Multi-model GWAS reveals key loci for leaf angle in maize inbred lines from southeast China Authors: Keyu Li, Jia Xie, Jinghan Cao, Changjin Wang, Wangfei He, Renjie She, Ruzhen Bao, Yuchen Chai, Haoze Yu, Yu Han, Weidong Huang, Xinxin Cheng, Mingliang Cui, Lei Chen, Haibing Yu Published: 2026 ABSTRACT AbstractBackgroundMaize is a vital grain and forage crop with continuously increasing yield demands, making the improvement of plant architecture for high-density planting particularly significant. Among architectural traits, leaf angle (LA) is a key factor influencing light interception and canopy architecture. To dissect the genetic basis of LA in maize, a genome-wide association study was conducted.ResultsA panel of 212 maize inbred lines from breeding programs in Southeast China was genotyped using the Maize6H-60 K SNP array. Employing five GWAS models (GLM, MLM, SUPER, FarmCPU, and BLINK), we identified 14 significant SNPs (p < 8.55 × 10⁻⁷) and 342 candidate genes within the 200 kb flanking regions, 134 of which were functionally annotated. Multi-model cross-validation revealed six stably associated SNPs, including four novel loci (AX-107936049, AX-108102817, AX-107953621, and AX-108061477). AX-86,305,869 was located within 2 Mb of a previously reported significant SNP, while AX-107,994,145 fell within a known QTL interval. Within these six stable loci, 11 high-confidence candidate genes are putatively associated with LA variation, potentially acting via hormone signaling, leaf morphogenesis, and midvein development. Notably, GRMZM2G036905 (homologous to Arabidopsis ELF1) may play a role in regulating upright leaf architecture, and GRMZM2G003565 may contribute to leaf rolling as previously suggested. Based on these stable SNPs, KASP markers were developed and validated in 52 varieties/lines (including Zhengdan 958, MY73, and Anke 985) and 20 inbred lines, confirming the association between these SNPs and LA variation.ConclusionOur study identified six stable SNPs and 11 candidate genes for maize LA through multi-model GWAS in 212 inbred lines. Four novel loci were discovered, and KASP markers were developed for three stable SNPs, providing genetic resources and tools for breeding maize varieties tolerant to high planting density.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12870-026-09162-w.Keywords: Maize, Inbred germplasm, Genome-wide association studies, Leaf angle, Candidate genes, KASP markers Introduction Maize (Zea mays L.) is a major global food crop with extensive applications in human consumption, animal feed, and industrial processing [1]. Its yield level is crucial for ensuring food security and supporting economic development. While global maize production has maintained steady growth [2], demand is projected to rise further due to population increase and economic development. By 2033, global grain consumption is projected to reach 3.2 billion metric tons, with maize accounting for the largest share [3]. To address future supply pressures, genetic improvement of plant architecture to enhance yield potential has become a key direction in maize breeding [4–6]. Leaf angle (LA), defined as the angle between the leaf midrib and the stem [7, 8], is a key determinant of ideal plant architecture in maize [6, 9]. LA directly influences light interception efficiency and tolerance to high planting density, thereby affecting overall canopy light utilization and ultimately grain yield [10–12]. Elucidating the genetic basis of LA is therefore essential for optimizing plant architecture and improving yield in maize. Recent advances in molecular biotechnology and high-throughput sequencing have greatly promoted the application of genome-wide association studies (GWAS) in dissecting the genetic architecture of agronomic traits in crops such as cotton [13], maize [14], rapeseed [15], rice [16], and wheat [17]. This approach detects statistical associations between phenotypic variation and genome-wide genetic markers across diverse germplasm panels, enabling the identification of loci underlying complex traits [18]. Accumulating evidence from GWAS has advanced our understanding of the genetic basis underlying maize LA. For example, Lu et al. performed GWAS using 80 key maize inbred lines across two years, identifying 22 significant SNPs and candidate genes, five of which were validated via qRT-PCR [19]. In another study, Duan et al. conducted GWAS for LA using 492 inbred lines evaluated in five environments, detecting 85 significant SNPs and revealing seven meta-QTLs [20]. Similarly, Peng et al. analyzed LA phenotypes collected over three years in 285 inbred lines, identifying 96 significant SNPs and predicting seven candidate genes [21]. Current evidence indicates that genes regulating LA function mainly through pathways controlling auricle and ligule development (e.g., LG1, LG2, ZmBEH1) [22–24], establishment of leaf polarity (e.g., MWP1, ZmHB53) [25, 26], phytohormone signal transduction (e.g., LGN, Brachytic2, ZmILI1, ZmACS7, UPA1, UPA2) [27–30], and regulation of mechanical tissue strength in the midrib (e.g., ZmDRL1, ZmDRL2) [31]. A recent study employing map‑based cloning identified two additional genes on chromosome 4 that regulate maize LA, encoding a 3‑hydroxyacyl‑CoA dehydrogenase (AIM1) and a RING‑type E3 ubiquitin ligase (SINAT4), respectively [32]. Although a number of regulatory genes have been characterized, they likely do not represent the full genetic architecture of this trait. Therefore, further systematic mining of associated loci and candidate genes is necessary to elucidate the molecular mechanisms underlying LA variation. Despite an increasing number of studies reporting LA-associated QTL and genes, only a limited subset has been effectively applied in maize breeding. A primary reason is that, following long-term selection during crop improvement, favorable QTLs and alleles may have reached high frequency, near fixation, or fixation within elite germplasm used in specific breeding programs. As a consequence, loci identified from diverse germplasm panels may not maintain sufficient polymorphism within these breeding populations to support further genetic improvement [33]. Moreover, due to differences in genetic materials, environmental conditions, and analytical methods across studies, the significant SNPs detected often vary [34]. When the target breeding environment differs substantially from the original conditions, the results tend to show limited reproducibility and reduced breeding effectiveness. Therefore, further in-depth investigation into the genetic mechanisms of LA in maize inbred lines derived from breeding programs in southeastern China is warranted. In this study, an association panel consisting of 212 maize inbred lines was used, together with LA phenotypic data collected over two consecutive years. Five models (GLM, MLM, SUPER, FarmCPU, and BLINK) were employed to identify QTL regulating LA and to screen candidate genes. Furthermore, KASP markers were developed based on significant SNP, providing a theoretical basis and practical tools for fine mapping, gene cloning, and molecular breeding of ideal plant architecture in maize. Materials and methods Plant materials and field experiment design The 212 maize inbred lines used in this study were obtained from the Maize Breeding Engineering Technology Research Institute of Anhui Province, China [35]. Field trials were conducted in Fengyang County, Anhui Province (32°N, 117°E) during the 2023 (23FY) and 2024 (24FY) growing seasons, constituting two distinct environments. A single-row plot design was adopted with the following specifications: row length of 5.00 m, row spacing of 0.60 m, and plant spacing within rows of 0.33 m. The experiment was arranged with two replications. All field management practices followed standard agronomic protocols for maize cultivation. Phenotypic measurements In this study, LA phenotyping was conducted at the early milk stage, when maize leaves remained green and had not yet senesced. Five representative plants per inbred line were randomly selected from each of the two replicates, and the measured leaves were ensured to be intact and free from severe senescence. The LA was measured as the angle (°) between the leaf midrib and the stem for three consecutive leaves above the ear [7, 8]. The average value of these measurements was considered the phenotypic value for each plant. Descriptive statistics—including mean, variance, standard deviation, coefficient of variation, range, skewness, and kurtosis—were computed using Microsoft Excel 2021 and IBM SPSS Statistics 23.0. Pearson correlation analysis and normality tests were also performed. For analysis across the two-year environments, best linear unbiased estimates (BLUEs) of LA across both years were estimated using the lmer function from the lme4 package in R [36]. Genotype was treated as a fixed effect, while replicate and year were included as random effects in the model. Broad-sense heritability was estimated using the following formula: In the formula, denotes genetic variation, represents the interaction between genotype and environment, indicates residual error, n refers to the number of environments, and r signifies the number of repetitions within each environment [37]. Genotyping Genotyping was performed using the Maize6H-60 K SNP array developed by the Maize Research Center, Beijing Academy of Agriculture and Forestry Sciences [38]. SNP loci with a minor allele frequency (MAF) below 5% and a missing rate exceeding 20% were excluded. A total of 58,455 high-quality SNP markers were retained for subsequent analyses. PopLDdecay 3.40 was used to evaluate genetic relationships among individuals and estimate linkage disequilibrium (LD) based on the coefficient (r²), from which an LD decay plot was generated [35]. Genome-wide association study Genome-wide association analysis of maize LA was conducted using five models implemented in the R package GAPIT version 3.4 [39, 40]: GLM (General Linear Model), MLM (Mixed Linear Model), SUPER (Settlement of MLM Under Progressively Exclusive Relationship), FarmCPU (Fixed and Random Model Circulating Probability Unification), and BLINK (Bayesian-information and Linkage-disequilibrium Iteratively Nested Keyway). Based on our previous work, the first three principal components (PCs) were incorporated as covariates in all models to control for population structure. GLM is computationally efficient but tends to produce inflated false positives due to the absence of kinship correction. The kinship matrix (K) was calculated using PLINK 1.9 and incorporated into the MLM model, which accounts for relatedness and reduces false positives, albeit often at the cost of statistical power [41]. Notably, the SUPER model does not rely on a genome-wide kinship matrix constructed from all markers; instead, it uses an optimized relationship matrix derived from a subset of associated markers [42]. In multi-locus approaches, FarmCPU iteratively fits fixed- and random-effect models, using pseudo quantitative trait nucleotides (QTNs) to control background confounding [43]. BLINK further replaces the random‑effect model with a fixed‑effect model and selects markers based on Bayesian information criteria and linkage disequilibrium [39, 40]. To control for false positives, the Bonferroni correction was applied, resulting in a significance threshold of –log₁₀(p) > 6.07. The validity of the association results was assessed using Q-Q plots and p-value distributions. To evaluate potential inflation of the GWAS test statistics, the genomic inflation factor lambda (λ) was calculated using R according to the following formula: λ = median(qchisq(1 − p, 1))/qchisq(0.5, 1), where p represents the vector of P-values from the GWAS results. A λ value close to 1 indicates no obvious inflation of the test statistics [44]. Candidate genes were predicted based on the linkage disequilibrium (LD) decay characteristics of the population, with gene annotations retrieved from the MaizeGDB database (B73 RefGen_v3 served as the reference genome for SNP design). The LD decay distance was estimated as 200 kb (r² = 0.15) [35]. Functional annotation of candidate genes For significant SNP loci identified through association analysis, candidate genes were defined as those located within a 200 kb genomic region (upstream and downstream) of each marker, based on LD structure characterization [35]. Functional prediction and annotation of candidate genes were performed using the following biological databases: MaizeGDB (https://maizegdb.org/gbrowse/maize_v3/), NCBI (https://www.ncbi.nlm.nih.gov/), TAIR (https://www.arabidopsis.org/), and UniProt (https://www.uniprot.org/). Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses were conducted using the agriGO tool (https://systemsbiology.cau.edu.cn/agriGOv2/), the KEGG database (https://www.kegg.jp/), and the GeneCloud platform (https://www.genescloud.cn/). RNA-seq data were obtained from the Maize Gene Function and Expression Database (MaizeGDB). A heatmap of candidate gene expression was generated using the R programming language. Gene expression was considered detectable at a threshold of FPKM ≥ 1, and expression levels were normalized using the formula log₂(FPKM + 1) [45]. Protein–protein interaction (PPI) networks were constructed using the STRING database (https://string-db.org/) and visualized with Cytoscape software (v3.10.3). Characterization of superior allelic variants at significant loci and KASP markers development Based on the association analysis results, allelic effect analysis was performed for the stable SNP loci identified. Group differences were assessed using t-tests, and corresponding box plots were generated with GraphPad Prism 8 and Origin 2021. KASP primers (Table S6) were designed using PrimerPicker version 6. Each 10 µL KASP reaction contained 1 µL of genomic DNA, 5 µL of 2× KASP Master Mix (Vazyme, Nanjing, Q113-03), 3.6 µL of deionized water, 0.1 µL each of the allele-specific primers SNP-F1 and SNP-F2, and 0.2 µL of the common reverse primer SNP-R. The amplification protocol consisted of an initial denaturation at 95 °C for 5 min; followed by 40 cycles of 95 °C for 10 s and 60 °C for 30 s; and a final extension at 72 °C for 2 min. The KASP assays were performed on a real-time PCR instrument. Fluorescence signals were monitored during the annealing/extension step of each PCR cycle. After amplification, genotype calling was automatically performed by the instrument’s software based on the kinetic fluorescence data. Results Phenotypic analysis of LA Phenotypic analysis revealed that LA varied between 4.98° and 76.23° across environments (Table 1; Table S1). The coefficients of variation were 49.03% in the 23FY environment and 46.19% in 24FY (Table 1), indicating substantial phenotypic variation among the 212 maize inbred lines. The skewness and kurtosis of LA distribution were 1.18 and 2.19 in 23FY, and 1.44 and 3.25 in 24FY, respectively (Table 1). Broad-sense heritability (H²) of LA reached 82.7% (Table 1), indicating that genetic factors predominantly control this trait. Correlation analysis showed a highly significant positive relationship (r = 0.71, P < 0.001) between LA measurements in 2023 (LA23) and 2024 (LA24) (Fig. 1). Moreover, LA values in both environments were strongly correlated with their best linear unbiased estimates (BLUEs; r = 0.91 for 23FY and r = 0.93 for 24FY, P < 0.001; Fig. 1). Table 1Phenotypic variation of leaf angle (LA) in 212 inbred lines of maizeTraitsenvironmentMean ± SEMRangeSkewnessKurtosisCV (%)H2(%)VarianceLA23FY20.36 ± 9.984.98–60.931.182.1949.0382.799.5924FY23.67 ± 10.936.00–76.231.443.2546.19119.50 Phenotypic variation of leaf angle (LA) in 212 inbred lines of maize Correlation analyses between leaf angle phenotypes of the 212 maize inbred lines. *** indicates P < 0.001 Genome-wide association analysis Genome-wide association analysis of LA was conducted using five models in the GAPIT package: GLM, MLM, SUPER, FarmCPU, and BLINK. A significance threshold of p < 8.55 × 10⁻⁷ (–log₁₀(p) > 6.07) was applied. The Q-Q plots indicated varying degrees of deviation among the models. The genomic inflation factors (λ) for the five models were as follows: GLM (2.0151), MLM (1.0176), SUPER (1.8006), FarmCPU (1.0472), and BLINK (1.1589). the genomic inflation factors for MLM, FarmCPU, and BLINK were within a marginally acceptable range, demonstrating improved control of false positives and population structure. Fourteen significant SNPs associated with LA were identified by four models (GLM, SUPER, FarmCPU, and BLINK), distributed across seven chromosomes (excluding 7, 9, and 10) (Fig. 2, Table S2). Among the detected SNPs, three were identified by two models and two by three models; These loci were considered stable due to their reproducible detection across models, and the well-behaved Q-Q plots observed in at least one model, without evident inflation or excessive conservativeness. Notably, the locus AX-86,305,869 on chromosome 3 was consistently identified by GLM, FarmCPU, and BLINK (Fig. 2; Table 2), with P-values of 7.08 × 10⁻⁸, 6.61 × 10⁻⁷, and 7.47 × 10⁻⁷, respectively, and exhibited relatively high phenotypic variance explained (PVE) across all three models (Table 2). Another key SNP, AX-107,953,621 on chromosome 5, was also detected by the same three models, with P-values of 3.50 × 10⁻⁸, 1.07 × 10⁻¹⁴, and 1.18 × 10⁻⁸. Additionally, three stable SNP—AX-107,936,049 (chromosome 1), AX-108,102,817 (chromosome 2), and AX-107,994,145 (chromosome 4) were co-detected by GLM and FarmCPU (Table 2; Table S2). Two SNP loci on chromosome 6, AX-108,061,477 (detected by the SUPER model) and AX-108,014,790 (detected by the FarmCPU model), were located within ± 200 kb of each other (Table 2; Table S2). Given their colocalization within the ± 200 kb interval, AX-108,061,477 was also designated as a stable SNP locus in subsequent analyses. Manhattan plots (left) and quantile–quantile (QQ) plots (right) for leaf angle. The Manhattan plots display SNP associations across chromosomes (color-coded by chromosome). The horizontal green line indicates the genome-wide significance threshold (− log₁₀(P) = 6.07, corresponding to P < 8.55 × 10⁻⁷). The QQ plots show the observed versus expected − log₁₀(P) distributions; the red line represents the expected line under the null hypothesis Table 2Identified stable SNP loci and predicted candidate genesTraitMarker nameChr.Position(bp)ModelP-valuePVE(%)Candidate geneGene AnnotationLAAX-107,936,04917652821GLM7.43E-082.19 GRMZM2G059428 NAC-transcription factor 53FarmCPU6.95E-071.65E-07AX-108,102,817249740954GLM6.76E-071.57 GRMZM2G003565 Regulator of chromosome condensation2FarmCPU9.31E-101.82 AC186329.3_FG009 Tetratricopeptide repeat (TPR)-like superfamily protein GRMZM2G436187 Transducin family protein/WD-40 repeat family proteinAX-86,305,8693162842787BLINK7.47E-0749.77 GRMZM2G133173 Ribophorin II (RPN2) family proteinFarmCPU6.61E-0724.93 GRMZM2G133398 VP1, dormancy regulatory proteinGLM7.08E-0812.86AX-107,994,1454163227890FarmCPU5.14E-074.84 GRMZM2G175983 Auxin efflux carrier family proteinGLM1.78E-076.28 GRMZM2G036905 Component of the RNA polymerase II elongation complex GRMZM2G092085 Transducin/WD40 repeat-like superfamily proteinAX-107,953,621520828657BLINK1.18E-0813.37 GRMZM2G011156 Cytochrome P450 36FarmCPU1.07E-140.00GLM3.50E-087.35AX-108,061,477/AX-108,014,7906146129498/146225884SUPER/FarmCPU4.21 E-07/7.14 E-124.71/0.00 GRMZM2G173630 Encodes a gibberellin (GA) receptor, has GA-binding activity, showing higher affinity to GA4.Chr. Chromosome, PVE Phenotypic variance explained Identified stable SNP loci and predicted candidate genes 4.21 E-07/ 7.14 E-12 4.71/ 0.00 Chr. Chromosome, PVE Phenotypic variance explained Candidate gene analysis Genome-wide association analysis identified 14 significant SNPs associated with maize LA. Within ± 200 kb regions of these SNPs, 342 potential candidate genes were identified, including 134 with functional annotations (Table S2). GO and KEGG enrichment analyses were performed to elucidate the biological functions of these candidates. A total of 83 annotated genes were enriched in 110 GO terms, including “membrane part,” “integral component of membrane,” and “transport” (Fig. 3; Table S3). Additionally, 23 genes were enriched in 28 KEGG pathways, such as sulfur metabolism, glycosphingolipid biosynthesis-ganglio series, and various types of N-glycan biosynthesis (Fig. 3; Table S4). These results provide a basis for the prioritization of candidate genes putatively associated with LA development in maize. GO and KEGG enrichment analysis of candidate genes. Enrichment results of candidate genes in Gene Ontology (GO) terms (biological process, BP; cellular component, CC; molecular function, MF) and KEGG pathways are shown. The x-axis indicates the number of enriched genes, and the y-axis lists the GO terms or KEGG pathways. Dot size corresponds to the gene count, and color intensity represents the enrichment significance level (-log₁₀(P-value)). KEGG terms marked with an asterisk (*) correspond to “Ganglioside biosynthesis - ganglio series”, and those marked with two asterisks (**) correspond to “Ganglioside biosynthesis - globo and isoglobo series” By integrating GO and KEGG enrichment results with maize (MaizeGDB and NCBI) and Arabidopsis ortholog annotations, we prioritized 17 high-confidence candidate genes potentially involved in regulating LA from the initial 134 annotated genes (Table S2). Among these, 11 candidate genes located within the confidence intervals of the six stable SNP loci were prioritized. Representative genes including: GRMZM2G059428 (NAC transcription factor), GRMZM2G003565 (regulator of chromosome condensation), AC186329.3_FG009 (tetratricopeptide repeat-like superfamily protein), GRMZM2G436187 (transducin family protein), GRMZM2G133173 (ribophorin II), GRMZM2G133398 (dormancy-associated regulatory protein), GRMZM2G175983 (auxin efflux carrier), GRMZM2G036905 (RNA polymerase II elongation complex component), GRMZM2G092085 (transducin/WD40 repeat protein), GRMZM2G011156 (cytochrome P450) and GRMZM2G173630 (Gibberellin-insensitive dwarf protein homolog) (Table 2). To further assess the association of these 17 candidate genes, identified through enrichment analyses and homology-based inference, with LA variation, we analyzed their spatiotemporal expression patterns using publicly available RNA-seq data (Table S5). The results revealed distinct expression profiles across multiple tissues—including coleoptiles, leaf blades, leaf tips, leaf bases, and shoot apical meristems—at various developmental stages (Fig. 4; Table S5). For instance, GRMZM2G133398 showed markedly higher expression in the coleoptile compared to other tissues. Six genes (GRMZM2G042250, GRMZM2G077356, GRMZM2G023858, GRMZM2G092085, GRMZM2G133173, and GRMZM2G077389) were highly expressed in coleoptiles, leaf bases, stem tips, and shoot apical meristems, suggesting roles in leaf growth and development. In contrast, GRMZM2G175983, GRMZM2G173630, GRMZM2G059428, GRMZM2G036905, GRMZM2G003565, and GRMZM2G083759 were more highly expressed in leaves and leaf tips (Fig. 4). No expression was detected for AC186329.3_FG009 or GRMZM2G436187 in any profiled tissues, while GRMZM2G011156 and GRMZM2G151205 were not expressed in the tissues included in Fig. 4. Expression patterns of candidate genes in different tissues of maize B73. Heatmap showing the expression levels of candidate genes across various developmental stages and tissues in the maize reference inbred line B73. The x-axis represents tissue types, and the y-axis lists candidate genes. The color gradient (blue to red) corresponds to expression levels from low to high Protein–protein interaction network analysis of candidate genes To provide a broader context for functional inference by examining interactions between proteins encoded by the candidate genes and other proteins, we constructed a protein–protein interaction (PPI) network using the STRING database (v12.0) with a medium-confidence interaction score (> 0.4) and a maximum of 10 interactors per node to ensure interpretability. Protein sequences for AC186329.3_FG009 and GRMZM2G436187 were unavailable in MaizeGDB, while GRMZM2G175983, GRMZM2G036905, and GRMZM2G077389 showed limited homology to annotated proteins in STRING, preventing reliable interaction predictions for these five genes. The resulting PPI network indicated that 12 candidate proteins function as network hubs (Fig. 5), forming distinct functional modules potentially involved in key regulatory processes during maize growth and development. Notably, RLD2 (encoded by GRMZM2G042250) and A0A1D6Q981 (encoded by GRMZM2G092085) both interacted with DCP2 (encoded by GRMZM2G051756), an mRNA decapping enzyme whose Arabidopsis ortholog mutants exhibit defective venation, impaired leaf development, and premature abscission [46, 47]. Proteins encoded by GRMZM2G059428 (NAC transcription factor), GRMZM2G133173 (ribophorin II), and GRMZM2G023858 formed a highly interconnected hub with multiple partners, suggesting roles in coordinating complex developmental processes. In contrast, GRMZM2G011156 and GRMZM2G083759 showed minimal connectivity, each interacting with only two partners, which may reflect specialized functions in specific metabolic pathways (Fig. 5). These differential interaction patterns highlight varying levels of functional integration among candidate genes within the maize developmental regulatory network. Protein–protein interaction network of candidate gene-encoded proteins. Nodes represent proteins, with those encoded by candidate genes highlighted in red. Black edges indicate known or predicted functional interactions between proteins Analysis of allelic variation effects on LA traits To evaluate the functional effects of allelic variation on LA regulation, we conducted a comprehensive analysis of allelic effects at stable SNP loci. As shown in Fig. 6, allelic variations at all six SNP loci showed highly significant associations (P < 0.01) with LA phenotypes in three independent environments: LA23, LA24, and LABLUE. At locus AX-107,936,049, inbred lines carrying the G/G allele showed average LA reductions of 13.00°, 13.83°, and 13.23° across the two environments and BLUE compared to those with the A/A allele, indicating that G/G represents a favorable allele. Similarly, for AX-108,102,817, the T/T allele reduced average LA by 10.09°–12.55° relative to the C/C allele. At AX-107,953,621, the C/C allele decreased average LA by 7.39°, 11.87°, and 9.75° compared to the A/A allele. For AX-107,994,145, the G/G allele reduced average LA by 11.97°–13.76° relative to the A/A allele. At AX-86,305,869, the T/T allele resulted in reductions of 15.01°–18.18° compared to the C/C allele. Finally, at AX-108,061,477, the C/C allele reduced LA by 4.78°–5.06° relative to the A/A allele (Fig. 6). These consistent reductions across environments confirm that the identified alleles represent favorable genetic variants for LA improvement. Phenotypic effect of SNP genotypes on leaf angle. Box plots show the distribution of leaf angle values across genotype classes. Blue boxes represent the superior (favorable) genotype at the locus, while red boxes represent the non-superior (unfavorable) genotype. Significance levels: *P < 0.05, **P < 0.01, ****P < 0.0001 In this study, the distribution of the six stable SNP loci was further analyzed using 212 inbred lines. The materials were classified into six categories based on the number of favorable alleles carried, ranging from one to six. The corresponding LA phenotypes for the six groups were 52.66 ± 9.17°, 41.77 ± 2.73°, 33.38 ± 2.54°, 26.24 ± 1.42°, 20.17 ± 0.72°, and 17.82 ± 0.66°, respectively (Fig. 7). Notably, materials carrying all six favorable alleles exhibited the smallest LA. Therefore, these six stable SNP loci can be effectively utilized for rapid screening of germplasm with minimal LA, demonstrating promising application potential in maize breeding. It should be noted that Group 1 (carrying 1 favorable alleles) included only three individuals, resulting in a limited sample size. Consequently, the statistical robustness of this observation is constrained, and further validation in larger, independent populations is required (Fig. 7). Phenotypic analysis of leaf angle based on the accumulation of favorable alleles at six stable SNP loci. The 212 maize inbred lines were categorized into six groups according to the number of favorable alleles (ranging from 1 to 6) carried at the six stable SNP loci. Leaf angle phenotypes (BLUE values) are compared among the groups. Groups sharing the same letter are not significantly different (Tukey’s test, P > 0.05); numbers in parentheses indicate the sample size (number of materials) per group Development and application of KASP markers Based on the six stable SNP loci, this study designed corresponding KASP primers. Effective KASP markers were successfully developed for the stable loci AX-86,305,869, AX-107,953,621, and AX-108,061,477 located on chromosomes 3, 5, and 6, respectively (Fig. 8A-C; Table S6). Using these markers, genotyping was performed on genomic DNA from 52 maize varieties/lines (including Zhengdan 958, MY73, Anke 985, etc.) and 20 inbred lines (Table S7). The three KASP primer pairs successfully genotyped all tested materials, and the resulting genotypes for the 20 inbred lines were highly consistent with those obtained from the SNP array. Further analysis of variance confirmed significant differences in LA phenotypes among genotype groups defined by these three KASP markers (Fig. 8A-C, Table S7). Validation and combined analysis of three KASP markers. A-C Genotyping results of KASP markers converted from three stable SNP loci (AX-86305869, AX-107953621, and AX-108061477). D Comparison of leaf angle phenotypes among haplotype groups defined by allele combinations at the three loci Furthermore, combined analysis using these three KASP markers classified all 72 materials into 13 distinct haplotypes, five of which were unique types represented by only a single material (Table S7). Notably, 13 germplasm accessions carrying the haplotype combination CC/CC/AA (corresponding to loci AX-86305869, AX-107953621, and AX-108061477, respectively) exhibited the smallest mean LA (12.46 ± 1.86°) (Fig. 8D; Table S7). The allele types within this combination were confirmed by prior allelic effect analysis to be superior variations favoring reduced LA. In summary, the stable SNP loci identified on chromosomes 3, 5, and 6 in this study represent superior allelic variations that can significantly reduce LA in maize. Their corresponding KASP markers provide an effective tool for marker-assisted selection in breeding density-tolerant maize varieties. Discussion Phenotypic analysis of maize LA In high-yield breeding, increasing planting density to enhance population yield generally holds greater potential than solely improving individual plant productivity [7, 9]. Therefore, developing ideal plant architecture adapted to high-density planting is a key direction in current maize breeding. Studies have shown that a smaller LA promotes more upright leaves, which reduces mutual shading within the canopy, improves light interception efficiency, and enhances ventilation, light penetration, and lodging resistance [7, 9]. Thus, elucidating the genetic basis of LA is of significant importance for ideal plant architecture breeding. In this study, the LA of three consecutive leaves above the ear was phenotypically evaluated in 212 maize inbred lines. The results showed that this trait was significantly and positively correlated between the two growing seasons, indicating good consistency across environments (Fig. 1). The phenotypic distribution showed a slight high-value tail, indicating limited deviation from normality (Fig. 1); however, mixed linear models are robust to such deviations, and the data were considered suitable for GWAS [48, 49]. The broad-sense heritability of leaf angle was above 80%, aligning with previous reports [50], further confirming that this trait is predominantly controlled by genetic factors and is suitable for genetic dissection and marker-assisted breeding. Comparative analysis of GWAS mapping results Different GWAS models may vary in their power to detect association signals due to distinct strategies for correcting population structure, kinship, and linkage disequilibrium [51, 52]. GLM tends to detect more associations but is prone to false positives, whereas MLM provides stricter control at the cost of reduced power. SUPER, FarmCPU, and BLINK achieve a better balance between detection power and false-positive control, with FarmCPU and BLINK generally showing more stable performance [40]. Therefore, in this study, five models were applied for cross-validation to improve the reliability of GWAS results. In this study, association analysis using 212 maize inbred lines identified 14 SNPs significantly associated with LA. Among them, five SNPs were consistently detected by multiple models. The corresponding Q-Q plots showed no evidence of inflation or over-conservativeness, supporting their classification as stable loci. (Table 2; Table S2), located on chromosomes 1, 2, 3, 4, and 5 respectively. AX-108,061,477 and AX-108,014,790 co-localized within a ± 200 kb interval on chromosome 6 (Table 2; Table S2). Accordingly, AX-108,061,477 was also classified as a stable SNP locus in this study. The fact that these loci were consistently detected across multiple models, indicating a robust association with the LA trait and providing mutual support across different analytical approaches. Both QTL mapping and GWAS are mainstream methods for dissecting the genetic architecture of complex traits. They both rely on statistical inference of phenotype-genotype associations to locate relevant loci, and their results can be mutually validated to enhance reliability [9, 53, 54]. Among the six stable SNPs identified here, four represent novel loci, while the remaining two show genomic colocalization or proximity (physical distance < 2 Mb) to previously reported QTLs or SNPs. For example, AX-86,305,869 on chromosome 3, detected concurrently by three models, is located at 162,842,787 bp, within 2 Mb of the SNP loci PZE03164583838 and PZE03164630089 reported by Tian et al. [9]. Furthermore, a recent study by Yao et al. reported a leaf-angle-associated locus on chromosome 3 at 163,726,634 bp, which is 883,847 bp away from AX-86,305,869 (i.e., < 1 Mb) [32]. AX-107,994,145 on chromosome 4 (163,227,890 bp), co-detected by the GLM and FarmCPU models, falls within the QTL interval qLA04-02 (135.34–181.26 Mb) reported by Zhang and Huang [55]. The consistency of these two loci with prior studies further strengthens the credibility of their associations and provides a theoretical basis for molecular marker breeding. The remaining four stable SNPs, although not directly reported in existing studies, showed robust association signals across multiple models, and candidate genes related to leaf development were identified within their respective ± 200 kb regions (Table S2). Such discrepancies may stem from differences in the genetic materials, environmental conditions, or analytical pipelines employed [56]. Nonetheless, these loci remain valuable references, offering new leads for subsequent functional validation and breeding applications (Table S2). Functional analysis of candidate genes and their physiological roles in hormone-mediated leaf development The regulation of maize LA may involve multiple phytohormonal signaling pathways that potentially regulate cell division, expansion, and differentiation within the leaf collar and midrib regions [7, 24, 57]. From the 134 functionally annotated candidate genes in this study, 17 key genes potentially regulating LA were further screened, 11 of which are located within the ± 200 kb confidence intervals of the six stable SNP loci. These genes are likely involved in regulatory processes integrating auxin (IAA), brassinosteroid (BR), gibberellin (GA), and abscisic acid (ABA) signaling, and may contribute to the modulation of leaf architecture [30, 58]. The gene GRMZM2G059428, located within the confidence interval of locus AX-107,936,049, encodes an NAC family transcription factor. Its Arabidopsis homolog ANAC036 is highly expressed in leaves, and its overexpression leads to shortened leaf blades and petioles [59]. NAC transcription factors have been increasingly recognized as integrators of hormonal signals during leaf morphogenesis [58]. These observations suggest that GRMZM2G059428 may be associated with LA, potentially through effects on adaxial–abaxial development or cell elongation. The AX-108,102,817 locus harbors several hormonally related candidate genes. GRMZM2G003565 may encode a regulator of chromosome condensation and could be a direct ortholog of the Arabidopsis UV-B receptor UVR8. UV-B perception via UVR8 triggers signaling that can result in shortened petioles and reduced leaf area, leading to a compact plant architecture [60, 61]. Notably, Yang et al. pinpointed the locus arl1, which affects maize leaf abaxial rolling, to Zm00001eb082500 (i.e., GRMZM2G003565) within this region, further suggesting a possible role in leaf morphogenesis [62]. AC186329.3_FG009, located in the same region, is a homolog of the Arabidopsis gene TPR2. TPR2 has been reported to be involved in axillary and leaf cell development through repression of the branching suppressor BRANCHED1 [63]. The Arabidopsis homolog of GRMZM2G436187 is TPL, which interacts with BR signaling transcription factors BES1/BZR1, inhibiting the expression of the boundary gene CUC3, thereby potentially affecting organ boundary formation [64]. Concurrently, TPL-related proteins have also been implicated in the IAA signaling pathway [65], pointing to a possible link between this locus and hormone crosstalk. At locus AX-86,305,869 on chromosome 3, two candidate genes are potentially related to hormone-associated processes. The Arabidopsis ortholog of GRMZM2G133173, HAP6, encodes a ribophorin implicated in ABA signaling [66], while GRMZM2G133398 encodes a transcription regulator that simultaneously enhances ABA signaling and negatively regulates GA signaling [67, 68]. These observations indicate that this locus may be associated with hormone-mediated regulation of LA variation, with the balance between GA and ABA signaling potentially contributing to this process, consistent with established mechanisms controlling organ growth [58]. At locus AX-107,994,145, the candidate gene GRMZM2G175983 encodes a membrane transporter potentially involved in intracellular IAA homeostasis. Its Arabidopsis homolog PIN5 influences leaf vein patterning and cell wall deposition by regulating auxin distribution [69] and interacts with GA signaling to regulate vascular development [70]. This highlights the role of auxin transport in establishing vascular architecture that may contribute to the mechanical support of LA. GRMZM2G036905 in the same region encodes a component of the RNA polymerase II elongation complex. Mutants of its Arabidopsis homolog ELF1 exhibit increased leaf erectness under specific genetic backgrounds [71], implying a potential role in coordinating transcriptional elongation with leaf inclination regulation. Another gene, GRMZM2G092085, is involved in mRNA decapping. Mutants of its Arabidopsis homolog VCS display leaf developmental defects and abnormal venation patterns [72], supporting its possible involvement in the post-transcriptional regulation of leaf morphogenesis. The candidate gene GRMZM2G011156 at locus AX-107,953,621 on chromosome 5 encodes a cytochrome P450 protein. Its Arabidopsis homolog participates in the metabolism of IAA biosynthetic precursors [73, 74], which may link it to the auxin biosynthesis pathway, a key regulator of leaf growth and orientation. The two SNPs on chromosome 6, AX-108,061,477 and AX-108,014,790, are physically adjacent (96.386 kb apart). Given their tight linkage, this study combined them for analysis as a single genetic locus. The candidate gene GRMZM2G173630 within this interval encodes a gibberellin receptor protein (belonging to the GID1C family) with high-affinity binding activity for bioactive GAs (e.g., GA4). Transcriptome evidence suggests that its expression responds to drought stress, accompanied by changes in DELLA-related genes [75]. In Arabidopsis, the mechanism of GID1 involves, upon GA binding, mediating the degradation of DELLA repressor proteins via the 26 S proteasome pathway. This relieves DELLA’s transcriptional repression of downstream growth-related genes, ultimately affecting plant development by regulating processes like cell elongation [76]. Given the role of GA in regulating cell expansion, this pathway may be indirectly associated with variation in LA, although the specific function of GRMZM2G173630 in maize requires further validation. Nevertheless, it should be noted that the predicted functions of these genes are primarily inferred from their orthologs and annotations in MaizeGDB and NCBI. As such, their precise biological roles in maize remain unconfirmed, and the mechanisms by which they influence LA are not fully understood. Further experimental validation will be necessary to substantiate these predictions and elucidate their contributions to this trait. Breeding implications and limitations of KASP marker development Several considerations should be noted regarding the breeding application of the identified markers. First, the stable SNPs were detected in a specific set of southeastern China breeding lines; their effects may vary in other genetic backgrounds or environments, and validation in broader germplasm is recommended. Second, some previously reported QTL were not detected in this study, possibly due to fixation of those loci in our panel or differences in linkage disequilibrium patterns. Third, while all six stable SNPs are potentially useful, we attempted to develop KASP markers for all six loci. However, only three loci located on chromosomes 3, 5, and 6 (AX-86305869, AX-107953621, and AX-108061477) successfully yielded KASP markers capable of effectively discriminating among the materials. The overall conversion success rate was 50%, which is lower than the previously reported conversion rate of KASP markers in maize (approximately 71%) [77]. For the unsuccessful loci, the failure can be primarily attributed to the presence of paralogous sequences flanking the target SNPs, leading to a lack of locus specificity in primer design. It is important to note that all six loci were subjected to KASP assay development attempts, rather than a subjective selection of a subset; the final availability of only three markers resulted naturally from technical feasibility screening. Finally, although LA has high heritability (82.7%), marker‑assisted selection offers complementary advantages over phenotypic selection alone, including early‑generation screening, reduced environmental influence, and accelerated line development. Therefore, the validated KASP markers provide practical tools for improving plant architecture in high‑density maize breeding, while the non‑converted SNPs remain targets for future marker optimization or functional validation. In summary, the candidate genes screened in this study are extensively involved in multiple key biological processes, including the integration of plant hormone signaling (IAA, BR, GA, ABA), transcriptional regulation, cell morphogenesis, and leaf vein development, thereby constituting a complex regulatory network for LA. These results lay a solid theoretical foundation for subsequent gene functional validation and the development of efficient molecular markers. Conclusion In this study, a panel of 212 maize inbred lines was used for GWAS with 58,455 SNP markers through five models (GLM, MLM, SUPER, FarmCPU, and BLINK) implemented in the GAPIT package. A total of 14 SNPs significantly associated with LA were identified, and 342 candidate genes were screened within the 200 kb flanking regions of these SNPs, among which 134 were functionally annotated. Further analysis revealed six stable SNP loci detected by multiple models, including four novel loci reported for the first time. Within the confidence intervals of these six stable loci, 11 functionally annotated candidate genes potentially involved in the variation of LA were identified. For instance, the candidate gene GRMZM2G036905 has an Arabidopsis homolog, ELF1, whose double mutant (elf1-1/iws1-1) exhibits a more upright leaf phenotype; The candidate gene GRMZM2G003565 has been reported to be related to maize leaf rolling. Additionally, KASP markers were successfully developed for three of the stable SNP loci and validated for their association with LA phenotypes in 52 varieties/lines (including Zhengdan 958, MY73, and Anke 985) and 20 inbred lines. These findings provide new loci and candidate genes for elucidating the genetic basis of LA in maize, laying an important foundation for functional validation of related genes, development of molecular markers, and breeding of density-tolerant maize varieties. It is worth noting that, although stable SNPs associated with LA in maize were identified and corresponding candidate genes were prioritized in this study, the findings are constrained by the specific breeding population analyzed. To fully elucidate the genetic architecture underlying LA, it will be necessary to conduct comparative analyses across diverse germplasm resources, including elite breeding lines, landraces, and wild relatives, and to pursue more comprehensive and in-depth investigations. Supplementary Information Supplementary Material 1 Authors’ contributions C.L. and Y.H. conceived and designed the study. L.K. conducted the experiment and wrote the manuscript. W.C. and Y.H. provided the experimental materials. X.J., C.J., H.W. (Wangfei He), S.R., B.R., C.Y., Y.H. and H.Y. performed the LA investigations. L.K. and H.W. (Weidong Huang) analyzed the data, C.X. and C.M. supervised the experiment, C.L. reviewed and modified the manuscript. All authors have read and agreed to the published version of the manuscript. Funding This work was supported by the Anhui Provincial Major Science and Technology Project (2022e03020008), the Key Discipline Construction Funds for Crop Science of Anhui Sciences and Technology University (No. XK-XJGF001), Science and Technology Innovation Groups of Anhui Science and Technology University (2023KJCXTD002), the Project No.2025SWYZ0200 supported by Bio-breeding Laboratory of Anhui province, the Talent Introduction Project of Anhui Science and Technology University (NXYJ201901), the research and development fund of Anhui Science and Technology University (FZ230121), the Chinese College Student Innovation Fund Project (Genome-wide Association Analysis of Maize Plant Architecture-Related Traits and Candidate Gene Identification). Data availability Data supporting the findings of this study are available within the manuscript and its Supporting Information files. Declarations Ethics approval and consent to participate All the methods involving plants and their materials complied with relevant institutional, national, and international guidelines and legislation. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Lei Chen, Email: chenlei@ahstu.edu.cn. Haibing Yu, Email: yuhb@ahstu.edu.cn. References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement Data supporting the findings of this study are available within the manuscript and its Supporting Information files.