QTL data on maize grain yield (GY) and GY-related traits were collected from 33 studies published between 2000 and 2016 (Table S1). The GY-related traits included two important components: (1) ERT, including ear weight (EW), ear length (EL), ear diameter (ED), cob weight (CW), cob diameter (CD), and kernel row number (KRN); and (2) kernel-related traits (KRT), including kernel length (KL), kernel width (KWI), kernel thickness (KT), kernel number (KN), kernel weight (KW), kernel ratio (KR), and kernel volume (KV) (Table 1). Detailed QTL information and the associated literature on maize grain yield and related traits were collected from three databases, MaizeGDB (http://www.maizegdb.org), Gramene (http://www.gramene.org), and the PubMed web server (http://www.ncbi.nlm.nih.gov/pubmed). In this study, we collected only QTLs that were identified under normal growth conditions.

The collected QTLs were projected onto the “IBM2 2008 Neighbors” maize reference map (http://curation.maizegdb.org/cgi-bin/displaymaprecord.cgi?id=1140201) to generate a consensus map. Then, a meta-analysis was performed to integrate the QTL data from different studies and to refine the confidence intervals. For the meta-analysis of many QTLs, five different models (1-, 2-, 3-, 4-, or N-QTL) with different Akaike information criterion (AIC) values have been proposed using BioMercator software V4 (http://moulon.inra.fr/index.php/en), where the model with the lowest AIC-value is considered optimal. Finally, the consensus QTL presented by the optimum model is regarded as the MQTL (Arcade et al., 2004).

The physical intervals of these MQTLs were identified using the MaizeGDB and Gramene databases. The physical locations of flanking markers for these MQTLs were confirmed on the IBM2 2008 Neighbors map. The annotated transcripts within these MQTL regions were mined in the MaizeGDB database. The physical locations of these MQTL regions were based on genome annotation version AGPv2 of the maize B73 reference map (http://curation.maizegdb.org/). The gene sequences located in the MQTL regions were aligned to the NCBI non-redundant (nr) database with Blastx, using an E-value of < 10⁻⁵, with a hit number threshold of 100. The best functional annotations were obtained using this process. With the Nr annotation, we used the Blast2GO program (version: v2.5.0) to obtain the gene ontology (GO) annotation for these genes.

In this study, we collected 25 genes related to grain yield and related traits in rice (Table S3). Homologous genes in maize were identified as follows: (1) the protein sequences of these 25 collected rice genes were obtained from http://www.ricedata.cn/gene; (2) BLASTP (protein-protein BLAST) searches using these protein sequences were performed against the maize “non-redundant protein sequences” database (https://www.ncbi.nlm.nih.gov/). The criteria for these searches included an E-value of < 10⁻¹⁰, identity >60%, and a coverage region >60% to select homologous genes of these rice yield-related genes in maize.

An association mapping panel with 627 maize inbred lines covering highly diverse maize germplasms was applied in this study to perform regional association mapping. The genotypes and phenotypes of this association panel have been provided in our previous reports (Chen et al., 2016a; Qin et al., 2016). We selected SNP markers with minor allele frequencies >0.05 in MQTL regions to perform the analysis. The association analysis was estimated using a mixed linear model (MLM) incorporated in TASSEL V5.0, controlling for population structure (Q) and kinship (K). The first three principal components (PCs), which have been analyzed in previous studies, were used as the covariant variables to control for the existing population structure in the association mapping panel. Significant marker-trait associations were declared at LOD >3. The expression data from different kernel development values of these candidate genes were collected from the MaizeGDB database (http://maizegdb.org/) (Winter et al., 2007; Sekhon et al., 2011).

A total of 999 QTLs related to GY, ERT, and KRT were collected (Table S1). The number of QTLs per trait ranged from 11 (kernel ratio, KR) to 218 (kernel weight, KW) (Table 1). The 999 collected QTLs were distributed unevenly across the ten chromosomes (Figure 1A). The greatest number of QTLs (182) were located on chromosome 1, while chromosome 6 exhibited the fewest, with 68 (Figure 1A), similar to previous meta-analyses of maize yield QTLs (Wang et al., 2016). A total of 75.08% of the collected QTLs for each trait exhibited an R² < 10%, implying that the proportion of phenotypic variance explained by each QTL was very small (Figure 1B). These results suggest that grain yield and related traits in maize are mainly controlled by numerous loci of minor effect and display a complex genetic architecture.

The collected QTLs for maize yield and its related traits were projected onto the target map IBM2 2008 Neighbors via meta-analysis to mine MQTLs and refine QTL intervals. A total of 76 MQTLs were identified according to the models with the lowest AIC-values (Table 2, Figure 2). These MQTLs were distributed unevenly across the genome, with the number per chromosome ranging from 4 on chromosome 4–10 on chromosome 5 (Table 2, Figure 2). These MQTLs were named sequentially from MQTL-1 to MQTL-76 according to their chromosomal locations. MQTL-31 covered just two QTLs, while MQTL-28 interestingly covered 45 QTLs (Table 2). We also found that 60 MQTLs were related to GY, while 4 MQTLs were related to only kernel-related traits (KRT), and MQTL-17, on chromosome 3, was associated with only ERTs (Table 2).

Furthermore, we found that several yield-related MQTLs tended to show a clustered distribution in the maize genome. One of these clustered regions was detected in the genomic region from 38.17 to 69.15 Mb on maize chromosome 5 and consisted of two MQTLs (MQTL-33 and MQTL-34) (Table 2). Two MQTLs (MQTL-39 to MQTL-40) were clustered in a 28.5-Mb region on maize chromosome 6 (Table 2). Other clusters were found on maize chromosome 7 (from 9.78 to 34.19 Mb), chromosome 8 (from 146.89 to 163.31 Mb), and chromosome 9 (from 16.24 to 22.68 Mb) (Table 2). Five MQTLs (from MQTL-72 to MQTL-76) were clustered in a 16.24-Mb genomic region on maize chromosome 10 (Table 2).

Among the 76 MQTLs, 44 MQTLs with an interval of less than 5 Mb were selected for GO analysis of annotated transcripts. A total of 4,501 genes were identified in the 44 MQTL regions, whose detailed gene IDs are presented in Table S2. Multiple genes associated with yield-related traits in maize were found to be located in MQTL intervals. Fea2 (Bommert et al., 2013) and Ub2 (Chuck et al., 2014), which are important genes in the control of kernel row numbers in maize, were located in the MQTL-27 and MQTL-5 regions, respectively. The UBL1 gene, which plays an important role in kernel and seedling development by influencing pre-mRNA splicing (Li et al., 2016a), was located in MQTL-37. Three genes related to maize kernel development, emp2 (Fu et al., 2002), ZmZHOUPI (Grimault et al., 2015), and ZmReas1 (Qi et al., 2016), were located in the MQTL-11, MQTL-9, and MQTL-43 regions, respectively. According to the two-level WEGO classification, 2,257 genes were annotated, which were divided into three categories (cellular component, molecular function, and biological pathway) and further divided into 48 classes (Figure 3). We found that more than one thousand genes were related to the cell, cell part, metabolic process, cellular process, organelle, binding, and catalytic activity terms (Figure 3).

To identify candidate genes related to grain yield in the MQTL regions, we collected 25 genes that were functionally characterized as being associated with grain yield and its components from the rice genome (Table S3). Homologous genes were not found in the maize genome for four of these rice genes (APG, GW5, PGL1, qGL3), while eight of the rice genes (An-1, Bsg1, D2, DEP1, GIF1, GW2, LP, SMG1, SRS3, and SRS5) exhibited more than one homologous gene, and the other genes exhibited a single homologous gene in maize (Table S3). Finally, 11 MQTL regions containing 11 maize orthologs of 10 rice genes related to grain yield and its related traits were identified in this study.

MQTL-19 contained ZmGS5 (Gene ID: GRMZM2G13815), an ortholog of the well-characterized rice yield gene GS5 (Table 3). In rice, GS5 encodes a serine carboxypeptidase family protein that controls grain size and weight by regulating grain width and filling (Li et al., 2011). One candidate gene (Gene ID: GRMZM2G097275), homologous to OsSPL14, was identified in the MQTL-35 region (Table 3). OsSPL14 is a member of the SBP (squamosa promoter-binding-like transcription activator) family, and can improve the grain yield by regulating plant architecture (Miura et al., 2010). GN1a encodes a cytokinin dehydrogenase 2 protein and regulates grain number to improve grain yield in rice (Ashikari et al., 2005). A GN1a homolog (Gene ID: GRMZM2G325612) was identified in maize in the MQTL-54 region (Table 3). A homologous gene of D61 (Gene ID: GRMZM2G048294), which encodes a brassinosteroid receptor kinase and can regulate plant height to produce small grains in rice (Morinaka et al., 2006), was identified in the MQTL-58 region (Table 3). Homologs of other grain yield-related genes, such as LP1, SRS5, Ghd7, and DEP1, were also identified in MQTL regions (Table 3). These homologs of yield-related genes may play important roles in ear and kernel development in maize.

In our study, a total of 10 MQTL regions were found to be related to KRT alone or to both GY and KRT (Table 2). These 10 MQTL regions were selected to implement a regional association analysis. The integration of SNP genotyping data from the 10 selected MQTL regions with phenotype information (10-kernel length, 10KL; 10-kernel width, 10KW; 100-kernel weight, HKW) for all 627 accessions successfully revealed significant associations for four MQTL regions (MQTL-10, MQTL-39, MQTL-49, and MQTL-73).

The combination of meta-analysis and regional association mapping indicated that three genes (GRMZM2G359974, GRMZM2G301884, and GRMZM2G083894) may play important roles in maize kernel development. The expression data for the three candidate genes were collected from the MaizeGDB database (http://www.maizegdb.org/). We found that these genes were all expressed at different stages of maize kernel development. For instance, GRMZM2G359974, which encodes a transferase family protein, was highly expressed in endosperm-20 DAP, while GRMZM2G301884 was highly expressed at the early stage of grain development, and GRMZM2G083894, which encodes an AN1-like zinc finger protein, was expressed at different levels in different stages of kernel development in maize (Figure S1). Therefore, these genes should be considered the most promising candidates for involvement in maize kernel development for further functional validation.

In maize, most important agronomic traits, such as grain yield, ERT, and kernel-related traits, are complex, quantitative and controlled by numerous small effect QTLs. With increasing numbers of identified QTLs, researchers tend to compare the results of QTL analysis from different backgrounds and environments (Peng et al., 2011; Li et al., 2013). However, the results of QTL mapping are often inconsistent due to different population types, population sizes, mapping methods and genetic backgrounds. Therefore, QTLs localized to inconsistent genomic regions should be investigated further to validate these QTLs and verify their precise positions. Integrating QTL data from different studies and using meta-analysis to identify MQTL regions will pave the way for further fine mapping of QTLs and gene cloning (Wang et al., 2013, 2016; Martinez et al., 2016). In this study, 999 QTLs for maize yield and related traits were collected from different populations (Table S1). Through meta-analysis, a total of 76 maize yield MQTLs were identified in this study (Table 2, Figure 2). Compared with previous studies, we identified many MQTL regions that were highly consistent across different studies (Table S4). We also identified nine genomic regions for yield-related traits that had been identified in previous studies (Wang et al., 2013, 2016; Martinez et al., 2016) (Table S4). Intriguingly, we found that a major QTL (qGW4.05) for maize kernel size and weight that we recently fine-mapped was located in the MQTL-27 region on maize chromosome 4. On the same chromosome, a QTL for maize grain yield, qYPP4-1, was recently identified using a high-density genetic map, which was placed in the MQTL-28 region by our meta-analysis (Chen et al., 2016b). These results suggest that some MQTL regions identified in our study can serve as major QTLs for fine mapping and gene cloning of important yield genes.

In crops with large and complex genomes, such as maize and wheat, homology-based cloning methods are useful tools for identifying important genes related to complex traits. In recent years, many crop genome sequences have been released based on the wide application of next-generation sequencing technologies (Liu et al., 2015). These public genome sequences aid in the identification of conserved genomic regions and key genes in different crops.

With the numerous grain yield-related genes identified in rice, many homologous genes related to grain yield have been cloned in maize based on a comparative genomics strategy. In maize, multiple genes related to kernel size and weight have been identified, and their function has been validated through linkage analysis and association mapping analysis (Li et al., 2010a,b; Liu et al., 2015). ZmGS5, a maize homolog of the rice grain size and weight gene GS5 was identified in the MQTL-19 region in this study (Table 3). Previous studies have demonstrated that brassinosteroids such as brd2, dwf11, and d61 play an important role in seed development in rice. One gene homologous to D61 was identified in MQTL-58 (Table 3). The dwarf d61 mutant of rice exhibits smaller grains and a lower grain weight due to defects in brassinolide metabolism (Morinaka et al., 2006). The members of the SBP gene family are important transcription factors in plants, involved in plant growth and development. OsSPL14, which encodes an SBP transcription factor, controls shoot branching in the vegetative stage, and a higher grain yield can be achieved in rice via over-expression of this gene in the reproductive stage (Miura et al., 2010). One candidate gene (GRMZM2G097275) was identified in the MQTL-35 region, which was homologous to OsSPL14 (Table 3). The rice gene GN1a, which encodes a cytokinin oxidase, can increase grain number and improve grain yield. When the expression of GN1a is reduced, the cytokinin accumulates in inflorescence meristems, and the number of reproductive organs is increased (Ashikari et al., 2005). Furthermore, GRMZM2G325612, a homolog of GN1a, was mapped to the MQTL-54 region (Table 3). The homologs of the 10 genes in the MQTL regions have been shown to improve grain yield in rice, suggesting their possible roles in maize. As in previous studies, multiple methods of analysis, such as expression pattern analysis, phylogenetic analysis, linkage analysis, candidate gene mapping analysis, and transgenic analysis, should be used to validate the function of homologous genes. These homologous genes in MQTL regions should to be validated in the future using a number of the techniques mentioned above.

Association mapping and linkage mapping are two effective strategies for dissection of the genetic basis of complex quantitative traits in crops. These two methods each have particular advantages, such as a higher relative power and lower false positive rate for linkage mapping and a relatively higher mapping resolution for association analysis (Sneller et al., 2009). In recent years, the combination of linkage and association mapping strategies has been widely used in crops. One strategy is to construct integrated mapping populations such as MAGIC (multiparent advanced generation inter-crosses) and NAM (nested association mapping) populations (Buckler et al., 2009; Meng et al., 2016). Based on these types of populations, many complex quantitative traits have been resolved, such as flowering time, plant height, and disease resistance (Buckler et al., 2009; Peiffer et al., 2014; Ding et al., 2015). Another strategy is to first determine confidence intervals related to target traits based on linkage mapping analysis. Then, genome-wide association mapping or regional association mapping methods can be used to narrow down the confidence intervals of major QTLs and identify candidate genes or loci. Many major QTLs related to kernel size and weight have been finely mapped to candidate genes in maize (Chen et al., 2016a; Li et al., 2016b; Qin et al., 2016) or rapeseed (Li et al., 2014).

Combining meta-analysis and regional association mapping can rapidly identify candidate genes associated with complex agronomic traits, such as grain size and weight, in rice (Daware et al., 2017). In this study, we identified three candidate genes (GRMZM2G359974, GRMZM2G301884, and GRMZM2G083894) for kernel size and weight by combining meta-analysis and regional association mapping (Table 4, Figures 4–6). One gene (GRMZM2G359974) located in MQTL-10 region encoding a transferase family protein was found to be significantly associated with kernel width and weight through regional association analysis (Table 4, Figure 4). GRMZM2G083894, which encodes an AN1-like zinc finger domain protein, was found to be associated with kernel width and exhibited different expression levels in different stages of kernel development (Table 4, Table S4). The three genes (GRMZM2G359974, GRMZM2G301884, and GRMZM2G083894) are identified by combined the meta analysis and regional association mapping. And, we found the three genes were all expressed at different stages of maize kernel development. To date, many GWAS studies have been done for maize yield and its related traits, but these results are very different for many reasons such as the mapping population, the phenotype variation and the population structure. The three genes identified in our study were near by the significant SNPs with GWAS for kernel size and weight traits through combining 10 RIL populations (Liu et al., 2017). For example, GRMZM2G359974 were significant associated with kernel width and weight in this study and a SNP which located on 0.9 Mb upstream of this gene were also found significantly associated with kernel width by GWAS (Liu et al., 2017), and a SNP associated with kernel width was located on 1.5 Mb upstream of GRMZM2G083894. GRMZM2G083894 were located in the QTL interval which is found by using the RIL population from the cross of B73 and BY804 (Liu et al., 2017). Based on these results, these three candidate genes should be considered the most promising candidates related to kernel size and weight in maize for further functional validation.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.