Genome-Wide Association Analysis Identifies Loci and Candidate Genes for 100-Kernel Weight in Maize

Abstract

Maize is an important food crop, and 100-kernel weight (HKW) is one of the three key components of yield. In this study, 200 maize inbred lines were used as the material, and HKW was evaluated over three consecutive years in two environments. A genome-wide association study (GWAS) was conducted using the Blink and FarmCPU models with 44,935 SNP markers evenly distributed across the maize genome. A total of 25 SNPs significantly associated with HKW were identified, with three SNPs detected in both models. Six significant SNPs were located within previously mapped QTL bins associated with grain weight. In the linkage disequilibrium (LD) regions of the significant SNP loci, 198 candidate genes were identified, of which 74 had annotation information. Further analysis revealed 21 candidate genes related to HKW, such as GRMZM2G010555 (alternative oxidase), GRMZM2G102471 (ubiquitin-conjugating enzyme), GRMZM2G060669 (histone deacetylase), GRMZM2G090156 (methyltransferase), GRMZM2G002075 (BZIP transcription factor), and GRMZM2G138454 (bHLH transcription factor). The SNP loci and candidate genes identified in this study provide important references for marker-assisted selection, fine mapping, and gene cloning related to HKW in maize.

1. Introduction

Maize (Zea mays L.) is the most important grain crop in China and serves as a critical raw material for feed, food, bioenergy, and industrial production. It plays a pivotal role in national food security and economic development [1]. Yield is the primary breeding target for maize breeders. However, yield is a quantitative trait with low heritability and is significantly influenced by environmental factors, making its direct genetic analysis challenging. Kernel traits, such as kernel weight (KW), are key components of yield and are less affected by the environment [2,3]. Therefore, enhancing the genetic analysis of kernel-related traits and identifying genes associated with kernel weight is of great practical significance for elucidating the molecular mechanisms of maize kernel development and increasing maize yield through molecular methods.

Significant progress has been made in identifying QTLs associated with kernel weight in crops such as soybean, rice, and wheat [4,5]. In rice, several key grain weight-related genes have been cloned, including GS3, GS5, qGL3, and GW2 [6,7,8]. In maize, hundreds of QTLs associated with 100-kernel weight have been identified. For example, Raihan et al. [9] mapped 16 major QTLs controlling kernel weight and size across multiple environments using an RIL population. Liu et al. [10] identified 729 QTLs associated with kernel traits and weight using three statistical models across 10 RIL populations. Zhao et al. [11] identified 13 QTLs related to grain weight in two F_2:3 populations across eight water conditions using composite interval mapping. Su et al. [12] used the Genotyping-by-Sequencing (GBS) method to genotype 199 F₂ individuals derived from a cross between the maize inbred lines SG-5 and SG-7, constructing a genetic linkage map and identifying four QTLs associated with 100-kernel weight. However, only a few grain weight-related genes have been cloned and studied in detail [13,14].

Genome-wide association studies (GWASs) have become an essential tool for systematically dissecting the genetic mechanisms underlying complex quantitative traits [15]. GWAS was first applied to maize to study the Dwarf8 gene, revealing the association between polymorphisms in this gene and flowering time [16]. Maize inbred lines are highly genetically diverse, and their linkage disequilibrium (LD) decays rapidly, making them particularly suitable for GWAS analysis. Studies have shown that the resolution of GWASs is 5000 times higher than that of traditional linkage mapping [16,17]. GWAS has been widely used to identify QTLs associated with complex traits such as plant architecture, yield, and stress resistance [18]. For example, Peng et al. [19] used GWAS to analyze 285 maize inbred lines and detected 27 SNPs significantly associated with leaf angle, explaining 5.54–8.73% of the phenotypic variance. Ma et al. [20] used GWAS to analyze 309 maize inbred lines and identified 12 SNPs significantly associated with ear shaft diameter, uncovering 17 candidate genes. Qu et al. [21] used GWAS to analyze the genetic structure of kernel size in 212 maize inbred lines and identified 38 significant SNPs, leading to the discovery of 58 candidate genes related to kernel development.

Currently, there are relatively few studies utilizing GWAS to analyze the genetic basis of 100-kernel weight in maize. In this study, a GWAS was conducted using 200 maize inbred lines, with phenotypic data collected over three years across two environments. Using 44,935 SNP markers covering the entire maize genome, significant loci associated with kernel weight were identified, and candidate genes were predicted. This study provides an important reference for breeding high-yield maize varieties and for further cloning of kernel weight-related genes.

2. Materials and Methods

2.1. Experimental Materials and Field Design

The 200 maize inbred lines used in this study were sourced from the molecular breeding group at the Institute of Organic Dryland Agriculture, Shanxi Agricultural University, including 13 parents of major expanded hybrids, 26 US imported lines and 161 newly developed inbred lines from different breeding institutions of China. Detailed information of accessions is listed in Supplementary File S1. The field trials were conducted over three years (2019, 2020, and 2021) at two locations: Jinzhong (Shanxi Province, 112.69° E, 37.56° N) and Xinzhou (Shanxi Province, 110.53° E, 38.60° N). A randomized complete block design (RCBD) was used with two replications. Each inbred line was planted in two rows of 5 m in length, with a spacing of 50 cm between rows and 23 individual plants per row. The planting density was around 6000 plants per mu. Standard field management practices were followed throughout the growing season.

2.2. Phenotypic Data Collection and Statistical Analysis

During the silking stage, 10 uniformly growing plants were selected from each plot and bagged for self-pollination, which was performed 1–2 days after silking. These plants were used for TKW determination. After the maize matured, five ears of similar size and shape from healthy plants in each plot were harvested. The kernels from the middle section of each ear were mixed, and 100 kernels were randomly selected and weighed. This process was repeated three times, and the average weight was used for further analysis. A mixed linear model was constructed using the R package lme4 to estimate the Best Linear Unbiased Prediction (BLUP) values for HKW. The broad-sense heritability (H²) across multiple years and locations was calculated using the formula:

$$H^2={\displaystyle \frac{V_g}{V_g+\frac{V_{gl}}{L}+\frac{V_{gy}}{Y}+\frac{V_{\sum }}{nLY}}}$$

where V_g is the genetic variance, V_gl is the variance of genotype-by-location interaction, V_gy is the variance of genotype-by-year interaction, V_Σ is the residual variance, and n, L, and Y are the number of replications, locations, and years, respectively [22]. The R package Performance Analytics was used to perform correlation analysis on the TKW data and BLUP values across different years and locations.

2.3. Genotyping and Quality Control

Genomic DNA was extracted from fresh young leaves using a EasyPure^® Plant Genomic DNA Kit (TransGEN Biotech., Beijing, China). The purity and integrity were assessed using 1% agarose gel electrophoresis, and the DNA concentration was quantified using Qubit2.0 (ThermoFisher Scientific Inc., Waltham, MA, USA). The genotyping was performed via a GenoBaits^® Maize 40K Panel (Boruidi Biotech., Beijing, China) with 44,935 SNPs. The raw data were filtered for quality using the FASTQ software [23] and then aligned to the maize reference genome (B73 RefGen_v3) using the BWA software (v0.1.17) [24]. The software GATK 4.0 was employed for SNP detection [25]. After removing the low-quality or SNPs with missing rate > 50% and minor allele frequency (MAF) <0.01 with the PLINK tool [26], 43,034 high-quality SNPs remained. Genotype imputation was performed using Beagle (V4.1).

2.4. Genome-Wide Association Analysis

Two models, linkage-disequilibrium iteratively nested keyway (Blink) and fixed and random model circulating probability unification (FarmCPU) in the GAPIT version 3.0, were employed for the genome-wide association study (GWAS) of HKW and BLUP for HKW. The population structure was controlled by the first three PCs. The phenotypic variation explained (PVE) by each SNP was calculated using a mixed linear model (MLM) conducted in GAPIT V3.0 R software. Q-Q and Manhattan plots were generated using the R package CMplot based on the GWAS results. A Bonferroni-corrected p-value of 2.32 × 10⁻⁵ (1/total number of markers) was employed as the significant threshold, with a −log10 (p-value) of 4.63.

2.5. Candidate Gene Identification

The linkage disequilibrium (LD) decay distance and the square of the correlation coefficient (r²) was calculated using the PopLDdecay software (v3.42). LD decay plots were generated using the Origin software (v2019b). In our study, the LD decay distance was 200 kb when r² = 0.2. Genes located within 200 kb upstream and downstream of a significant SNP were retrieved as candidate genes for HKW based on the reference genome (B73 RefGen_v3, available at http://plants.ensembl.org (accessed on 12 May 2023)). Candidate genes were further annotated using databases MaizeGDB (https://staging.maizegdb.org/ (accessed on 10 January 2024)), NCBI (https://blast.ncbi.nlm.nih.gov/ (accessed on 10 January 2024 )), and UniProt (https://www.uniprot.org/ (accessed on 10 January 2024)).

3. Results

3.1. Phenotypic Analysis of 100-Kernel Weight

The HKW of the 200 maize inbred lines was analyzed over three years at two locations. The broad-sense heritability of HKW across three years and two locations was 80.78%. The phenotypic variation ranged from 14.23 g to 55.00 g. The highest HKW was observed in 2020 at both Yuci and Xinzhou, with means of 30.79 g and 30.02 g, respectively. The inbred line DH103-3 had the highest HKW across all years and locations, ranging between 39.46 g and 45.00 g. The coefficients of variation for HKW across different years and locations ranged from 16.17% to 22.49%, with the largest variation observed in Yuci in 2020. The skewness and kurtosis values of HKW in all environments were less than 1, indicating the HKW followed a normal distribution (

Table 1. Statistical analysis of 100-grain weight in different years and locations.

Year	Location	Mean	Max	Min	KURT	SKEW	SD	CV (%)	H2 (%)
2019	YC	26.01	44.52	15.81	0.79	0.58	5.30	20.35	85.64
	XZ	26.97	44.58	16.72	0.77	0.48	5.22	19.35
2020	YC	30.79	55.00	17.83	0.44	0.38	6.92	22.49
	XZ	30.02	44.21	14.23	−0.17	−0.29	6.47	21.54
2021	YC	28.12	42.54	16.76	−0.15	0.34	5.19	18.46
	XZ	29.34	41.56	16.34	0.08	−0.24	5.42	19.93
BLUP	—	28.24	42.34	18.43	0.12	0.41	4.56	16.17

Note: The units of the mean, maximum and minimum values are grams (g); YC and XZ represent the Yuci test area and Xinzhou test area.

). The combined variance analysis demonstrates that the effects of year, genotype, Y × L interaction, and Y × G interaction have all achieved extremely significant levels. However, the location, L × G interaction, and Y × L × G interaction had a relatively minor impact on the HKW (

Table 2. The joint variance analysis of 200 maize inbred lines’ HKW.

Source of Variation	Sum of Squares	DF	Mean Squares	F-Value
Location (L)	6604	1	3302	149.803 ***
Year (Y)	16	2	16	0.710
Genotype (G)	49,549	198	250	11.352 ***
Y × L	545	2	273	12.368 ***
L × G	11,372	330	34	1.563 ***
Y × G	3119	168	19.00	0.842
Y × L × G	6867	326	21	0.969
Residuals	21,691	984	22

Note: The significance levels of 0.001 are marked with ***.

).

Correlation analysis showed significant positive correlations (p < 0.001) between the HKW values of the 200 maize inbred lines across multiple years and locations, with correlation coefficients ranging from 0.468 to 0.901. The highest correlations were observed between the BLUP values and the HKW from Xinzhou in 2021 and 2020, with correlation coefficients of 0.901 and 0.907, respectively. The lowest correlation was found between the HKW data from Yuci in 2019 and 2020, with a correlation coefficient of 0.468 (Figure 1).

The diagonal in the figure represents the traits, and the upper triangle represents the correlation coefficient and the degree of significance. The significance levels of 0.05, 0.01 and 0.001 are marked with *, **, ***, respectively. The lower triangle is a scatter plot of the weight per hundred grains in different years and environments. YC and XZ represent Yuci and Xinzhou, respectively, and 2019–2021 represent the years.

3.2. Genome-Wide Association Analysis of 100-Kernel Weight

To accurately identify the significant loci associated with maize TKW, two models, Blink and FarmCPU, were employed for GWAS. With the threshold at −log10 p > 4.63, a total of 25 SNPs significantly associated with HKW were detected across chromosomes 1 to 5 (Figure 2). The Blink model detected 15 significant loci, with the p-values ranging from 5.34 × 10⁻¹⁴ to 9.97 × 10⁻⁶. The SNP 4_228498570 located on chromosome 4 had the lowest p-value of 5.34 × 10⁻¹⁴. Four significant SNP loci were identified based on BLUP values. The SNP with the largest phenotypic variance explained (PVE) in the Blink model was 2_186957636 located on chromosome 2, explaining 10.77% of phenotypic variation. The FarmCPU model detected 13 significant SNPs, with the p-values ranging from 7.45 × 10⁻¹⁰ to 1.93 × 10⁻⁵. The most significant SNP 5_142616635, which explained 9.05% of the phenotypic variance, was detected in both XZ and YC in 2021. One significant SNP was detected based on BLUP values under this model. SNPs 3_125129946, 4_238711294 and 5_142616635 were co-located by both models.

3.3. HKW Candidate Gene Analysis

According to the significant SNP combined with the linkage disequilibrium (LD) decay distance of 200 kb (r² = 0.2) (Figure 3), candidate genes were searched in B73_RefGen_v3 reference genome. A total of 198 candidate genes were identified. Functional annotation revealed that 74 of these genes had annotation information. Through literature retrieval, 21 candidate genes affecting HKW were identified (

Table 3. Candidate genes for 100-kernel weight and functional annotation.

Method	SNPID	Chr.	p-Value	R2 (%)	Envir.	Candidate Genes
FarmCPU	1_274302441	1	1.09 × 10−8	4.91	2021YC	GRMZM2G135283 (Serine hydroxymethyltransferase)
FarmCPU	2_2636802	2	1.17 × 10−5	1.02	2020YC	GRMZM2G010555 (alternative oxidase); GRMZM2G102471 (ubiquitin-conjugating enzyme 30)
FarmCPU	3_122617408	3	1.93 × 10−5	1.50	2019XZ	--
FarmCPU	3_122617431	3	1.93 × 10−5	7.01	2019XZ	--
FarmCPU	3_122928920	3	1.93 × 10−5	2.10	2019XZ	GRMZM2G045435 (beta-1,4-N-acetylglucosaminyltransferase family protein)
FarmCPU	3_122929643	3	1.93 × 10−5	4.01	2019XZ	-
FarmCPU/Blink	3_125129946	3	1.88 × 10−5	8.12	2019XZ,2019XZ	-
FarmCPU	3_229327953	3	2.22 × 10−6	0.01	2021YC	GRMZM2G093175 (Asparagine synthetase); GRMZM2G354558 (Omega-3 fatty acid desaturase)
FarmCPU	3_36009926	3	1.74 × 10−5	2.22	2021YC
FarmCPU/Blink	4_238711294	4	8.27 × 10−6/1.47 × 10−8	7.44	BLUP/2019YC	GRMZM2G060669 (Histone deacetylase complex subunit SAP18, GRMZM2G090156 (Methyltransferases)
FarmCPU/Blink	5_142616635	5	7.45 × 10−10/1.27 × 10−6	9.05	2021XZ/2021YC	GRMZM2G162748 (Cytochrome b6-f complex iron-sulfur subunit)
FarmCPU	5_18123084	5	1.63 × 10−6	1.32	2019XZ	GRMZM2G061292 (Dof transcription factor gene); GRMZM2G320754 (B3 domain transcription factors)
FarmCPU	5_201092307	5	5.99 × 10−7	0.15	2020YC	GRMZM2G101287 (Adenine nucleotide alpha hydrolases-like superfamily protein); GRMZM2G347767 (Isoflavone reductase)
Blink	1_50864066	1	5.66 × 10−8	7.17	2019XZ	-
Blink	4_1490461	4	4.10 × 10−6	4.46	2019XZ	GRMZM2G015232 (Putative cytochrome P450 superfamily protein); GRMZM2G349791(O-methyltransferase ZRP4)
Blink	5_210354193	5	9.97 × 10−6	1.06	2019XZ	GRMZM2G113418 (Glutaredoxin 2); GRMZM2G154332 (SAUR12-auxin-responsive SAUR family member)
Blink	2_186957636	2	3.55 × 10−7	10.77	2019YC	GRMZM2G002075 (BZIP transcription factor;); GRMZM2G138454 (bHLH transcription factor)
Blink	1_298487278	1	3.54 × 10−7	7.28	2021XZ	GRMZM2G159353 (putative cytochrome P450 superfamily protein)
Blink	2_65602039	2	1.19 × 10−7	6.05	2021XZ	GRMZM2G015024 (50S ribosomal protein L22,ribosomal protein L22)
Blink	2_79335071	2	2.076 × 10−6	6.55	2021XZ	-
Blink	3_177543426	3	4.31 × 10−7	6.17	2021XZ	-
Blink	4_188053528	4	1.16 × 10−6	1.34	2021YC	-
Blink	3_178779239	3	1.35 × 10−8	7.31	BLUP	-
Blink	4_228498570	4	5.34 × 10−14	6.75	BLUP	-
Blink	5_68455590	5	1.12 × 10−9	4.95	BLUP	-

), including 3 transcription factor genes: GRMZM2G002075 (BZIP transcription factor) and GRMZM2G138454 (bHLH transcription factor) on chromosome 2 and GRMZM2G061292 (Dof transcription factor) on chromosome 5. Additionally, two cytochrome P450-encoding genes were identified: GRMZM2G159353 on chromosome 1 and GRMZM2G015232 on chromosome 4. Other key candidate genes identified included those encoding important plant enzymes such as GRMZM2G010555 (alternative oxidase) and GRMZM2G102471 (ubiquitin-conjugating enzyme) on chromosome 2 and GRMZM2G060669 (histone deacetylase) and GRMZM2G090156 (methyltransferase) on chromosome 4.

The horizontal axis represents the physical distance between single nucleotide loci (SNP) on the same chromosome, and the vertical axis represents the linkage disequilibrium parameter r² value.

3.4. Allelic Variation Effects

Four significant SNPs with a larger PVE were selected for allelic effect analysis, including 2_186957636 (10.77%), 3_125129946 (8.12%), 4_238711294 (7.44%), and 5_142616635 (9.05%). These four loci showed a significantly affected HKW across all seven environments (Figure 4). For instance, the G/G allele of 2_186957636 increased the HKW by 1.12–10.37 g compared to the C/C allele across all seven environments, identifying G/G as the superior allele. Similarly, the G/G allele of 3_125129946 increased the HKW by 4.44–10.84 g compared to the T/T allele, thus identifying G/G as the superior allelic variant, while the C/C allele of 4_238711294 and the T/T allele of 5_142616635 also contributed to increased HKW.

4. Discussion

High grain yield is one of the most important goals persistently pursued by breeders in their professional endeavors. However, maize yield is an extremely intricate quantitative trait regulated by numerous genes with minor effects. Compared to other quantitative traits, it manifests a lower heritability (H²). Owing to its complexity, yield has often been replaced by ear traits in genetic studies, such as HKW which has a higher H² [27]. In this study, HKW was evaluated over three consecutive years in two environments. The H² of HKW was 85.64%, indicating that HKW is mainly controlled by genetic factors and is suitable for early generation selection [28,29]. The ANOVA indicated highly significant differences among genotypes (p < 0.001). The high genetic variation that exists among genotypes is very useful for maize breeders to select the highest HKW genotypes efficiently. Genotype (G) × Location (L) interactions (GLIs) were significant (p < 0.001) for HKW, inferring that the different genotypes exhibit variable responses to environmental conditions at each location. Genotype (G) × year (Y) was not significant, suggesting that the impact of years on HKW is relatively small.

Genome-wide association studies (GWASs) are an effective method for analyzing the genetic mechanisms underlying complex quantitative traits [30]. In this study, 25 SNPs significantly associated with HKW were identified. The Blink model detected 14 SNPs, and the FarmCPU model detected 16 SNPs. Several of these SNPs were located in regions consistent with previously identified QTLs. For example, the two SNPs on chromosome 1 (1_274302441 and 1_298487278) were located in bins 1.10 and 1.12, explaining 4.91% and 7.28% of the phenotypic variation. Previous studies have also detected grain weight-related QTLs in this region using different populations [31,32]. For instance, Peng et al. [33] identified a kernel density QTL (Qqkden1) and a kernel thickness (Yqkthi1) in bin 1.10. Similarly, the significant SNPs on chromosome 4 (4_188053528 and 4_228498570) located in bins 4.08 and 4.09 explained 1.34% and 6.75% of the phenotypic variation. Zhao et al. [34] mapped three stable QTLs controlling ear weight, cob weight, and HKW under different water stress conditions in this region. Additionally, the significant SNP 3_229327953 was located in bin 3.09. Within this interval, a QTL controlling grain protein and starch content was identified using three RIL populations in three different environments [35]. The significant SNP 5_142616635 was located in bin 5.04, and it accounts for 9.05% of the total phenotypic variation. Previous research indicated this area contains QTLs that control yield and starch content, which can be utilized for yield-related genetic improvement [36,37]. Five significant SNPs (3_122617408, 3_122617431, 3_122928920, 3_122929643, and 3_125129946) located in bin 3.04 explained a maximum phenotypic variation of 7.01%. However, no previously reported QTLs have been found in this region, suggesting that they might be novel loci associated with HKW.

Based on the physical locations of the significant SNPs and LD decay distances, 180 candidate genes for HKW were identified. Candidate genes for the significant SNP 2_2636802 include GRMZM2G010555 (alternative oxidase, AOX) and GRMZM2G102471 (encoding ubiquitin-conjugating enzyme). GRMZM2G010555 encodes an alternative oxidase (AOX), which is upregulated under low nitrogen conditions [38]. Mutations in these genes can alter the absorption efficiency of nitrogen and carbon in Arabidopsis [39]. Knockdown of the GmAOX2b gene leads to a reduction in seed yield [40]. GRMZM2G102471 encodes a ubiquitin-conjugating enzyme. Relevant studies have indicated that the ubiquity–proteasome pathway plays a critical role in regulating and controlling the grain size [41]. Overexpression of the rice ubiquitin-conjugating enzyme gene OsUBC45 enhanced disease resistance and increased grain size, thus improving yield [42]. The candidate genes for SNP 4_238711294 include GRMZM2G060669 and GRMZM2G090156. GRMZM2G060669 encodes a histone deacetylase (SAP18), which is involved in the regulation process of seed dormancy, maturity, and germination in maize [43]. Overexpression of hda101 in maize led to a reduced grain size [44]. GRMZM2G090156 encodes methyltransferase. Overexpression of a caffeic acid O-methyltransferase (OsCOMT) gene increases grain yield per plant even in a high-yield variety background [45], and silencing of the wheat SAM-dependent methyltransferases enzymes (TaSAM) can result in reduced grain yield [46].

As for SNP 5_18123084, candidate gene GRMZM2G061292 encodes a Dof transcription factor. These transcription factors are involved in the accumulation of endosperm storage proteins and starch, which are closely related to seed yield and quality. For instance, ZmDOF36 positively regulates starch accumulation in maize kernels [47]. GRMZM2G002075 and GRMZM2G138454, associated with SNP 2_186957636, are BZIP and bHLH transcription factors, respectively, and both are associated with grain development. ZmbZIP22 negatively regulates starch synthesis in maize kernels, and overexpression of ZmbZIP22 results in smaller grains and reduced grain weight [48]. Overexpression of the bHLH transcription factor ZmBES1/BZR1 significantly increases seed size and weight in Arabidopsis and rice [49]. All of the above genes are important candidate genes that control the hundred-grain weight of maize. This research provides information for the fine mapping and cloning of 100-kernel weight genes, contributing to the understanding of the developmental mechanism of maize kernels.

5. Conclusions

A genome-wide association study (GWAS) was conducted in this study using two models with 44,935 SNP markers. A total of 25 SNPs significantly associated with HKW were identified. Six significant loci were located within previously mapped QTL bins associated with grain weight. Further analysis revealed 98 candidate genes related to HKW. Among them, 21 candidate genes are most likely be related to HKW, such as GRMZM2G010555, GRMZM2G102471, GRMZM2G060669, GRMZM2G090156, GRMZM2G002075 and GRMZM2G138454. The SNP loci and candidate genes identified in this study provide important references for marker-assisted selection, fine mapping, and gene cloning related to HKW in maize.