Genome-wide association study reveals stable loci and candidate genes for cottonseed protein content in upland cotton (Gossypium hirsutum L.)

Abstract

Cottonseed protein is a valuable yet underexploited resource. However, efforts to genetically improve this trait have been hindered by the scarcity of genetic loci that are reproducible across diverse environments. In this study, we evaluated the crude protein content in a natural population consisting of 259 upland cotton (Gossypium hirsutum L.) accessions across seven environments in Xinjiang during the 2023–2024 growing seasons and performed a genome-wide association study (GWAS) using 1,144,681 single-nucleotide polymorphism (SNP) markers derived from whole-genome resequencing. The results revealed a wide range of variation in protein content (27.00%–50.76%), a relatively high heritability estimate (H² = 0.68), and significant genotype-by-environment interaction effects, collectively underscoring the necessity of multi-environment association mapping. Using 12 phenotypic datasets comprising single-environment measurements and BLUP-derived integrated phenotypes, a mixed linear model identified 211 significant SNP–trait associations corresponding to 195 unique SNPs after removing recurrent detections across datasets. Among these, 11 SNPs were reproducibly detected in at least two datasets and were further consolidated into nine candidate association intervals distributed across six chromosomes. Haplotype analysis of the LD block surrounding D11:19,580,385, combined with functional annotation, enabled the prioritization of 13 putative candidate genes with annotation-level roles in lysine biosynthesis, protein translation, vesicle-mediated transport, ubiquitin-related regulation, and protein interaction processes. These genes warrant further functional validation in future studies. Collectively, the findings of this study provide recurrent association loci and a practical foundation for accelerating marker-assisted selection breeding and functional validation of genes underlying the cottonseed protein content.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00122-026-05261-2.

Introduction

Cotton (Gossypium spp.) is cultivated extensively worldwide, primarily as the leading source of natural fibers for the textile industry. Beyond fiber production, cottonseed is a rich source of high-quality protein and provides an important nutritional supplement for human consumption and livestock feed (Cheng et al. 2020; Kumar et al. 2021b). Cottonseed protein is enriched in basic amino acids associated with diverse biological activities, exhibits favorable palatability, and constitutes a high-yield and high-quality plant protein. Accordingly, it is increasingly regarded as a promising candidate for alternative protein development with potential applications in the food industry (Tan et al. 2022). The content and composition of cottonseed protein directly determine its nutritional value and physicochemical properties (Ma et al. 2018). Therefore, improving seed nutrient profiles to enhance cotton productivity and enable the broader utilization of cottonseed protein has become an important objective in modern breeding programs.

Cottonseed protein biosynthesis relies on amino acids as precursors, with glutamate serving as the central metabolic node within this network (Forde and Lea 2007). Glutamate is converted into multiple amino acids, including aspartate and alanine, through reactions catalyzed by aspartate aminotransferase (AST) and alanine aminotransferase (ALT) (Torre et al. 2014; Xu et al. 2017). Glutamate dehydrogenase (GDH) mediates the reversible interconversion of glutamate and α-ketoglutarate, thereby coupling carbon and nitrogen metabolism. Seed nitrogen is largely derived from the reductive assimilation of nitrates. Nitrate reductase (NR) and nitrite reductase (NiR) sequentially reduce nitrate to ammonium (NH₄⁺) (Masclaux-Daubresse et al. 2010). Ammonium is subsequently assimilated into amino acids through the glutamine synthetase (GS) and glutamate synthase (GOGAT) cycles, and the activities of these enzymes are widely used as indicators of the intensity of protein synthesis (Liang et al. 2011). Evidence further indicates that the GhGS gene participates in regulating seed embryo development (Yajun et al. 2008), whereas the high expression of GhASN is closely associated with free amino acid accumulation and protein synthesis (Iqbal et al. 2022) (Fig. 1). Genome-wide association studies (GWAS) are a powerful approach for identifying associations between genetic variations and complex traits. Compared with traditional quantitative trait locus (QTL) mapping, GWAS offers a higher resolution for localizing trait-associated loci (Wei et al. 2015). In cotton breeding, cottonseed protein content and composition are key determinants of nutritional quality, and GWAS has been increasingly applied to dissect the genetic architecture of these traits and inform improvement strategies. Substantial progress has been made in elucidating the genetic basis of cottonseed protein content. Multiple studies have used diverse genetic populations and high-throughput marker platforms to identify numerous associated QTLs in various species. Hu et al. (Hu et al. 2022) analyzed 316 upland cotton accessions with more than 1.87 million single nucleotide polymorphism (SNP) markers and identified 27 QTLs associated with cottonseed protein content on chromosomes A01, A03, A05, A09, A11, D01, D03, D07, and D08. Du et al. (Du et al. 2018) conducted GWAS in a germplasm population of comparable size and detected 21 protein-related QTLs, with most loci located on chromosome D12. Using a recombinant inbred line (RIL) population of 196 families, Gong et al. (Gong et al. 2022) integrated 8,186 SNP markers derived from simple sequence repeats (SSR), microarrays, and specific-locus amplified fragment sequencing (SLAF-seq) and identified 44 QTLs distributed across all chromosomes that explained 4.87–13.02% of the phenotypic variation. Chromosome segment substitution lines (CSSLs) developed from wild cotton germplasm provide an effective platform for identifying superior alleles. Yue et al. (Yue 2022) used 552 SSR markers to map 19 protein content QTLs in 553 CSSLs derived from G. darwinii and G. hirsutum. Cui et al. (Cui 2022) identified 31 QTLs in 559 CSSLs derived from G. tomentosum and G. hirsutum using 486 SSR markers. Liu et al. (Liu 2023) mapped 11 QTLs in a BC₃F₂ population of 564 lines derived from G. mustelinum and G. hirsutum using 150 SSR markers. In addition, Tang (Tang 2023) and Bao (Bao 2023) identified 46 and 20 protein-related QTLs, respectively, in populations of 186 RILs and 178 high-quality fiber RILs using high-density SNP markers. Several studies focusing on seed oil traits have provided relevant evidence regarding seed metabolism. Wang et al. (Wang et al. 2019) constructed a high-density map with 7,033 SLAF-SNP markers in a 180-family RIL population and identified 13 oil content-related QTLs. Liu et al. (Liu et al. 2020) mapped 8 QTLs for oil content in 376 F₂ populations and suggested potential links to protein metabolism. Zhang et al. (Zhang et al. 2022) analyzed QTLs associated with oil and fatty acid content in 188 RILs using 388 SSR markers. Collectively, these investigations have generated high-density genetic resources and underscored the complex genetic architecture governing cottonseed protein traits, thereby supporting marker-assisted selection (MAS) and gene cloning efforts.

Fig. 1.

Schematic representation of the determination workflow and biosynthetic pathway of cottonseed protein. Note: The key enzymes and metabolites depicted in this figure include nitrate reductase (NR) and nitrite reductase (NiR), which are involved in nitrate assimilation; glutamine synthetase (GS) and glutamate synthase (GOGAT), which constitute the primary GS/GOGAT cycle for ammonium (NH₄⁺) assimilation into glutamine (Gln) and glutamate (Glu); glutamate dehydrogenase (GDH), which is an additional node linking carbon and nitrogen metabolism; and aminotransferases, including aspartate aminotransferase (AST) and alanine aminotransferase (ALT), which mediate amino-group transfer between amino acids and their corresponding keto acids. Related genes, such as the Gossypium hirsutum glutamine synthetase gene (GhGS) and asparagine synthetase gene (GhASN), are associated with nitrogen assimilation/allocation during seed development and contribute to nitrogen storage/transport (e.g., via asparagine synthesis) and protein biosynthesis

Contemporary cotton production remains primarily oriented toward fiber utilization, whereas the value of cottonseed remains undeveloped. Cottonseed has broad application potential in the food and feed industries, and its high protein content represents a cost-effective and sustainable resource with substantial development potential. However, GWAS targeting cottonseed protein composition remains limited. To address this gap, the present study evaluated the crude protein content of cottonseed in a natural population of 259 upland cotton accessions across seven diverse environments. The phenotypic variation observed in this population was used to evaluate its potential for germplasm screening and genetic dissection of cottonseed protein contents. GWAS was further applied to identify key loci and candidate genes associated with cottonseed protein content. These results provide a theoretical basis for the high-value utilization of cottonseed resources and future molecular breeding.

Materials and methods

Plant materials

A panel of 259 upland cotton (Gossypium hirsutum L.) germplasm accessions was used (Supplementary Table S1-1). The population comprised 97 conventional cultivars from the Northwest Inland cotton region, 77 from the Yellow River Valley cotton region, 34 from the Yangtze River Valley cotton region, 20 landraces, and 31 foreign varieties. All plant materials were provided by the Xinjiang Crop Genetics and Breeding Laboratory.

Field experiment design

Field experiments were conducted across diverse ecological zones in Xinjiang during the 2023 and 2024 growing seasons. The distribution of the experimental sites is shown in Figure 2A. In 2023, trials were conducted at three locations in this region. These included Yuepuhu County in Kashgar Prefecture, Southern Xinjiang, designated as Site I and coded as 23YPH (39°15′N, 76°48′E); Wensu County in Aksu Prefecture, Southern Xinjiang, designated as Site II and coded as 23WS (41°18′N, 80°26′E); and the Liuhudi Experimental Base in Manas County, Changji Prefecture, Northern Xinjiang, designated as Site IV and coded as 23LHD (44°39′N, 86°08′E). In 2024, the experiment was expanded to four locations, including Yuepuhu County (Site I) coded as 24YPH and Wensu County (Site II) coded as 24WS in Southern Xinjiang. Trials in Northern Xinjiang were conducted at the Kuitun City Experimental Base in Ili Prefecture, designated as Site V and coded as 24KT (44°26′N, 84°56′E), and the Sanping Experimental Base of Xinjiang Agricultural University in Urumqi, designated as Site III and coded as 24SP (43°56′N, 87°21′E). At all sites, sowing was conducted from mid-April to late April, and harvesting was completed from mid-October to late October each year. A randomized complete block design (RCBD) with three replicates was used to evaluate phenotypic variations in cottonseed protein content across environments. For each plot and each replicate, 20 naturally open bolls were hand-harvested from the middle canopy of representative plants. Defoliants were applied prior to harvest, following local standard agronomic practices. The seeds were thoroughly mixed after ginning. Prior to protein content determination, only fully developed and plump seeds were screened and selected for analysis to ensure accurate measurements.

Fig. 2.

Distribution of experimental sites and standard curve for protein quantification. (A) Cotton experimental sites from 2023 to 2024. (B) Standard curve for protein quantification using the Biuret method

Determination of protein content

Cottonseed protein content was determined using the Biuret method (Mesa and Megerssa 2024) with minor modifications. Three biological replicates were measured for each sample.

Reagent preparation and standard curve

Biuret reagent was prepared by dissolving 0.375 g CuSO₄·5H₂O and 1.5 g potassium sodium tartrate in 125 mL distilled water, followed by the addition of 75 mL of 10% NaOH. The mixture was diluted to 250 mL and stored in a light-protected bottle, and any reagent exhibiting precipitation was discarded. A standard curve was generated using Bovine Serum Albumin (BSA) as the reference standard (Fig. 2B). Aliquots of standard BSA solution, ranging from 0.1 to 1.6 mL, were prepared from a stock solution containing 0.4 g BSA dissolved in 25 mL of 0.1 N NaOH. Each aliquot was mixed with 8 mL Biuret reagent and brought to a final volume of 10 mL using 0.1 N NaOH. After incubation at room temperature (20 °C) for 40 min, the absorbance was measured at 555 nm using a spectrophotometer.

Sample measurement

Cottonseed samples were milled and passed through an 80-mesh sieve. The air-dried powder (200–500 mg) was defatted with 5 mL diethyl ether, followed by extraction in 10 mL of 0.1 N NaOH for 40 min. The extract was centrifuged at 3000 rpm for 10 min. A 2 mL aliquot of the supernatant was reacted with Biuret reagent using the same procedure as that applied for the standard curve. Three biological replicates were used for each experiment. The protein content was calculated using the following equation:

$$\text{Protein Content}\left(\%\right)=\left(C\times V\times 10\right)/\left(a\times W\right)\times 100$$

In this equation, C represents the concentration of the standard solution in mg/mL, V represents the volume obtained from the standard curve in mL, a represents the supernatant volume in mL, and W represents the sample weight in mg.

Whole-Genome Re-sequencing

A total of 259 upland cotton germplasm accessions were included in this study. Genotypic data were obtained from whole-genome resequencing previously completed by our research group using the Illumina HiSeq 2500 platform. For sampling and library construction, one true leaf was collected from each accession with three biological replicates, placed in a centrifuge tube containing a desiccant, and used for the extraction of genomic DNA. The mean sequencing depth was approximately 10 × , and the reference genome coverage exceeded 90%. The upland cotton standard line TM-1 (CRI version) was used as a reference genome. Initial variant calling yielded 6,607,760 SNPs, which were subjected to stringent quality control using PLINK software. After removing loci with a minor allele frequency (MAF) below 0.05 and a missing rate at or above 0.05, 1,144,681 high-quality SNPs were retained for downstream analyses.

Population structure and linkage disequilibrium (LD) analysis

To characterize genome-wide phylogenetic relationships, a phylogenetic tree was constructed using the Neighbor-Joining (NJ) method implemented in TASSEL software (Bradbury et al. 2007). The population structure was inferred using ADMIXTURE with K values ranging from 2 to 9 and 10,000 iterations per run (Alexander and Lange 2011). Principal Component Analysis (PCA) was performed using GCTA to further evaluate population stratification (Yang et al. 2011). LD was estimated by calculating the pairwise LD coefficient r² among high-quality SNPs, and LD decay was assessed using PopLDdecay (Zhang et al. 2019).

Genome-wide association study (GWAS)

A GWAS for cottonseed protein content was conducted using 1,144,681 high-quality SNPs with the genome-wide efficient mixed model association package (GEMMA), version 0.94.1 (Zhou and Stephens 2012). A Mixed Linear Model (MLM) incorporating Principal Components (P) and a kinship (K) matrix as covariates (P + K model) was applied to account for population structure and relatedness. The GWAS significance threshold was determined based on the effective number of independent SNPs. Because SNPs across the genome are not completely independent due to linkage disequilibrium (LD), the conventional Bonferroni correction (0.05/1,144,681 = 4.37 × 10^–8) was overly conservative and yielded very few significant associations (Wen et al. 2025). To account for this, LD pruning was conducted in PLINK using –indep-pairwise 50 10 0.01, resulting in 21,062 approximately independent SNPs (Kanai et al. 2016). Following the P = 1/N threshold strategy commonly used in cotton GWAS studies (Sun et al. 2017), the significance threshold was set to P = 1/21,062 = 4.75 × 10^–5, corresponding to -log10(P) = 4.32.

GO and KEGG enrichment of candidate genes

Functional enrichment analysis of candidate genes was conducted using the annotation resources of the Gossypium hirsutum TM-1 (CRI v1) reference genome ( https://www.cottongen.org/node/13354433). The 93 genes harboring nonsynonymous variants within recurrent GWAS intervals were designated as the input set and subsequently mapped against the TM-1 background gene set with available Gene Ontology (GO) or Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway annotations. After removing duplicate annotation records, GO and KEGG enrichment analyses were performed independently. The statistical significance of enrichment was determined using a hypergeometric test, and multiple testing correction was applied using the Benjamini–Hochberg procedure to control the false discovery rate (FDR). A nominal threshold of P < 0.05 was adopted for unadjusted significance, while an FDR-adjusted threshold of FDR < 0.05 was considered to be statistically robust. GO terms were classified according to three primary ontologies: Biological Process (BP), Cellular Component (CC), and Molecular Function (MF). KEGG pathways were stratified based on their first‑level functional hierarchy. All enrichment outputs were visualized using the ggplot2 package in the R statistical computing environment. In the graphical representation, "Gene Number" corresponds to the count of input candidate genes assigned to a specific term, and "Rich Factor" represents the ratio of these assigned genes to the total number of annotated background genes for that term.

Haplotype analysis and candidate gene prediction

Significant SNPs located within the 500-kb LD decay window were merged into candidate association intervals. Stable association intervals detected across multiple environments were prioritized for the haplotype analysis. LD blocks and haplotypes were visualized using LDBlockShow, and phenotypic differences among haplotypes were evaluated using non-parametric statistical tests in R. Only accessions with complete genotype information within the target LD block were retained for the haplotype grouping. Candidate genes were annotated based on the TM-1 reference genome in the CottonMD database (https://yanglab.hzau.edu.cn/CottonMD), and putative functions were inferred using homologous gene annotations from The Arabidopsis Information Resource (TAIR, https://www.arabidopsis.org/). Tissue-specific expression profiles were examined using publicly available transcriptomic datasets from CottonMD to support inferences regarding their potential roles in cottonseed development.

Results

Phenotypic variation of cottonseed protein content in 259 materials

Across the seven environments, the mean cottonseed protein content ranged from 35.24% to 39.04%, with the highest mean in 23LHD (39.04%) and the lowest in 24KT (35.24%). This trait exhibited broad phenotypic dispersion (overall 27.00% ~ 50.76%) and environment-dependent variability, with coefficients of variation spanning 6.77% ~ 11.64%. Distributional metrics indicated mild right-skewness (0.01 ~ 0.74) and modest kurtosis (− 0.26 to 0.87), supporting an approximately continuous quantitative distribution (Table 1; Fig. 3).

Table 1.

Phenotypic Analysis of Cottonseed Protein Content

Trait	Env	Mean	SD	Min	Median	Max	Skewness	Kurtosis	CV(%)
Protein (%)	23WS	36.039	3.307	27.792	36.182	44.669	0.005	-0.258	9.175
	23YPH	38.206	3.904	29.448	38.097	50.733	0.287	0.021	10.217
	23LHD	39.038	3.810	28.292	38.791	50.756	0.303	0.171	9.760
	24WS	36.442	2.466	31.465	36.216	45.542	0.732	0.871	6.766
	24YPH	37.259	4.337	29.461	36.582	49.500	0.739	0.036	11.641
	24SP	37.064	3.300	27.590	36.842	48.196	0.349	0.597	8.903
	24KT	35.237	3.317	27.000	35.052	45.643	0.055	-0.145	9.414

Fig. 3.

Beeswarm plot showing the distribution of cottonseed protein content in 259 upland cotton accessions across seven environments

Analysis of variance revealed highly significant effects of environment (E), genotype (G), and genotype-by-environment interaction (G × E) (all P < 0.001), with F-values of 45.22, 3.15, and 26.97, respectively. The broad-sense heritability (H²) was 0.68 (Table S1-2). Collectively, these properties support the use of mixed-model–derived phenotypes and provide sufficient phenotypic contrast for GWAS.

Chromosomal distribution of genome-wide SNP markers in 259 upland cotton accessions

Whole-genome resequencing of 259 upland cotton accessions yielded 1,144,681 high-quality SNPs distributed across 26 chromosomes. These variants spanned approximately 2.22 Gb of the genome, with a mean density of 514.18 SNPs per Mb, indicating extensive genetic polymorphism and providing a robust marker resource for genetic dissection (Fig. 4). The chromosomal distribution was markedly asymmetric between the two sub-genomes. The A sub-genome contributed 848,358 SNPs, representing 74.11% of the total, whereas the D sub-genome contained 296,323 SNPs, representing 25.89%. Chromosome A08 exhibited the highest SNP density, harboring 154,546 SNPs, which accounted for 13.50% of the total, with a density of 1,230.80 SNPs per Mb, approximately 2.4-fold higher than the genome-wide average.

Fig. 4.

Chromosomal distribution of genome-wide single-nucleotide polymorphism (SNP) markers in 259 upland cotton accessions

Population structure and linkage disequilibrium (LD) analysis

Population structure analysis performed using ADMIXTURE identified K = 2 as the optimal number of subpopulations based on cross-validation (Fig. 5C, E). This stratification was corroborated by both the Neighbor-Joining phylogenetic tree and principal component analysis (Fig. 5B, D), indicating a detectable population structure within the panel. Genome-wide linkage disequilibrium (LD) decay analysis revealed that the LD coefficient (r²) approached baseline levels at approximately 500 kb (Fig. 5A), suggesting that the SNP marker density was sufficient to ensure adequate mapping resolution for subsequent genome-wide association studies (GWAS). Population structure analysis partitioned the 259 cotton accessions into two subpopulations. Group1 comprised 88 accessions (33.98%), including 6 exotic introductions (6.82%). Within Group1, 66 accessions (75.00%) originated from the Northwest Inland Cotton Region, 12 (13.64%) from the Yellow River Valley Cotton Region, 2 (2.27%) from the Yangtze River Valley Cotton Region, and 2 (2.27%) were landraces. Group2 comprised 171 accessions (66.02%), including 25 exotic introductions (14.62%). Within Group2, 65 accessions (38.01%) originated from the Yellow River Valley Cotton Region, 32 (18.71%) from the Yangtze River Valley Cotton Region, 31 (18.13%) from the Northwest Inland Cotton Region, and 18 (10.53%) were landraces.

Fig. 5.

Population genetic analyses of 259 upland cotton accessions. (A) Genome-wide linkage disequilibrium (LD) decay curve. (B) Principal component analysis (PCA) plot. (C) Cross-validation error used to determine the optimal number of subpopulations (K = 2). (D) Phylogenetic tree based on the K = 2 model. (E) Population structure inferred for K = 2

Genome-wide association study (GWAS) for cottonseed protein content

Best linear unbiased prediction (BLUP) was used to derive multi-environment phenotypes by accounting for environmental effects and genotype-by-environment variation. For GWAS, 12 phenotype datasets were analyzed, comprising seven single-environment datasets and five BLUP-based multi-environment summaries (23N, 24N, BJ, NJ, and ALL) (Fig. 6). The BLUP-derived phenotypes showed approximately normal distributions (Fig. 6A–E), supporting their suitability for subsequent mixed-model GWAS.

Fig. 6.

Frequency distribution of best linear unbiased prediction (BLUP) values for cottonseed protein content in 259 upland cotton accessions across five environmental sets: (A) Xinjiang, 2023; (B) Xinjiang, 2024; (C) Northern Xinjiang; (D) Southern Xinjiang; and (E) all seven environments combined

Utilizing 1,144,681 high-quality SNP markers, a genome-wide association study (GWAS) was conducted on 12 phenotypic datasets within a mixed linear model (MLM) framework that accounted for population structure and familial relatedness (P + K). The datasets comprised seven single-environment phenotypic observations (23LHD, 23WS, 23YPH, 24KT, 24SP, 24WS, and 24YPH) and five BLUP-based multi-environment combined estimates (23N, 24N, BJ, NJ, and ALL). Applying a significance threshold of − log₁₀(P) ≥ 4.32, 211 significant SNP–trait associations were identified across the 12 phenotypic datasets (Supplementary Table S2), corresponding to 195 unique SNPs after removing recurrent detections across datasets (Fig. 7A–L). Genome-wide, the significance of associated SNPs ranged from − log₁₀(P) = 4.33 to 5.73, and the proportion of phenotypic variance explained (PVE) by individual associations varied from 6.28% to 8.52%. The number of associations varied considerably across the datasets. Among the seven single-environment analyses, the 23WS dataset yielded the largest number of associations (67; PVE = 6.30%–8.41%), followed by 24YPH (38; 6.28%–8.10%), 24KT (25; 6.29%–7.79%), and 24WS (20; 6.29%–8.42%), whereas the remaining single-environment datasets detected between 2 and 12 associations. Among the BLUP-derived datasets, 24N yielded 13 associations, followed by NJ with 10, 23N with 6, and ALL and BJ with 4 each. With respect to chromosomal distribution, the significant associations were predominantly concentrated on chromosome D10 (38 associations, accounting for 18.01% of the total), followed by D01 (18, 8.53%), A02 (16, 7.58%), and A04 (15, 7.11%). To prioritize robust and reproducible association signals, we retained SNPs that exceeded the significance threshold in at least two of the datasets. This filtering step yielded 11 recurrent SNPs (− log₁₀(P) = 4.33–5.46; PVE = 6.28%–8.02%) (Table 2). Notably, two strong association signals on chromosome D02, D02:6,376,281 and D02:6,376,276, were consistently detected across the three datasets (24N, ALL, and 23WS), with the PVE reaching up to 8.02%. Loci A04:77,547,223 and A06:79,541,933 were also detected across the three datasets. Merging these recurrent loci using a ± 500 kb flanking window resulted in nine candidate association intervals (Table 2).

Fig. 7.

Manhattan (left) and quantile–quantile (QQ; right) plots of the genome-wide association analyses for cottonseed protein content in upland cotton. (A–C) Single-environment GWAS in 2023: Wensu (23WS), Yuepuhu (23YPH), and Liuhudi (23LHD). (D–G) Single-environment GWAS in 2024: Wensu (24WS), Yuepuhu (24YPH), Sanping (24SP), and Kuitun (24KT). (H–L) Multi-environment GWAS using BLUP-derived phenotypes: 2023 BLUP (23N), 2024 BLUP (24N), Northern Xinjiang BLUP (BJ), Southern Xinjiang BLUP (NJ), and combined BLUP across seven environments (ALL)

Table 2.

Summary of 11 recurrent significant SNPs within nine merged association intervals

NO	Chr	SNP	Position(bp)	Locus_ID	Merged interval (bp)	N	Environments	-Log10(P)	PVE(%)
1	A04	A04:765,080	765,080	Locus_01	265,079–1265080	2	23N;23YPH	4.35–4.59	6.43–6.83
2	A04	A04:77,547,223	77,547,223	Locus_02	77,047,222–78047223	3	ALL;NJ;24YPH	4.65–5.03	6.77–7.38
3	A06	A06:10,009,230	10,009,230	Locus_03	9,509,229–10509230	2	23LHD;24YPH	4.54–4.58	6.76–6.84
4	A06	A06:79,541,933	79,541,933	Locus_04	79,041,932–80041933	3	24N;ALL;NJ	4.73–5.19	6.94–7.68
5	A12	A12:84,691,384	84,691,384	Locus_05	84,191,383–85,191,406	3	23N;NJ;24YPH	4.33–4.87	6.28–7.15
6	A12	A12:84,691,406	84,691,406			2	23N;NJ	4.37–4.45	6.41–6.55
7	D02	D02:6,376,276	6,376,276	Locus_06	5,876,275–6,876,281	3	24N;ALL;23WS	4.48–5.36	6.47–7.87
8	D02	D02:6,376,281	6,376,281			3	24N;ALL;23WS	4.71–5.46	6.83–8.02
9	D11	D11:19,580,385	19,580,385	Locus_07	19,080,384–20080385	2	NJ;23YPH	4.61–4.78	6.78–7.05
10	D11	D11:68,672,753	68,672,753	Locus_08	68,172,752–69,172,753	2	23LHD;24YPH	4.40–4.55	6.37–6.61
11	D13	D13:45,398,461	45,398,461	Locus_09	44,898,460–45898461	2	24N;BJ	4.41–4.49	6.52–6.66

Haplotype analysis of recurrent SNPs

Among the 11 stable significant SNPs detected, the locus D11:19,580,385 on chromosome D11 was situated within an LD block comprising nine linked SNPs (r² ≥ 0.8; Fig. 8A), indicating a strong local linkage disequilibrium in the panel. The remaining ten significant SNPs were distributed across different chromosomes or LD blocks with low LD (r² < 0.2) and were, therefore, excluded from the haplotype analysis. Based on the allelic combinations of the nine SNPs within the block, three major haplotypes with frequencies greater than 5% were identified among the 158 accessions with complete genotype data points. Accessions with missing genotypes in this LD block were excluded from the haplotype grouping. and were designated as Hap1, Hap2, and Hap3 (Fig. 8C), collectively accounting for all the effective samples. Hap1 was the most frequent (73.4%, 116/158), followed by Hap2 (15.8%, 25/158), and Hap3 was the least frequent (10.8%, 17/158). The Kruskal–Wallis non-parametric test was employed to compare phenotypic differences among haplotype groups, revealing significant overall variations (P < 0.05; Fig. 8D). Accessions carrying Hap3 tended to show higher phenotypic means than those carrying Hap1 in several datasets, including 24WS, ALL, 24N, and BJ. In contrast, the phenotypic performance of Hap2 was generally intermediate between Hap1 and Hap3 and did not show a consistent significant difference between either group. These results suggest that Hap3, although the least frequent haplotype in this population, is associated with increased cottonseed protein content in several datasets and may represent a favorable haplotype for further validation.

Fig. 8.

Linkage disequilibrium, haplotype analysis, and phenotypic associations of the stable significant SNP locus D11:19,580,385 and its surrounding LD block. (A) LD heatmap of stable SNPs (R²). (B) Sequence motif analysis showing nucleotide preferences. (C) Three major haplotypes (Hap1–Hap3) and their genotype distribution. (D) Phenotypic variation among haplotypes across different environments. Significant overall differences were assessed using the Kruskal–Wallis test

GO and KEGG functional enrichment analysis of candidate genes

To characterize the functional annotation profiles of candidate genes associated with cottonseed protein content, GO enrichment analysis was performed and visualized (Fig. 9), whereas KEGG pathway results were summarized as supplementary functional evidence (Supplementary Figure S1). In total, 71 candidate genes with valid GO annotations were included in the GO enrichment analysis, and 28 candidate genes with valid KEGG pathway annotations were included in the KEGG pathway analysis. The GO enrichment results showed that the candidate genes were assigned to three GO ontologies: Molecular Function, Biological Process, and Cellular Component. The Molecular Function ontology contained 42 enriched GO terms, of which eight terms were significant at P < 0.05 and two terms remained significant at FDR < 0.05. The Biological Process ontology contained 31 enriched GO terms, of which 8 terms were significant at P < 0.05 and 1 term remained significant at FDR < 0.05. The Cellular Component ontology contained eight enriched GO terms, of which two terms reached nominal significance, but none remained significant after FDR correction. Among the FDR-significant GO terms, protein binding, ADP binding, and signal transduction were the most prominent terms. Protein binding included 26 candidate genes (P = 2.02 × 10⁻⁵, FDR = 0.00163), ADP binding included 6 candidate genes (P = 5.55 × 10⁻⁴, FDR = 0.0183), and signal transduction included 5 candidate genes (P = 6.79 × 10⁻⁴, FDR = 0.0183). In contrast, no KEGG pathway remained significant after the FDR correction. At the nominal P < 0.05 level, lysine biosynthesis was detected, and Gh_D11G350500 was involved, providing limited auxiliary evidence for amino acid metabolism. Therefore, GO enrichment was used as one of the main sources of functional interpretation, whereas KEGG results were interpreted cautiously as supplementary annotation evidence rather than as statistically robust pathway enrichment results (Fig. 10).

Fig. 9.

Gene Ontology (GO) functional enrichment analysis of candidate genes. (A) GO enrichment bubble plot. The x-axis represents the Rich Factor, the bubble size indicates the number of enriched genes, and the color gradient denotes the significance level (P‑value) of enrichment. (B) GO enrichment bar plot. The distribution of enriched candidate genes across GO terms is presented according to the three major ontologies: Biological Process, Cellular Component, and Molecular Function

Fig. 10.

Expression heatmap of the selected candidate genes. Spatiotemporal expression profiles based on TM-1 (CRI) RNA-seq data from CottonMD. The samples cover vegetative tissues, reproductive tissues, fibers and ovules at different developmental stages (DPA, days post anthesis). Gene expression levels were normalized by log2(TPM + 1), and the color gradient indicates the level of gene expression (blue for low expression and red for high expression)

Candidate gene identification

To further refine the candidate genes associated with cottonseed protein content, a stepwise filtering strategy was applied based on repeatedly detected GWAS signals across multiple environments. Initially, 11 stable SNPs were selected from the recurrent significant association signals, and adjacent or overlapping signals were merged into nine candidate association intervals. Based on the gene annotation of the upland cotton TM-1 reference genome, 437 genes were identified within these intervals. SNPs located in the candidate intervals were functionally annotated using SnpEff, with missense variants as the criterion for nonsynonymous mutation screening (Supplementary Table S3). This analysis identified 93 genes harboring nonsynonymous variants. By integrating CottonMD public expression profiles with functional annotations of Arabidopsis homologs, these genes were further prioritized, resulting in 13 candidate genes. The predicted functional annotations are summarized in Table 3.

Table 3.

Gene Function Annotation

GeneName	Gene annotation
Gh_D11G350500	4-hydroxy-tetrahydrodipicolinate reductase; nominally associated with lysine biosynthesis-related GO/KEGG annotations
Gh_D02G053000	Encodes a di- and tri-peptide transporter involved in responses to wounding, virulent bacterial pathogens, and high NaCl concentrations. The protein is predicted to have 12 transmembrane helices
Gh_A12G157300	Serine carboxypeptidase-like 44; (source: Araport11)
Gh_A12G158700	Encodes a nuclear-localized protein with similarity to animal polycomb repressive core complex1 (PRC1) core component RING. Appears to function redundantly with ATRING1a, a close paralog. Both interact physically with CLF and LHP1 and appear to function together to repress class I KNOX gene expression
Gh_D02G054000	Encodes MAC3A, a U-box protein with homology to the yeast and human E3 ubiquitin ligase Prp19. Associated with the MOS4-Associated Complex (MAC). Involved in plant innate immunity
Gh_A04G004500	Got1/Sft2-like vesicle transport protein family;(source: Araport11)
Gh_A04G010500	Encodes SEC24a/ERMO2. Required for endoplasmic reticulum (ER) morphology. Has epistatic interactions with AT1G55350, AT3G59420, and AT3G10525
Gh_D02G054800	SNARE protein located in Golgi apparatus
Gh_A04G005400	Thermosensor which primes heat-induced stress granule formation via biomolecular condensation
Gh_A04G005100	Serine carboxypeptidase-like 44; (source: Araport11)
Gh_D11G185300	Eukaryotic aspartyl protease family protein;(source: Araport11)
Gh_D02G052600	Ribosomal protein-related gene; annotated with ribosome/translation-related terms, but not significantly enriched after correction
Gh_A12G157400	Other names: IST1-LIKE 10; ISTL10 (Ist1p); (source: Araport11)

Expression profile analysis indicated that the 13 prioritized genes displayed distinct expression patterns across roots, stems, leaves, floral organs, fibers, and ovules at different developmental stages. Among them, Gh_A12G157300 and Gh_A12G158700 were expressed in ovules at 0, 1, 3, 5, 10, 15, 20, and 25 days post anthesis (DPA), with relatively high transcript abundance at multiple stages. Gh_D02G054000 was consistently expressed in ovules from 0 to 25 DPA and also showed relatively high expression in roots, stems, leaves, floral organs, and fibers. Gh_D02G052600 was expressed at several ovule developmental stages, with relatively pronounced expression at approximately 10 DPA. In addition, Gh_A04G004500, Gh_A04G010500, and Gh_A04G005400 were expressed in ovules and other tissues, such as roots, stems, leaves, and floral organs. In contrast, Gh_A04G005100, Gh_D11G185300, Gh_A12G157400, Gh_D02G053000, and Gh_D02G054800 exhibited weak ovule expression or expression restricted to a limited number of tissues. Collectively, these expression profiles suggest that Gh_A12G157300, Gh_A12G158700, Gh_D02G054000, and Gh_D02G052600 are expression-supported candidates for further investigation of cottonseed protein accumulation.

Discussion

Comprehensive utilization value of cottonseed protein

The rising global demand for protein has intensified the pressure on conventional animal protein systems, which are increasingly constrained by supply capacity and production costs. Cottonseed meal is an important complementary protein resource because of its high protein content and widespread availability as a byproduct of cotton production (Li et al. 2012). Previous studies have also reported broad variations in cottonseed protein content across diverse upland cotton germplasm panels (Yuan et al. 2018), whereas the present study observed a range of 27.00% ~ 50.76%. This breadth of variation indicates substantial phenotypic diversity that can support genetic dissection and trait improvements. Beyond its abundance, cottonseed protein exhibits favorable nutritional value, digestibility, cost-effectiveness, and desirable functional properties, including solubility and water absorption (Ma et al. 2018). These attributes support their potential applications as food ingredients (Kumar et al. 2022), meat substitutes (Cutroneo et al. 2024), alternative protein ingredients (Kumar et al. 2021a), and food packaging materials (Biswas et al. 2023). Nevertheless, large-scale utilization is limited by two major constraints. One limitation is the nutritional imbalance associated with lysine, which is the first limiting amino acid (Świątkiewicz et al. 2016). The other is gossypol, a toxic polyphenolic compound that accumulates in cottonseeds (Gadelha et al. 2014). Recent advances in breeding and processing technologies, including ultra-low-gossypol cottonseed development (Rathore et al. 2020) and microbial fermentation detoxification of cottonseed meal (Dharmakar et al. 2023), have alleviated these constraints and expanded the prospects for cottonseed protein utilization. In this context, genetic approaches that enable the systematic dissection of protein traits have become increasingly important. GWAS can facilitate the identification of loci, candidate genes, and molecular markers underlying variations in protein content and quality, thereby supporting molecular design breeding and marker-assisted selection. Such advances may accelerate the development of cotton germplasm combining reduced gossypol accumulation with improved essential amino acid profiles, strengthening the basis for the comprehensive utilization of cottonseed protein and contributing to protein-supply security.

SNP loci identification and comparison with previous studies

In this study, 11 stable SNP loci were identified, corresponding to 9 candidate association intervals distributed on chromosomes A04, A06, A12, D02, D11, and D13. The phenotypic variance explained (PVE) by these significant loci ranged from 6.28% to 8.02%, consistent with the genetic characteristics of cottonseed protein content as a complex quantitative trait controlled by multiple genes. Previous studies have shown that multiple association loci or QTLs for cottonseed protein content can be detected across different genetic backgrounds and environmental conditions. For example, Yuan et al. (Yuan et al. 2018) used 196 upland cotton germplasm accessions and the CottonSNP80K array to identify SNPs and QTL intervals associated with cottonseed protein content across multiple environments and reported that the phenotypic contribution of individual loci was generally moderate. Yu et al. (Yu et al. 2012) identified 22 QTLs for protein content in a backcross inbred line population derived from upland cotton and sea-island cotton, further supporting the polygenic genetic basis of this trait. In the present study, loci A04:77,547,223, A06:79,541,933, A12:84,691,384/A12:84,691,406, and D02:6,376,276/D02:6,376,281 were repeatedly detected in at least two environments or BLUP-derived integrated phenotypes, indicating relatively strong environmental robustness.

Compared with previous studies, two highly adjacent SNPs, A12:84,691,384 and A12:84,691,406, were detected within the interval A12:84,191,383–85,191,406. These two loci were repeatedly identified in multiple environments or integrated phenotypes, including 23N, NJ, and 24YPH, with PVE values ranging from 6.28% to 7.15%. Liu et al. (Liu et al. 2015), based on a multi-environment association analysis of 180 upland cotton cultivars, also detected repeatedly occurring protein-content-associated markers on chromosome A12. In addition, Ai et al.(Ai et al. 2026) recently conducted a GWAS for cottonseed protein content in upland cotton and reported protein-content-associated QTNs, including TM42986 (A12:83,905,690) and TM42987 (A12:83,911,906). These previously reported QTNs are located approximately 0.78 Mb from the A12:84,691,384/A12:84,691,406 signals identified in the present study, indicating that this region of chromosome A12 may repeatedly harbor protein-content-associated signal. These results indicate that this interval on chromosome A12 shows stable association signals in GWAS screening and is valuable for further investigation. In contrast, D02:6,376,276 and D02:6,376,281 were located within the interval D02:5,876,275–6,876,281 and were repeatedly detected in 24N, ALL, and 23WS. Among them, D02:6,376,281 showed the highest PVE value of 8.02%, making it one of the loci with a relatively high explanatory power in this study. Liu et al. (Liu et al. 2015) detected repeatedly occurring SSR markers associated with protein content on chromosome D02 through GWAS analysis. Wang et al. (Wang et al. 2019) constructed a high-density SLAF-seq SNP genetic map and performed QTL mapping for cottonseed kernel weight, oil content, and protein content, and their results also indicated the presence of multiple protein-content-related QTL regions on chromosome D02. Therefore, the results of the present study are consistent with previous findings at the chromosomal level, providing additional support for GWAS results. In contrast, loci A04:765,080, A04:77,547,223, A06:10,009,230, A06:79,541,933, D11:19,580,385, D11:68,672,753, and D13:45,398,461 were not identical to the protein-content-associated loci discussed above and may represent additional association signals identified in this study. Among them, A06:79,541,933, located within the interval A06:79,041,932–80041933, was stably detected in 24N, ALL, and NJ, with -log10(P) values ranging from 4.73 to 5.19 and PVE values ranging from 6.94% to 7.68%, indicating a good cross-dataset repeatability. Ai et al. (Ai et al. 2026) also detected QTNs or QTN-by-environment interaction loci associated with cottonseed protein content on chromosomes A06 and D13, and reported that candidate genes such as GH_A06G1663 and GH_D13G0601 showed relatively high expression levels in ovules. These findings suggest that chromosomes A06 and D13 may harbor loci associated with cottonseed protein accumulation.

In summary, the significant SNPs detected in this study were repeatedly identified across multiple environments, indicating that the genetic improvement of cottonseed protein content may depend on the cumulative effects of multiple SNP loci. Among them, A12:84,691,384 and A12:84,691,406 were located near previously reported protein-content-associated signals on chromosome A12, whereas D02:6,376,276 and D02:6,376,281 showed good repeatability and relatively high phenotypic explanatory power, and were also consistent with previous studies reporting protein-content-related loci on chromosome D02. Therefore, these loci may serve as priority target intervals for subsequent candidate gene mining.

GWAS analysis and candidate gene identification

Functional annotation of SNP variants was performed using SnpEff, with missense_variant as the criterion for identifying nonsynonymous variants (Cingolani et al. 2012). Based on this criterion, 93 candidate genes harboring nonsynonymous variants were identified and further examined by integrating GO annotation, KEGG pathway information, public expression profiles from CottonMD, and functional information inferred from Arabidopsis homologs (Aleksander et al. 2023; Yang et al. 2023). Among these genes, 71 had valid GO annotations and were included in the GO enrichment analysis, accounting for 76.34% of the total candidate set, whereas 28 genes had valid KEGG pathway annotations and were incorporated into the KEGG pathway enrichment analysis, accounting for 30.11%. Therefore, the GO and KEGG enrichment results should be interpreted as functional tendencies of the annotatable subset rather than as evidence that all 93 candidate genes were simultaneously represented in both analyses. Among the 13 prioritized candidate genes, 11 had GO annotations, 8 had KEGG pathway annotations, and 12 possessed at least one GO or KEGG annotation, providing a functional basis for the subsequent candidate gene classification.

GO enrichment analysis showed that the annotatable candidate genes were assigned to the Molecular Function, Biological Process, and Cellular Component categories. In particular, protein binding, ADP binding, and signal transduction remained significant after multiple-testing correction. The protein binding term contained 26 candidate genes (P = 2.02 × 10⁻⁵, FDR = 0.00163), ADP binding contained 6 candidate genes (P = 5.55 × 10⁻⁴, FDR = 0.0183), and signal transduction contained 5 candidate genes (P = 6.79 × 10⁻⁴, FDR = 0.0183). These results suggest that the genes harboring nonsynonymous variants were mainly associated with protein interactions, nucleotide-binding-related molecular functions, and regulatory processes within the annotatable subset. In contrast, although KEGG pathways such as lysine biosynthesis and monobactam biosynthesis reached nominal significance at the unadjusted level of P < 0.05, no KEGG pathway remained significant after FDR correction. Therefore, KEGG pathway information was used only as auxiliary functional evidence and was not considered a statistically robust conclusion.

According to the integrated annotation results, the 13 prioritized candidate genes were broadly assigned to four functional categories. The first category included genes putatively associated with amino acid biosynthesis or organic nitrogen transport, represented by Gh_D11G350500 and Gh_D02G053000, respectively. Gh_D11G350500 was annotated as 4-hydroxy-tetrahydrodipicolinate reductase and was nominally assigned to lysine biosynthesis-related GO/KEGG terms, suggesting a possible association with lysine biosynthesis-related amino acid metabolism. Previous studies have shown that lysine biosynthesis pathway genes are involved in seed amino acid metabolism in crop species, providing a biological context for interpreting this annotation (Liu et al. 2016). Gh_D02G053000 was annotated as a peptide transporter-related gene, implying its possible involvement in organic nitrogen or small peptide transport. Amino acid and peptide transporters participate in nitrogen uptake, allocation, and source-to-sink nutrient partitioning in plants, providing a reasonable framework for interpreting the function of this candidate gene (Rentsch et al. 2007; Tegeder and Rentsch 2010). The second category contained Gh_D02G052600, which was annotated with ribosome/translation-related terms and may be associated with protein translation at the annotation level. The third category comprised genes related to protein transport and vesicle-mediated trafficking, including Gh_A04G010500, Gh_D02G054800, Gh_A04G004500, and Gh_A12G157400. Among them, Gh_A04G010500 was annotated with COPII-related vesicle transport terms, Gh_D02G054800 was annotated as a Golgi-localized SNARE protein, and Gh_A12G157400 showed an annotation-level link to endocytosis-related processes. These annotations suggest that intracellular protein transport, vesicle sorting, and membrane trafficking may be associated with the accumulation of cottonseed proteins. This interpretation is consistent with previous studies showing that seed storage protein deposition depends on endomembrane trafficking, vesicle-mediated sorting, and protein storage vacuole targeting (Ping et al. 2021). In addition, SNARE-mediated membrane fusion has been reported to be required for protein storage vacuole biogenesis and seed development in Arabidopsis (Kazuo et al. 2008). Therefore, although these genes cannot yet be regarded as functionally validated regulators of cottonseed protein content, they provide plausible candidates for further investigation of intracellular transport processes potentially related to seed protein deposition. The fourth category included genes annotated as being related to protein processing, degradation, or proteostasis, namely Gh_A04G005100, Gh_D11G185300, Gh_D02G054000, Gh_A12G157300, Gh_A12G158700, and Gh_A04G005400. Gh_A04G005100 and Gh_D11G185300 were annotated as protease-related genes, whereas Gh_D02G054000, Gh_A12G157300, and Gh_A12G158700 were associated with ubiquitination and E3 ubiquitin ligase-related functions. The ubiquitin–proteasome system plays a central role in protein turnover, post-translational regulation, and the formation of several agronomic and seed-related traits in plants (Linden and Callis 2020). Accordingly, these genes may be more closely related to protein turnover and proteostasis than to the direct structural synthesis of seed storage proteins. In addition, Gh_A04G005400 was assigned to the GO term protein binding, suggesting that it may be associated with protein interaction-related processes. However, because protein binding is a relatively broad functional term, this gene should be further evaluated in combination with its expression pattern and variant effect before being considered a functionally validated candidate gene.

Taken together, the GO and KEGG analyses of the 93 candidate genes harboring nonsynonymous variants provided useful functional clues for interpreting the genetic basis of cottonseed protein content. The GO enrichment results based on 71 annotatable genes highlighted protein interaction, ADP-binding-related molecular functions, and signal regulation, whereas KEGG pathway information provided only nominal auxiliary evidence, mainly for the lysine biosynthesis annotation of Gh_D11G350500. Among the 13 prioritized candidate genes, 12 had GO or KEGG functional annotations, and their predicted functions were distributed across amino acid substrate supply, protein translation, vesicle-mediated transport, protein processing, and proteostasis regulation. These results suggest that cottonseed protein content is unlikely to be controlled by a single linear protein synthesis pathway but may involve multiple coordinated biological processes. Based on association mapping, functional annotation, and public expression profile evidence, Gh_D11G350500, Gh_D02G052600, Gh_D02G054000, Gh_A12G157300, Gh_A12G158700, Gh_A04G010500, and Gh_D02G054800 were selected as priority candidate genes for subsequent studies. However, these genes remain annotation-based candidates, and their biological roles in cottonseed protein accumulation require further validation using qRT-PCR, gene editing, genetic transformation, or other functional experimental approaches.

Conclusion

By integrating phenotyping across seven environments with resequencing-based genome-wide association analysis, this study dissected the complex genetic architecture of cottonseed protein content in a natural population of 259 upland cotton accessions. Based on 12 phenotypic datasets, 211 significant SNP–trait associations were detected, corresponding to 195 unique SNPs, among which 11 recurrent SNPs were consolidated into nine association intervals. These results reveal a polygenic basis shaped by both stable genetic effects and environment-sensitive genetic components. Notably, haplotype analysis of a representative stable interval on chromosome D11 (D11:19,580,385) revealed haplotype-dependent variation in protein content in several environments and BLUP-derived datasets, suggesting that this haplotype block may be a useful target for further validation in marker-assisted selection. Through the integration of functional variant annotation and ovule transcriptomic data, we preliminarily identified 13 biologically plausible candidate genes. Among these, priority candidates, including Gh_D11G350500, Gh_D02G052600, Gh_D02G054000, Gh_A12G157300, Gh_A12G158700, Gh_A04G010500, and Gh_D02G054800, were annotated with putative roles in lysine biosynthesis, protein translation, vesicle-mediated transport, ubiquitin-related regulation, and protein interaction-related processes, implying their possible involvement in cottonseed protein accumulation. However, the precise biological roles of these genes remain to be experimentally validated. Collectively, the stable loci and candidate genes identified in this study may serve as a useful reference for marker‑assisted selection breeding and future functional characterization of genes underlying cottonseed protein content.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

ANOVA

Analysis of variance

BLUP

Best linear unbiased prediction

CRI

Cotton Research Institute (version/assembly of the TM-1 reference genome)

CRISPR

Clustered regularly interspaced short palindromic repeats

CSSLs

Chromosome segment substitution lines

CV

Coefficient of variation

Cas9

CRISPR-associated protein 9

DA1

DA1 (organ size regulator; DA1 peptidase family)

DAR

DA1-related

DPA

Days post anthesis

E3

E3 ubiquitin-protein ligase

FPKM

Fragments per kilobase of transcript per million mapped reads

Fe–S

Iron–sulfur

GDH

Glutamate dehydrogenase

GEMMA

Genome-wide Efficient Mixed Model Association

GOGAT

Glutamate synthase (glutamate:2-oxoglutarate aminotransferase)

GS

Glutamine synthetase

GWAS

Genome-wide association study

H²

Broad-sense heritability

HSD

Honestly significant difference (Tukey’s HSD test)

HSP40

Heat shock protein 40

LD

Linkage disequilibrium

MAC

MOS4-associated complex

MAC3A/3B

MOS4-associated complex subunits 3A/3B

MAF

Minor allele frequency

MAS

Marker-assisted selection

MLM

Mixed linear model

MOS4

Modifier of snc1 4

NPF

NRT1/PTR family (nitrate transporter 1/peptide transporter family)

NR

Nitrate reductase

NRT1.7

Nitrate transporter 1.7

NiR

Nitrite reductase

PCA

Principal component analysis

PLINK

Tool set for whole-genome association and population-based linkage analyses

PTR

Peptide transporter

PTR/NPF

Peptide transporter / NRT1-PTR family

PVE

Phenotypic variance explained

QTL

Quantitative trait locus

RGF

Root meristem growth factor

RILs

Recombinant inbred lines

RNA-seq

RNA sequencing

SLAF-seq

Specific-locus amplified fragment sequencing

SNP

Single-nucleotide polymorphism

SSR

Simple sequence repeat(s)

SUFE

SufE-like protein (Fe–S cluster assembly factor)

TM-1

Upland cotton genetic standard line TM-1 (reference genome)

VIGS

Virus-induced gene silencing

r²

Squared correlation coefficient (LD measure)

Author contributions

Z.W., H.L., and Y.Z.: methodology, data curation, software, formal analysis, validation, writing—original draft. H.S., and Q.W.: investigation, validation, supervision. K.Z., Q.C., Q.Z., and X.D.: investigation, supervision, resources, project administration, writing—review & editing, funding acquisition. All authors have read and agreed to the final manuscript.

Funding

This work was supported by the Xinjiang Key Research and Development Program (2024B02001-1/-2), the Xinjiang Major Science and Technology Project (2024A02003-3).

Data availability

The data presented in this study are included in the article and its supplementary materials. Additional data and inquiries can be directed to the corresponding author upon reasonable request.

Declarations

Conflicts of interest

The authors declare no conflicts of interest.