Introduction

Maize is an important cereal crop worldwide, playing a significant role in human and livestock nutrition (Blummel et al. 2013). Physical traits of maize kernel, including kernel size and kernel weight, directly affect grain yield and food quality (Raihan et al. 2016). When kernels are used for seeds, their physical traits show correlations to seed vigor. Studies on maize and cotton showed that the seed mass and oil content were positively associated with seed germination performances and seedling vigor (Revilla et al. 1999; Snider et al. 2016). Moreover, studies from pea seeds provided the opposite result, which showed that small seeds have better seedling vigor (Peksen et al. 2004; Ambika et al. 2014).

Kernel weight is usually indicated by hundred-kernel weight (HKW), while kernel size can be reflected by kernel length (KL), kernel width (KW) and kernel thickness (KT). Recently, genetic studies have emphasized kernel size and weight in crops, which revealed that both were quantitative traits regulated by many genes (Huang et al. 2012; Li and Li 2016; Chen et al. 2016a; Liu et al. 2017b; Wu et al. 2017). Several grain size- and weight-related genes have been cloned from rice, which could be clustered into four functional groups: (1) RING-type E3 ubiquitin ligase genes, including GW2 and GW5/qSW5 (Song et al. 2007; Li et al. 2010a; Xu et al. 2015); (2) hormone signaling regulator genes, including GL2, GSK2, and TGW6 (Ishimaru et al. 2013; Che et al. 2015; Liu et al. 2017a); (3) transcription factor genes, including the SBP-domain transcription factor GW8/OsSPL16 (Wang et al. 2012; Wang et al. 2015a) and MADS-box transcription factor OsLG3b (Yu et al. 2018); and (4) other genes encoding the TONNEAU1-recruiting motif protein GW7/GL7 (Wang et al. 2015b), plant-specific organ size regulation domain protein GS3 (Fan et al. 2006; Li et al. 2010b; Gao et al. 2015), and putative serine carboxypeptidase-like protein OsGS5 (Li et al. 2011; Xu et al. 2015).

Maize seeds exhibit substantial diversity in size and weight within and among populations, which enables the identification of a substantial number of trait-related QTL (Chen et al. 2016b, 2016c; Peng et al. 2011; Li et al. 2013, Li et al. 2016; Qin et al. 2016). Based on these identified QTL, a meta-analysis, combined with a genome-wide association study (GWAS), identified several key genes showing potential associations with these traits (Chen et al. 2017; Liu et al. 2017a). However, the underlying gene/QTL has been seldomly cloned, probably due to the complex genome structure of maize (Liu et al. 2015).

Epistasis is a phenomenon where the effect of a gene/locus is dependent on the presence of other gene(s) or locus (loci). Epistasis arises due to the interactions among genes, which leads to nonlinear effects and substantially contributes to the variation of quantitative traits (Mather and Jinks 1982; Stich and Gebhardt 2010). Epistasis analysis has recently been applied in maize for some important traits, such as plant height, but the application in seed trait has not been published (Ma et al. 2018; Wu et al. 2018).

In this work, a set of QTL for seed size and weight were identified from an F2:3 population. After comparing to the previously identified QTL and trait-associated SNPs, several promising chromosome regions were highlighted for further fine mapping of the underlying genes and the acceleration of the breeding of large-seed maize. In addition, a number of potential epistatic interactions were identified to understand the possible genetic epistatic effects on these traits.

Materials and methods

Plant materials

An F2:3 population previously generated from 220 (herein named L220) × PH4CV was used for this study (Li et al. 2018b). The seeds of 223 F2 lines were sowed in Shunyi, Beijing, China (116°65′E longitude, 40°13′N latitude) in 2014. DNA from F2 plants were used for genotyping (Li et al. 2018b), and the harvested F3 seeds were used for trait evaluation.

Phenotype evaluation

Ninety F3 seeds were randomly sampled from each F2 plant and divided into three groups (replicates). Each group of 30 seeds were scanned using an EPSON J221A scanner (Seiko Epson Corporation, Nagano-ken, Japan), followed by an image analysis to score the data for kernel length (KL), kernel width (KW), kernel thickness (KT), and projected area (PA) by using the Seeds Identification and Photoshop software package (Wen et al. 2015). The kernel volume (KV) was calculated as the formula: KV (mm3) = KT × PA. The hundred-kernel weight (HKW) was the dry mass of 100 seeds that was measured after drying the seeds in a 130 °C oven for 1 h (ISTA 2018). Total seed starch was measured for 30 individual seeds by drying the seeds in a freeze dryer (Labconco FreeZone, Kansas City, MO) and grinding the seeds into a powder, which was then used to perform the starch measurement according to the user manual of the Total Starch Assay Kit (Megazyme, Wicklow, Ireland) (McCleary et al. 1994).

Statistical analysis and QTL identification

The collected phenotypic data were analyzed with IBM SPSS 20.0 (IBM corp., Armonk, NY) and the R statistical package (R Core Team 2016). A mean value of three replicates was used for correlation analysis and QTL mapping. The coefficients of variation (CV, %) were calculated as follows: CV = s / x, (s, the standard deviation; x, the mean for each trait). Linear regression analysis was used to determine the phenotypic correlation coefficients (Pearson’s) at the significance level of P = 0.05.

A genetic linkage map was constructed for this F2:3 population with SNP markers generated from a 6 K Array Chip, with a total length of 1689.8 cM and an average interval length of 1.3 cM (Li et al. 2018b). QTL mapping was conducted with the software QTL IciMapping 4.1 by using the inclusive composite interval mapping (ICIM) algorithm (Li et al. 2007). The cut-off threshold likelihood of odds (LOD) value at a P = 0.05 was determined separately for each trait using the output from 1000 permutations of the original trait data (Churchill and Doerge 1994).

Meta-QTL analysis

Based on the physical localizations of the genetic markers on the reference genome (B73 RefGen_v2) in each population, a meta-QTL analysis was performed by syntenic comparison of the identified QTL intervals from this work with the intervals for the published grain yield, kernel size, and kernel weight traits.

Analysis of epistatic interactions

The possibility of epistatic interactions between each pair of SNP markers was calculated using a two-way ANOVA test in R package (Li et al. 2007; Supplementary Data 1). A permutation test was conducted by randomly sampling 1000 pairs of markers with 1000 times to determine the genome-wide significant P value threshold. After recording the minimum p value observed for each set of permutations, the genome-wide threshold was set as the 5th percentile of all observed minimum values. The interactions with p values less than the threshold value for each trait were taken as the epistatic interactions, and the involved adjacent SNP markers on both sides of the interaction were merged into a chromosome region (locus).

Results

Seeds physical traits in the parents and the F2:3 population

PH4CV showed significantly larger seeds than L220, which was reflected by both seed size and seed weight (Fig. 1). The HKW and KV of PH4CV were approximately doubled compared to those of L220, with 28.39 g and 405.95 mm3 in PH4CV and 11.47 g and 244.37 mm3 in L220, respectively. The KL, KW, KT, and PA in PH4CV were 1.2–1.4 times higher than those of L220. Interestingly, the embryo size between PH4CV and L220 was similar, but their endosperm showed a large difference. The PH4CV endosperm was approximately 3 times higher than the L220 endosperm (Fig. 1c, e). As a consequence of the large endosperm, the starch content in PH4CV was significantly higher than that in L220 (Fig. 1d).

Fig. 1
figure 1

Seed physical-related traits in parental lines L220 and PH4CV. a Pictures of 50 seeds. b Seed transverse and longitudinal sections. c Pictures of 5 embryos. d Total starch content. e Weight of 30 seeds. f–k Kernel length (KL), kernel width (KW), kernel thickness (KT), project area (PA), hundred-kernel weight (HKW), and kernel volume (KV). ** and * represent significant differences at P < 0.01 and P < 0.05, respectively

Full size image

Among the population lines, all six investigated traits were continuously and normally distributed and showed quantitative inheritance, with absolute vales of both skewness and kurtosis being less than 1.0, with the exception of kurtosis for KW (Table 1). Low CV values were observed in all six traits, with the highest level for HKW (13.81%) and the lowest level for KW (6.84%). Notably, the phenotypic data of all six traits exhibited obvious bidirectional transgressive segregation in this population, indicating the polygenic control of the seed physical traits (Table 1; Fig. S1).

Table 1 Seed physical trait performances in population L220 × PH4CV
Full size table

QTL mapping of seed physical traits in the F2:3 population

A total of 9 confidential QTL were identified with the explained phenotypic variance of 6.71–13.77% and the LOD values of 3.61–6.77 (Table S1; Fig. 2). In addition to the confidential QTL, 11 QTL were identified with LOD values lower than the threshold levels but above 2.5, explaining phenotypic variances of 3.97–7.99% (Table 2, S1; Fig. 2). These 20 QTL corresponded to the six seed physical traits, with 5 for KV, 4 for KT, 3 for KL, KW and PA, and 2 for HKW. The highest explained phenotypic variance was found for a KW QTL (qKW4-1), and the highest LOD value was found for a KV QTL (qKV1-1). Nine QTL (45%) inherited the contributable alleles from the large-seed parent PH4CV, while 11 (55%) inherited the contributable alleles from the small-seed parent L220 (Table S1).

Fig. 2
figure 2

QTL distribution and epistatically interacting loci for seed physical traits detected from the L220 × PH4CV F2:3 population. Different colored lines within the chromosome cycle represent the chromosome positions and LOD values for QTL detected for kernel length (KL), kernel width (KW), kernel thickness (KT), hundred-kernel weight (HKW), seed projected area (PA), and kernel volume (KV). Detailed QTL information is listed in Table S2. The detected epistatic interactions that contributed to the above 6 traits were labeled with corresponding colored belt curves. Width of the belt represented the width of interacted QTL region

Full size image
Table 2 Summary of the chromosome regions harboring multiple QTL for the 6 seed physical traits identified in the L220 × PH4CV population
Full size table

Among the 20 identified QTL, 15 were colocated on 6 chromosome regions, with 3 on Chr. 2: 7.6–10.6 Mb, Chr. 4: 188.3–193.2 Mb and Chr. 9: 13.9–18.3 Mb, and 2 on Chr. 1: 195.2–195.8 Mb, Chr. 3: 8.88–9.58 Mb and Chr. 8: 18.77–24.11 Mb (Table 2; Fig. 3). It was interesting to find that QTL colocated at the same regions inherited the favorable alleles from the same parents, with the exception of the QTL at Chr. 8: 18.77–24.11 Mb (Table 2, S1). The QTL on Chr. 1: 195.2–195.8 Mb, Chr. 3: 8.88–9.58 Mb, and Chr. 4: 188.3–193.2 Mb inherited the favorable alleles from the small-seed parent L220 that contributed 3.97–9.90% phenotypic variances, while those on Chr. 3: 8.88–9.58 Mb and Chr. 9: 13.9–18.3 Mb inherited the favorable alleles from the large-seed parent PH4CV (Table 2).

Fig. 3
figure 3

Meta-QTL (mQTL) detected that relate to seed physical traits from the L220 × PH4CV population and previously published populations. Different symbols represent different traits of kernel length (KL), kernel width (KW), kernel thickness (KT), seed projected area (PA), kernel volume (KV) and hundred-kernel weight (HKW), and/or their mQTL. Different colors stand for different studies with different populations. The chromosomal localizations of the 8 mQTL are shadowed with a box

Full size image

Comparative meta-QTL analysis and gene prediction within promising QTL regions

From 27 biparental populations published in the past decade, 1321 QTL were collected, with 1228 from F2:3 populations and 93 from recombination inbred line (RIL) populations (Lima et al. 2006; Peng et al. 2011; Semagn et al. 2013; Liu et al. 2014; Jiang et al. 2015; Chen et al. 2016a; Raihan et al. 2016; Liu et al. 2017b; Zhang et al. 2017). Together with the 20 QTL detected in this work, a meta-QTL analysis was conducted to identify 8 chromosome regions (8 mQTL) harboring 5–29 initial QTL in each region (Table S2; Fig. 3). It was interesting to find that 19 out of the identified 20 QTL from this work were located in the 8 mQTL regions (Table S2).

mQTL9-1 contained 12 QTL with 3 (qHKW9-1, qKV9-1, qPA9-1) from this work and 9 from previous publications for both seed size and weight traits. The contributable alleles for the QTL from this study were inherited from the large-kernel parent PH4CV, explaining phenotypic variance of 7.37–10.86% (Table S2). Most of the previously identified QTL in mQTL9-1 were also contributed by the large-kernel parents (Zhang et al. 2017; Liu et al. 2017b; Semagn et al. 2013), suggesting an important role of this region in the regulation of seed size and weight contributed by the larger-kernel parents.

mQTL2-1 was a seed size-specific mQTL that contained 3 initial QTL (qKL2-1, qKT2-1, and qPA2-1) from this work and 7 from previous works (Table S2). The contributable alleles were inherited from both the small-kernel parents and large-kernel parents (Zhang et al. 2017; Liu et al. 2017b; Semagn et al. 2013).

mQTL1-1 harbored 4 seed size QTL (qKT1-1, qKV1-1, qKV1-2, and qKT1-2) from this population and 25 size and weight QTL from the previous populations (Table S2). mQTL4-1 contained 3 QTL (qKV4-1, qPA4-1, and qHKW4-1) from this work and 12 (9 for KW, 2 for HKW, and 1 for KV) from previous works. The other 6 mQTL contained 5–8 QTL identified from both this work and previous works, controlling both seed size and weight traits. The contributable alleles for QTL in these 6 mQTL regions were inherited from both the small-kernel parents and large-kernel parents.

Detection of digenic epistatic effects on seed size and weight traits

A two-way ANOVA test identified 10 epistatic interactions that had significant effects on the 6 seed physical traits (P < 0.01), with 3 for HKW, 2 for both KW and PA, and one each for KL, KT, and KV (Table S3; Fig. 2). All interactions were transinteractions with the two loci located in different chromosomes (Table S3; Fig. S2). One interaction between locus M5c33968310 on Chr. 5 and locus M6c145990751 on Chr. 6 affected both the KL and PA traits, while the remaining 9 interactions affected only one seed physical trait. Among the 10 interactions, 6 showed an additive × additive epistatic effect (AA), 3 showed dominant × additive epistatic (DA) effects in addition to an AA effect, and 1 showed an additive × dominant effect (AD) in addition to an AA effect (Table S3).

Interestingly, 6 out of the 20 identified QTL were located in the interacting loci. Locus Chr. 9: 173.16 harbored three QTL (qKV9-1, qPA9-1, and qHKW9-1) and interacted with the M3c2743270 locus to contribute the HKW trait. Locus Chr. 2: 176.02 and Chr. 7: 168.29 harbored two (qKW3-1 and qKV3-1) QTL and one (qHKW4) QTL and interacted with the M2c176020325 locus and M7c168290649 locus to contribute PA and KT traits, respectively. These results suggested that these 6 QTL might explain phenotypic variations by both direct effects and epistatic interactions.

Discussion

Relationship of the identified QTL and the epistatically interacting loci

Epistatic interactions between loci or genes may play important roles not only in understanding the genetic basis of complex quantitative traits but also in the formation of heterosis (Phillips 2008). The epistatic interactions involved in plant height, starch synthesis, and cell extension have been published, but they have seldomly been concerned with seed physical traits (Stich et al. 2007; Ma et al. 2018; Wu et al. 2018). A previous study found that some embryo-sac, pollen, and spikelet fertility QTL had pleotropic effects on several traits by transinteracting with some loci on the other chromosome in rice (Li et al. 2017). In this study, 10 interaction pairs were found to be involved in 6 seed physical traits with 1–3 interactions for each trait. Six out of the 20 identified QTL located in these interacting loci (Table S3; Fig. 2), but only one QTL (qHKW9-1) controlled the same trait as the interacting loci, suggesting that the epistatically interacting loci might contain multiple genes with tightly genetic linkages but did not contain a pleiotropic gene with both direct effects and epistatic interaction effects on trait variation.

Candidate gene discovery for seeds physical traits

A total of 20 QTL were detected for KL, KW, KT, PA, KV, and HKW from this F2:3 population (Table S1). In addition, 6 chromosome regions were identified to harbor at least two QTL that controlled different traits (Table 2). After a meta-QTL analysis of 1321 seed physical trait-related QTL identified from this work and the previous publications, 8 mQTL were identified to contain 95 initial QTL (Table S2). It was interesting to find that the 6 chromosome regions highlighted by this work all fell into these 8 mQTL regions, suggesting important roles of these regions in controlling seed physical traits that could function in different genetic backgrounds.

Among the 8 mQTL, mQTL4-1 included two overlapped QTL (qKV4-1, qPA4-1) at the region 188.3–188.7 Mb on Chr. 4, and a neighboring QTL (qHKW4-1) at the 191.1–193.2 Mb region from this population (Table 2; S1). This region also harbored three KW QTL with two from a Mc × V671 F2:3 population and one from a Zheng58 × SK RIL population (Liu et al. 2014; Raihan et al. 2016), in addition to a KV QTL from a Ye478 × Qi319 RIL population (Zhang et al. 2017). Combining the QTL regions identified from all populations, the overlapped interval for these QTL could be narrowed down to a 0.2 Mb region (188.3–188.5 Mb), which contained 10 genes. Among them, the gene GRMZM2G132956, encoding a pentatricopeptide repeat superfamily protein (PPR), was a possible candidate that affected seed physical traits by participating in RNA intron splicing during seed development. The gene GRMZM2G030877, a bZIP transcription factor, might also influence seed development by regulating targeted gene expression (Table S4; Li et al. 2014; Wang et al. 2019; Li et al. 2018a).

The mQTL4-1 region showed an epistatic effect on KT by interacting with a translocus on Chr. 7: 167.9–173.8 Mb (Fig. 2; Table S3). A ubiquitin receptor gene, GRMZM2G009438, encoding the DA1-related protein 2 (DAR2), was found to locate in this region. Its homolog in Arabidopsis acted redundantly with DA1 to set final seed and organ size by restricting cell proliferation (Peng et al. 2013a, b). As the ubiquitin pathway has been known to play an important role in plant seed size determination (Li and Li 2014), four ubiquitin-related genes were noted to locate in the epistatically interacting loci of Chr. 7: 167.9–173.8 Mb. These genes encode a ubiquitin-like superfamily protein (GRMZM2G055527), an E3 ubiquitin ligase family protein (GRMZM2G020574), a ubiquitin-conjugating enzyme 34 (GRMZM2G045039), and a ubiquitin-specific protease 24 (GRMZM2G330690) (Table S4). They are possible interaction targets of DAR2 to affect the seed size or weight.

Additionally, qKL2-1, qKT2-1, and qPA2-1, identified in this study, overlapped with previous QTL for KL, KT, KS (kernel shape, calculated as formula of KL: KW: KT) and KRN (kernel row number) in a Ye478 × Qi319 RIL population (Zhang et al. 2017). These QTL belonged to the mQTL2-1 region and shared an overlapping region of 0.4 Mb (8.9–9.3 Mb), where 21 genes were located. A starch synthase Ш protein gene, GRMZM5G897776, could serve as a candidate for further functional analysis within this region (Table S4), as starch synthase plays a key role in endosperm development (Lv et al. 2008).

Relationship between seed physical characteristics and seed vigor performances

Previous works raised a controversial question about the correlations between seed physical traits and seed vigor traits, with positive correlations obtained from maize and cotton (Revilla et al. 1999; Ambika et al. 2014; Snider et al. 2016), while negative from pea (Peksen et al. 2004; Singh et al. 2005; Rastegar and Kandi 2011). In this work, seed physical traits (KW, PA, KV and HKW) showed positive correlations with seeds vigor traits (TL, RL, and SL; P < 0.01; Table S5), which seemed to support the opinion that large seeds have better germination performance and tend to produce more vigorous seedlings than small seeds (Ambika et al. 2014). However, data from this F2:3 population suggested that the small-seed parent L220 contained higher seed vigor than the large-seed parent PH4CV under low-temperature conditions (Li et al. 2018b). A possible explanation was that L220 contained more contributable QTL involved in seed vigor, which were not linked to the QTL for seed size or weight. Future work should be continued to identify the underlying genes controlling seed vigor and seed physical traits, which could highlight the genetic mechanism of seed physical traits in relation to seed vigor.