the 55 significant markers in three panels. Figure S1, Frequency distribution of heading days, plant height, seed set rate and panicle length in Panel 1 in 2008. The height of black bar represents the number of varieties in different range of traits. Figure S2, Frequency distribution of grain length, grain width, grain length/width and 1000 grain weight in Panel 1 in 2008. The height of black bar represents the number of varieties in different range of traits. Figure S3, Frequency distribution of flag leaf length, flag leaf width, flag leaf length/width and panicle number per plant in Panel 1 in 2008. The height of black bar represents the number of varieties in different range of traits. Figure S4, Frequency distribution of heading days, plant height, seed set rate and panicle length in Panel 1 in 2009. The height of black bar represents the number of varieties in different range of traits. Figure S5, Frequency distribution of grain length, grain width, grain length/width and 1000 grain weight in Panel 1 in 2009. The height of black bar represents the number of varieties in different range of traits. Figure S6, Frequency distribution of flag leaf length, flag leaf width, flag leaf length/width and panicle number per plant in Panel 1 in 2009. The height of black bar represents the number of varieties in different range of traits. Figure S7, Distribution of pairwise relative kinship values in Panel 2 and 3. The height of black bar represents the percentage of varieties in different range of kinships. Figure S8, Delta K change according to different K among Panel 2 and Panel 3 identified by STRUCTURE under Admixture model.
https://doi.org/10.1371/journal.pone.0111508.s001
(DOC)
Acknowledgments
We are grateful to Dr. Guoyou Ye from International Rice Research Institute, Dr. Xiaoling Li, Dr. Lan Wang, Dr. Zhixiong Chen, Dr. Xuelin Fu, Dr. Youxin Yang, Ms Xingjuan Zhao and Ms. Shuhong Yu from South China Agricultural University for their assistance in the experiment, and thank Miss Anja Bus from Max Planck institute for plant breeding research for the improvement of English writing.
Author Contributions
Conceived and designed the experiments: JL PZ. Performed the experiments: PZ. Analyzed the data: PZ JL. Contributed reagents/materials/analysis tools: JL XL HT. Wrote the paper: PZ JL YL.
References
- 1. Mather KA, Caicedo AL, Polato NR, Olsen KM, McCouch S, et al. (2007) The extent of linkage disequilibrium in rice (Oryza sativa L.). Genetics 177: 2223–2232.
- 2. Londo JP, Chiang YC, Hung KH, Chiang TY, Schaal BA (2006) Phylogeography of Asian wild rice, Oryza rufipogon, reveals multiple independent domestications of cultivated rice, Oryza sativa. Proc Natl Acad Sci U S A 103: 9578–9583.
- 3. Flint-Garcia SA, Thornsberry JM, Buckler ES (2003) Structure of linkage disequilibrium in plants. Annu Rev Plant Biol 54: 357–374.
- 4. Kraakman ATW, Niks RE, Van den Berg PMMM, Stam P, Van Eeuwijk FA (2004) Linkage disequilibrium mapping of yield and yield stability in modern spring barley cultivars. Genetics 168: 435–446.
- 5. Zhu C, Gore M, Buckler ES, Yu J (2008) Status and Prospects of association mapping in plants. The Plant Genome 1: 5–20.
- 6. Remington DL, Thornsberry JM, Matsuoka Y, Wilson LM, Whitt SR, et al. (2001) Structure of linkage disequilibrium and phenotypic associations in the maize genome. Proc Natl Acad Sci U S A 98: 11479–11484.
- 7. Huang X, Han B (2014) Natural variations and genome-wide association studies in crop plants. Annu Rev Plant Biol 65: 531–551.
- 8. Abdurakhmonov IY, Abdukarimov A (2008) Application of association mapping to understanding the genetic diversity of plant germplasm resources. Int J Plant Genomics 2008: 574927.
- 9. Han B, Huang X (2013) Sequencing-based genome-wide association study in rice. Curr Opin Plant Biol 16: 133–138.
- 10. Huang X, Zhao Y, Wei X, Li C, Wang A, et al. (2012) Genome-wide association study of flowering time and grain yield traits in a worldwide collection of rice germplasm. Nat Genet 44: 32–39.
- 11. Zhao K, Tung CW, Eizenga GC, Wright MH, Ali ML, et al. (2011) Genome-wide association mapping reveals a rich genetic architecture of complex traits in Oryza sativa. Nat Commun 2: 467.
- 12. Famoso AN, Zhao K, Clark RT, Tung CW, Wright MH, et al. (2011) Genetic architecture of aluminum tolerance in rice (Oryza sativa) determined through genome-wide association analysis and QTL mapping. Plos Genetics 7.
- 13. Zhang N, Xu Y, Akash M, McCouch S, Oard JH (2005) Identification of candidate markers associated with agronomic traits in rice using discriminant analysis. Theoretical and Applied Genetics 110: 721–729.
- 14. Agrama HA, Eizenga GC, Yan W (2007) Association mapping of yield and its components in rice cultivars. Molecular Breeding 19: 341–356.
- 15. Iwata H, Ebana K, Uga Y, Hayashi T, Jannink JL (2010) Genome-wide association study of grain shape variation among Oryza sativa L. germplasms based on elliptic Fourier analysis. Molecular Breeding 25: 203–215.
- 16. Jin L, Lu Y, Xiao P, Sun M, Corke H, et al. (2010) Genetic diversity and population structure of a diverse set of rice germplasm for association mapping. Theoretical and Applied Genetics 121: 475–487.
- 17. Huang X, Wei X, Sang T, Zhao Q, Feng Q, et al. (2010) Genome-wide association studies of 14 agronomic traits in rice landraces. Nat Genet 42: 961–967.
- 18. Yan JB, Warburton M, Crouch J (2011) Association Mapping for Enhancing Maize (Zea mays L.) Genetic Improvement. Crop Science 51: 433–449.
- 19.
Frankel OH (1984) Genetic perspectives of germplasm conservation. In: Arber W, Llimensee K, Peacock WJ (Eds.), Genetic Manipulation: Impact on Man and Society. Cambridge University Press, UK, pp.161–170.
- 20.
Frankel OH, Brown AHD (1984a) Current plant genetic resources—a critical appraisal. In: Genetics: new frontiers, vol 4. Oxford and IBH Publ, New Delhi, India, pp. 1–11.
- 21.
Frankel OH, Brown AHD (1984b) Plant genetic resources today: acritical appraisal. In: Hoden HW, Williams JT (eds) Crop genetic resources: conservation and evaluation. George Allen and Urwin, London, pp. 249–257.
- 22.
Brown AHD (1995) The core collection at the crossroads. In: Hodgkin T, Brown AHD, van Hintum TJL, Morales EAV (Eds.), Core Collections of Plant Genetic Resources. John Wiley and Sons, Chichester, UK, pp. 3–19.
- 23. Li XB, Yan WG, Agrama H, Jia LM, Shen XH, et al. (2011) Mapping QTLs for improving grain yield using the USDA rice mini-core collection. Planta 234: 347–361.
- 24. Agrama HA, Yan W (2009) Association mapping of straighthead disorder induced by arsenic in Oryza sativa. Plant Breeding 128: 551–558.
- 25. Yu JM, Pressoir G, Briggs WH, Bi IV, Yamasaki M, et al. (2006) A unified mixed-model method for association mapping that accounts for multiple levels of relatedness. Nature Genetics 38: 203–208.
- 26. Zhu CS, Yu JM (2009) Nonmetric multidimensional scaling corrects for population structure in association mapping with different sample types. Genetics 182: 875–888.
- 27. Wang ML, Zhu CS, Barkley NA, Chen ZB, Erpelding JE, et al. (2009) Genetic diversity and population structure analysis of accessions in the US historic sweet sorghum collection. Theoretical and Applied Genetics 120: 13–23.
- 28. Yang XH, Yan JB, Shah T, Warburton ML, Li Q, et al. (2010) Genetic analysis and characterization of a new maize association mapping panel for quantitative trait loci dissection. Theoretical and Applied Genetics 121: 417–431.
- 29. Wray NR, Yang J, Hayes BJ, Price AL, Goddard ME, et al. (2013) Pitfalls of predicting complex traits from SNPs. Nat Rev Genet 14: 507–515.
- 30. Li XL, Lu YG, Li JQ, Xu HM, Muhammad QS (2011) Strategies on sample size determination and qualitative and quantitative traits integration to construct core collection of rice (Oryza sativa). Rice Science 18: 46–55.
- 31. Zhang P, Li JQ, Li XL, Liu XD, Zhao XJ, et al. (2011) Population structure and genetic diversity in a rice core collection (Oryza sativa L.) investigated with SSR markers. PLoS One 6 (12)
- 32.
Li JQ, Zhang P (2012) Genetic diversity in plants. In: Çalişkan M, editor. Chapter5: assessment and utilization of the genetic diversity in rice. Hard cover: InTech-Open Access Publisher.
- 33. Chen X, Temnykh S, Xu Y, Cho YG, McCouch SR (1997) Development of a microsatellite framework map providing genome-wide coverage in rice, Oryza sativa L. Theoretical and Applied Genetics 553–567.
- 34. Temnykh S, Park WD, Ayes N, Cartinhour S, Hauck N, et al. (2000) Mapping and genome organization of microsatellite sequences in rice, Oryza sativa L. Theoretical and Applied Genetics 697–712.
- 35. Temnykh S, Declerck G, Lukashova A, Lipovich L, Cartinhour S, et al. (2001) Computational and experimental analysis of microsatellites in rice (Oryza sativa L.): frenquency, length variation, transposon associations, and genetic marker potential. Genetics Research 1441–1452.
- 36. McCouch SR, Teytelman L, Xu Y, Lobos KB, Clare K, et al. (2002) Development and mapping of 2240 new SSR markers for rice (Oryza sativa L.). DNA Res 9: 199–207.
- 37.
Huang CF (2003) Development of position-specific microsatellite markers and molecular mapping lf insect resistant genes in rice (Oryza sativa L.). M.Sc. Thesis, South China Agricultural University.
- 38. Zheng KL, Huang N, Bennett J, Khush GS (1995) PCR-based phylogenetic analysis of wide compatibility varieties in Oryza sativa L. Theoretical and Applied Genetics 65–69.
- 39. Panaud O, Chen X, McCouch SR (1996) Development of microsatellite markers and characterization of simple sequence length polymorphism (SSLP) in rice (Oryza sativa L.). Mol Gen Genet 252: 597–607.
- 40. Pritchard JK, Stephens M, Donnelly P (2000a) Inference of population structure using multilocus genotype data. Genetics 155: 945–959.
- 41. Pritchard JK, Stephens M, Rosenberg NA, Donnelly P (2000b) Association mapping in structured populations. Am J Hum Genet 67: 170–181.
- 42. Hardy OJ, Vekemans X (2002) SPAGeDi: a versatile computer program to analyse spatial genetic structure at the individual or population levels. Molecular Ecology Notes 2: 618–620.
- 43. Yu J, Buckler ES (2006) Genetic association mapping and genome organization of maize. Curr Opin Biotechnol 17: 155–160.
- 44. Goodman SN (2001) Of P-values and Bayes: a modest proposal. Epidemiology 12: 295–297.
- 45. Katki HA (2008) Invited commentary: Evidence-based evaluation of p values and Bayes factors. American Journal of Epidemiology 168: 384–388.
- 46. Moran MD (2003) Arguments for rejecting the sequential Bonferroni in ecological studies. Oikos 100: 403–405.
- 47. Xiao J, Li J, Grandillo S, Ahn SN, Yuan L, et al. (1998) Identification of trait-improving quantitative trait loci alleles from a wild rice relative, Oryza rufipogon. Genetics 150: 899–909.
- 48. Mei HW, Li ZK, Shu QY, Guo LB, Wang YP, et al. (2005) Gene actions of QTLs affecting several agronomic traits resolved in a recombinant inbred rice population and two backcross populations. Theoretical and Applied Genetics 110: 649–659.
- 49. Yang GH, Xing YZ, Li SQ, Ding JZ, Yue B, et al. (2006) Molecular dissection of developmental behavior of tiller number and plant height and their relationship in rice (Oryza sativa L.). Hereditas 143: 236–245.
- 50. Kunihiro Y, Qian Q, Sato H, Teng S, Zeng DL, et al. (2002) QTL analysis of sheath blight resistance in rice (Oryza sativa L.). Yi Chuan Xue Bao 29: 50–55.
- 51. Flint-Garcia SA, Thuillet AC, Yu JM, Pressoir G, Romero SM, et al. (2005) Maize association population: a high-resolution platform for quantitative trait locus dissection. Plant J 44: 1054–1064.
- 52. Agrama HA, Eizenga GC (2008) Molecular diversity and genome-wide linkage disequilibrium patterns in a worldwide collection of Oryza sativa and its wild relatives. Euphytica 160: 339–355.
- 53. Mei HW, Luo LJ, Ying CS, Wang YP, Yu XQ, et al. (2003) Gene actions of QTLs affecting several agronomic traits resolved in a recombinant inbred rice population and two testcross populations. Theoretical and Applied Genetics 107: 89–101.
- 54. Cao G, Zhu J, He C, Gao Y, Yan J, et al. (2001) Impact of epistasis and QTL×environment interaction on the developmental behavior of plant height in rice (Oryza sativa L.). Theoretical and Applied Genetics 103: 153–160.
- 55. Lanceras JC, Pantuwan G, Jongdee B, Toojinda T (2004) Quantitative trait loci associated with drought tolerance at reproductive stage in rice. Plant Physiology 135: 384–399.
- 56. Jiang GH, Xu CG, Li XH, He YQ (2004) Characterization of the genetic basis for yield and its component traits of rice revealed by doubled haploid population. Yi Chuan Xue Bao 31: 63–72.
- 57. Tan YF, Xing YZ, Li JX, Yu SB, Xu CG, et al. (2000) Genetic bases of appearance quality of rice grains in Shanyou 63, an elite rice hybrid. Theoretical and Applied Genetics 101: 823–829.
- 58. Xing YZ, Tan YF, Xu CG, Hua JP, Sun XL (2001) Mapping quantitative trait loci for grain appearance traits of rice using a recombinant inbred line population. Acta Botanica Sinica 43: 840–845.
- 59. Yoshida S, Ikegami M, Kuze J, Sawada K, Hashimoto Z, et al. (2002) QTL analysis for plant and grain characters of sake-brewing rice using a doubled haploid population. Breeding Science 52: 309–317.
- 60. Hittalmani S, Huang N, Courtois B, Venuprasad R, Shashidhar HE, et al. (2003) Identification of QTL for growth- and grain yield-related traits in rice across nine locations of Asia. Theoretical and Applied Genetics 107: 679–690.
- 61. Gao Y, Zhu J, Song Y, He C, Shi C, et al. (2004) Analysis of digenic epistatic effects and QE interaction effects QTL controlling grain weight in rice. Journal of Zhejiang University Science 5: 371–377.
- 62. Kobayashi S, Fukuta Y, Sato T, Osaki M, Khush GS (2003) Molecular marker dissection of rice (Oryza sativa L.) plant architecture under temperate and tropical climates. Theoretical and Applied Genetics 107: 1350–1356.
- 63. Huang N, Courtois B, Khush GS, Lin HX, Wang GL, et al. (1996) Association of quantitative trait loci for plant height with major dwarfing genes in rice. Heredity 77: 130–137.
- 64. Septiningsih EM, Prasetiyono J, Lubis E, Tai TH, Tjubaryat T, et al. (2003) Identification of quantitative trait loci for yield and yield components in an advanced backcross population derived from the Oryza sativa variety IR64 and the wild relative O-rufipogon. Theoretical and Applied Genetics 107: 1419–1432.
- 65. Thomson MJ, Tai TH, McClung AM, Lai XH, Hinga ME, et al. (2003) Mapping quantitative trait loci for yield, yield components and morphological traits in an advanced backcross population between Oryza rufipogon and the Oryza sativa cultivar Jefferson. Theoretical and Applied Genetics 107: 479–493.