A total of 240 G. barbadense accessions from diverse countries and regions were collected for seedling-stage salt tolerance evaluation. These included 220 mainstream cultivars from Xinjiang, China, three wild G. barbadense accessions collected from Yunnan and Hainan, China, six Pima cotton germplasm lines from the United States, six cultivated materials from Egypt, and five G. barbadense varieties from Central Asia.

Sulfuric acid-delinted seeds were sown in 10 × 5 seedling trays. After 3 days of germination, seedlings were transferred to a hydroponic system containing 1/2-strength Murashige and Skoog (MS) nutrient solution (pH 5.8). The nutrient solution was replaced every 3 days, and continuous aeration was maintained using an air pump. Plants were grown under controlled environmental conditions in the greenhouse at Zhejiang University Agricultural Experiment Station. NaCl treatment (200 mmol/L) was initiated at the two true leaves and one apical bud stage, while control groups remained untreated, both the control groups and the salt-stress groups were cultured synchronously in the hydroponic system. After 7 days of salt stress, senesced cotyledons were removed. Plant height (cotyledonary node to apical meristem) and shoot fresh weight were measured. Roots and shoots were then oven-dried at 105 °C for 60 min followed by 80 °C to constant weight for dry weight determination. Three biological replicates per treatment were maintained to ensure experimental reliability. All measured parameters were converted to the Salt Tolerance Index (STI), which was calculated using Equation 1:

The traits included relative plant height (RPH), relative shoot fresh weight (RSFW), relative shoot dry weight (RSDW), and relative root dry weight (RRDW). To minimize environmental variance across years and emphasize genetic effects, best linear unbiased estimates (BLUEs) for four traits (2022–2023 data) were calculated using the lme4 R package:

In Equation 2, STI serves as the dependent variable, with Sample (genotype) designated as the fixed-effect independent variable. The term 1|Rep denotes experimental replicates modeled as random effects, 1|Year represents year-specific random effects, and 1|Year: Rep specifies the nested random effects of replicates within years.

BLUE values were analyzed for descriptive statistics and ANOVA using SPSS v26. Phenotypic frequency distributions and correlations were visualized with the Hmisc R package. Graphs were generated using GraphPad Prism.

A membership function method was applied to comprehensively evaluate seedling-stage salt tolerance across traits.

In Equation 3, F i denotes the comprehensive index factor score of genotype i. E i j represents the eigenvector corresponding to the j-th individual indicator in the i-th principal component. S T I j indicates the salt tolerance index of the j-th individual indicator for genotype i. In Equation 4, μ ( F i ) refers to the membership value of the comprehensive index for genotype i, where F i m a x and F i m i n represent the maximum and minimum values of the comprehensive index, respectively. Equation 5 defines W i as the weight (relative importance) of the i-th comprehensive index among all indices, calculated from its contribution rate ( P i ). In Equation 6, D quantifies the integrated salt tolerance coefficient.

Correlation analysis was performed using SPSS. Hierarchical clustering analysis was conducted in R with the hclust function, employing Euclidean distance and Ward’s minimum variance method for cluster aggregation.

Oxidative stress markers, including superoxide dismutase (SOD) activities, malondialdehyde (MDA) and proline (Pro) content were measurement using established protocols outlined (Qian et al., 2024). All assays utilized a 0.1 g fresh sample of leaves.

Resequencing data PRJNA728217 (Yu et al., 2021) for 240 G. barbadense accessions was downloaded from the NCBI SRA database using SRA Toolkit. Raw Illumina paired-end reads were quality-filtered using Fastp (Chen et al., 2018), with parameters “-c -n 15 -u 50 -q 15” to retain high quality sequences (Shao et al., 2022). Clean reads were aligned to the G. barbadense (AD2) ‘3-79’ reference genome HAU (Wang et al., 2019) using BWA (Li and Durbin, 2009), following index construction with the same tool. Alignment files were converted to binary BAM format using SAMtools (Li et al., 2009), sorted with sambamba (Tarasov et al., 2015), and PCR duplicates were removed.

SNPs and InDels were identified using the HaplotypeCaller module in GATK (Mckenna et al., 2010). GVCF files were merged with CombineGVCFs and converted to VCF format. Variants were filtered using VariantFiltration module in GATK (Mckenna et al., 2010), followed by additional filtering (maf > 0.05, max-missing > 0.8) to obtain GWAS-compatible SNPs (Yu et al., 2021). Functional annotation was performed using ANNOVAR. SNP/InDel densities were calculated with VCFtools (Danecek et al., 2011), and chromosomal distributions were visualized using RColorBrewer and stringr R packages.

A genetic distance matrix generated by VCF2Dis was used to construct a Neighbor-Joining (NJ) phylogenetic tree via the FastME online platform (http://www.atgc-montpellier.fr/fastme). Population structure was inferred using Admixture (Alexander et al., 2009), and PCA was conducted with GCTA (Yang et al., 2011) to resolve substructure and mitigate false positives in association studies. Genome-wide linkage disequilibrium (LD) decay was assessed with PopLDdecay (Zhang et al., 2019a).

To quantify genetic divergence and variation among subpopulations, pairwise population differentiation index (F _ST) and nucleotide diversity (π) were calculated genome-wide using VCFtools (Danecek et al., 2011) with 100-kb sliding windows and 20-kb steps.

Three types of salt stress phenotypic data for the seedling stage of G. barbadense were generated based on the calculated BLUE values and the phenotypes observed in 2022 and 2023. The genome-wide efficient mixed-model association (GEMMA) software (Zhou and Stephens, 2012), was used to correct for population stratification by incorporating both PCA and kinship matrices. Manhattan plots were generated using the R package CMplot to represent the distribution of SNPs and their corresponding P-values, while quantile-quantile (QQ) plots were constructed to evaluate the model’s performance. SNP filtering was performed using Plink software based on linkage disequilibrium criteria (window size = 50, step size = 50, r² ≥ 0.2), resulting in a total of 213,990 effective SNPs. A stringent threshold of p < 4.67 × 10^-6 was set to identify significant association loci (Yu et al., 2021). However, due to the risk of overly stringent thresholds excluding true trait-associated genetic loci with p-values that do not meet the strict cutoff, SNPs with p < 1.0 × 10^-5 identified in at least two environments or phenotypes were also retained to capture more candidate genes (Zhao et al., 2022).

RNA-seq data were downloaded from NCBI under project numbers PRJNA490626 (Hu et al., 2019) (salt stress treatments at 1 h, 3 h, 6 h, 12 h, and 24 h) and PRJNA601953 (salt stress treatment at 14 days) (Dong et al., 2022). Additionally, the G. barbadense genome (version 379_HAU) was obtained from COTTONGENE (https://www.cottongen.org/species/Gossypium_barbadense/nbi-AD2_genome_v1.0) and the genome index was built using ​​HISAT2 (Kim et al., 2019). Quality control and filtering were performed using ​​Fastp​​ (Chen et al., 2018) with the following criteria (Song et al., 2023): paired reads were removed if any read met the following criteria: ambiguous “N” bases exceeded 10% of the read length; >50% of bases had low quality (Q ≤ 5); or adapter sequences were detected. Reads were aligned to the reference genome using SAMtools (Li et al., 2009), and gene expression levels were quantified as FPKM values using StringTie (Pertea et al., 2015).

Two contrasting materials, the highly salt-tolerant line H160 and the salt-sensitive line H20, were subjected to salt stress (200 mmol/L NaCl) at 0 h, 24 h, and 48 h, with three biological replicates per time point. Total RNA was extracted from leaf samples using an RNA extraction kit (TIANGEN), reverse-transcribed into cDNA using a reverse transcription kit (TOROIVD), and subjected to qRT-PCR analysis using enzymes from TOROGreen^® qPCR Master Mix (TOROVID). The qPCR reactions were performed in a 20 μL reaction system containing 2 μL of cDNA, 10 μL of qPCR master mix, 4 μL of forward primer, and 4 μL of reverse primer, with each primer at a final concentration of 1–2 micromolars. The housekeeping gene UBQ7 (Ubiquitin extension protein 7) was employed as an internal control gene (Wang et al., 2013). Each template was analyzed in triplicate technical replicates. The qPCR protocol was conducted using the LightCycler 96 real-time PCR system (Roche) with the following cycling conditions: initial denaturation at 96°C for 3 minutes, followed by 40 cycles of denaturation at 95°C for 10 seconds and annealing/extension at 60°C for 30 seconds. Primers used for quantitative analysis are listed in Supplementary Table S1 .

Virus-induced gene silencing (VIGS) was performed using the tobacco rattle virus (TRV)-based pTRV1/2 vector system.​​ Target gene fragments were amplified from G. barbadense cDNA and cloned into the pTRV2 vector via homologous recombination using EcoRI and KpnI restriction sites. The constructed vectors were transformed into Escherichia coli DH5α cells and validated by plasmid sequencing. Verified vectors were introduced into Agrobacterium tumefaciens strain GV3101 via heat-shock transformation. ​​Agrobacterial cultures​​ harboring the vectors were grown in liquid medium supplemented with rifampicin (50 μg/mL) and kanamycin (50 μg/mL) at 28°C with 200 rpm agitation for 12 h. Bacterial cells were harvested, resuspended in infiltration buffer (10 mM MgCl2, 10 mM MES, 200 μM acetosyringone), and adjusted to OD600 = 1.0. After 2–3 h of dark incubation at 28°C, Agrobacterium suspensions carrying pTRV1 and pTRV2 (negative control), pTRV1 and pTRV2: CLA1 (positive control), or pTRV1 and pTRV2: GbXTH27 were mixed at 1:1 ratio. The mixtures were infiltrated into cotyledons of 7-day old G. barbadense 3–79 seedlings using sterile syringes. ​​Post-infiltration​​, plants were maintained in darkness for 24 h, then transferred to a growth chamber at 25°C under 16 h light/8 h dark cycles. The empty pTRV2 vector served as a negative control, while pTRV2: CLA1 (essential for chloroplast development) was used as a positive control, inducing characteristic leaf whitening within two weeks due to chloroplast defects (Mandel et al., 1996). Leaves from pTRV2 and pTRV2: GbXTH27 infiltrated plants were collected at two weeks post-infiltration for RNA extraction and qPCR validation of GbXTH27 silencing efficiency. ​​All primers used for VIGS vector construction are listed in Supplementary Table S1 .

The DAB solution was formulated by dissolving DAB powder in distilled water to reach a concentration of 1 mg/mL, with its pH adjusted to 3.8. For the staining procedure, leaves were immersed in the prepared DAB solution and incubated at 28°C in the dark for 12 hours. Following this, the leaves underwent a 10-minute boiling treatment in 95% ethanol to remove chlorophyll.

Sodium (Na) and potassium (K) contents were measured via inductively coupled plasma atomic emission spectroscopy (ICP-AES) (Zhang et al., 2019b). Roots, stems, and leaves of cotton seedlings were collected, dried, and ground to pass through a 40-mesh sieve. The homogenized samples were thoroughly mixed prior to analysis to ensure representativeness.

This study employed a population of 240 G. barbadense accessions to investigate genetic variation and construct a high-density variation map. A total of 2,983,855 high-quality SNPs were identified, which were unevenly distributed across chromosomes ( Supplementary Figure S1 , Supplementary Table S2 ). The At and Dt subgenomes contained 1,947,267 and 1,036,588 SNPs, respectively, with the At subgenome harboring approximately 1.88 times more SNPs than the Dt subgenome, which was consistent with the At subgenome being roughly twice the size of the Dt subgenome (Hu et al., 2019). The average SNP density across the genome was 1.40 SNPs/kb, with densities of 1.45 SNPs/kb and 1.32 SNPs/kb in the At and Dt subgenomes, respectively. Chromosome A07 showed the highest SNP density (4.35 SNPs/kb), followed by D10 (3.48 SNPs/kb). Conversely, A03 exhibited the lowest density (0.62 SNPs/kb), with D12 marginally higher (0.64 SNPs/kb). Annotation of SNPs using ANNOVAR revealed 29,495 non-synonymous SNPs, 16,606 synonymous SNPs, 136,073 upstream/downstream SNPs, 741 stop-gain SNPs, 112 stop-loss SNPs, and 356 splicing SNPs ( Supplementary Table S3 ). Linkage disequilibrium (LD) decay was estimated using the r² coefficient between SNPs. LD decay distances at which r² dropped to half-maximum (0.5) were approximately 3,000 kb for the whole genome, with 5,200 kb for the At subgenome and 1,300 kb for the Dt subgenome ( Supplementary Figure S2 ). The slower LD decay in the At subgenome compared to the Dt subgenome may reflect differential selection pressures during domestication.

To investigate the origin, genetic diversity, and differentiation among subpopulations in the G. barbadense population, phylogenetic tree construction, population structure analysis, and PCA were performed. Population structure is a major factor influencing GWAS results and can lead to false positives (Price et al., 2006). Therefore, PCA and kinship ( Supplementary Figure S3 ) matrices were incorporated to correct for population stratification and reduce spurious associations. A neighbor-joining (NJ) phylogenetic tree divided the population into four subgroups, including G1 (20 accessions), G2 (73 accessions), G3 (36 accessions), and G4 (111 accessions) ( Figure 1A , Supplementary Table S4 ). All G1 accessions originated from regions outside Xinjiang, China. Cross-validation (CV) results showed the lowest error at K = 8, with CV errors stabilizing from K = 4 onward ( Supplementary Figure S4 ). PCA results were consistent with the phylogenetic tree, dividing the 240 accessions into four groups ( Figure 1B ). Admixture analysis at K = 4 confirmed these findings, showing that accessions from regions outside Xinjiang clustered together. Notably, G4 was distinct even at K = 2, with no admixture from other groups. This indicates low genetic diversity and limited hybridization, likely due to independent artificial selection during breeding ( Figure 1C ). The F _ST and π ratios were calculated for each subgroup using SNP data to assess the differences among groups. The F _ST value between G1 and G2 was the highest (0.39), while the F _ST value between G2 and G3 was the lowest (0.08), suggesting frequent genetic exchange between G2 and G3 during breeding process. G2 exhibited the highest nucleotide diversity (π = 4.7×10^-4), indicating greater genetic resources, whereas G4 had the lowest π (1.6×10^-4) ( Figure 1D ). These results highlight significant differences among the four subgroups, with G4 showing the lowest genetic diversity and experiencing the strongest selection pressure, which provides valuable insights into the breeding history of G. barbadense.

Data of four traits were collected from 239 G. barbadense accessions under 200 mmol/L NaCl salt stress in both 2022 and 2023. The traits included relative plant height (RPH), relative shoot fresh weight (RSFW), relative shoot dry weight (RSDW), and relative root dry weight (RRDW). To minimize the influence of environmental and batch effects, the best linear unbiased estimates (BLUEs) for each trait were calculated ( Table 1 ). BLUE-adjusted trait values showed differential sensitivity to salinity: RPH (0.49-0.79, mean ± SD = 0.62 ± 0.05), RSFW (0.27-0.79, 0.49 ± 0.09), RSDW (0.35-0.88, 0.58 ± 0.10), RRDW (0.35-0.85, 0.59 ± 0.08). All four metrics were less than 1, indicating that the 200 mmol/L NaCl treatment reduced plant height, shoot fresh weight, shoot dry weight, and root dry weight during the seedling stage of G. barbadense. Among the traits, the coefficient of variation (CV) of RPH was the smallest (7.99%) and RSFW was the largest (18.42%).

All four traits exhibited a normal distribution across replicates under salt stress, indicating their suitability for GWAS analysis. These traits showed significant positive correlations between each other ( Figure 2A ). Among them, the strongest correlation was observed between RSFW and RSDW (r = 0.80), while the weakest correlation was found between RPH and RRDW (r = 0.16). In terms of RRDW, G3 showed significantly higher values than G2, and G1 performed significantly better than G2 ( Figure 2B ). For the other three traits, no significant differences were observed among the subgroups ( Figures 2C-E ). Therefore, under salt stress, the materials in subgroup G3 demonstrated better salt tolerance.

PCA was conducted on the four relative salt-tolerance indices of the 239 G. barbadense accessions. The eigenvalues and contribution rates of each principal component are shown in Table 2 . Variance partitioning showed PC1 accounted for 66.98% (λ = 2.68), PC2 21.96% (λ = 0.88), PC3 6.53% (λ = 0.26), and PC4 4.53% (λ=0.18) of total variance. The cumulative contribution rate of the first two principal components reached 88.94%, which exceeds the threshold of 85% for selecting principal components. Therefore, the first two principal components can adequately represent the information from the original four traits. PC1 (66.98% variance) showed strongest positive loading for RSDW (loading = 0.35), followed by RSFW (0.34) and RPH (0.31). PC2 (21.96% variance) was predominantly loaded by RRDW (loading = 0.94) with negative correlation to RPH (-0.47) and RSFW (-0.17).

Based on the contribution rates of the PC1 and PC2, the comprehensive salt tolerance index (D value) for the 239 G. barbadense accessions was calculated using membership function analysis. Hierarchical clustering was then performed to classify the accessions into five distinct groups ( Figure 3 , Supplementary Table S4 ): 23 highly salt-tolerant, 42 moderately salt-tolerant, 110 intermediate, 39 moderately salt-sensitive, and 25 highly salt-sensitive accessions, labeled as Groups I to V, respectively. The corresponding D value ranges for the five groups: I (0.63–0.91), II (0.52–0.61), III (0.38–0.51), IV (0.28-0.37), V (0.09-0.28). Group III (intermediate tolerance) comprised 46.0% of the panel (110/239), whereas Group I (high tolerance) represented only 9.6% (23/239), reflecting the polygenic nature of salt tolerance. This classification method facilitates the screening and identification of salt-tolerant G. barbadense accessions (Group I), providing a foundation for developing salt-tolerant G. barbadense cultivars.

Association analysis of four traits under two environments and breeding values was performed using the Mixed Linear Model (MLM). Based on the significance thresholds of p < 4.67×10^-6 [1/n, where n = 213,990 effective SNPs calculated using PLINK software (Yu et al., 2021)] for SNPs detected in a single environment and p < 1×10^–5 for repeatedly identifing SNPs, a total of 1,577 SNP loci were identified ( Supplementary Table S5 , Supplementary Figure S5 ). Among these, 34 SNPs were uniquely detected in single environments, while 1,543 SNPs were repeatedly identified across two or more environments or traits. Trait-specific associations showed varying genetic architectures: RPH (1,453 SNPs), RSFW (1,407), RSDW (41), RRDW (50). SNP distribution showed significant subgenome bias, with 90.8% (1,433/1,577) localized to the At subgenome versus 9.2% (144/1,577) in Dt. This disparity may be attributed to the higher linkage disequilibrium (LD) decay rate and larger LD blocks observed in the At subgenome ( Supplementary Figure S2 ). LD decay intervals were employed to refine candidate gene selection. By defining 100-kb genomic regions upstream and downstream of significant SNPs as LD blocks (with overlapping regions merged), we identified 132 salt stress-related QTLs ( Supplementary Table S6 ) spanning approximately 47.78 Mb collectively. The At subgenome containing 81.2% (38.80/47.78 Mb) of QTL regions versus 20.9% (9.98 Mb) in Dt. These QTL regions, representing ~2.3% of the total genome length, encompassed 811 annotated genes ( Supplementary Table S7 ).

GWAS of 2022-RPH and BLUE-RPH identified a significant SNP cluster on chromosome D02 ( Figure 4 ), delineating the QTL-SALT98 locus (Gbar_D02: 45.56-46.52 Mb; 958 kb interval) spanning approximately 958 kb. The lead SNP (Gbar_D02_45674375) marked the association peak. Within QTL-SALT98, 1,596 SNPs were subjected to linkage disequilibrium (LD) analysis using LDBlockShow, which demonstrated strong linkage disequilibrium across this genomic region ( Figure 5A ). Haplotype analysis partitioned accessions into two major haplotypes (Hap1 and Hap2) ( Figure 5B ). Notably, Hap1 exhibited significantly higher RPH values compared to Hap2 ( Figure 5C ), and Hap1 also showed significantly higher RSFW, RSDW, and D value than Hap2 ( Supplementary Figure S6 ), indicating a robust association between haplotype variation and salt tolerance capacity.

The QTL-SALT98 interval harbors 17 candidate genes. Analysis of published RNA-seq data from salt-stressed G. barbadense revealed dynamic expression patterns of these genes across 6 time points [1 h, 3 h, 6 h, 12 h, 24 h, and 14 days post-treatment (Hu et al., 2019; Dong et al., 2022)]. Notably, only five genes: Gbar_D02G014580, Gbar_D02G014590, Gbar_D02G014610, Gbar_D02G014670, and Gbar_D02G014700 exhibited significant differential expression under salt stress ( Figure 5D ).

To validate candidate gene expression in extreme phenotypic materials, we selected the salt-tolerant genotype H160 and salt-sensitive genotype H20. Physiological characterization revealed contrasts in their salt tolerance. Under 7-day salt stress, H160 exhibited less structural alterations with mostly upright stems and partially turgid leaves, showing only slight wilting compared to controls, whereas H20 displayed severe wilting and pronounced stem bending ( Supplementary Figures S7A, B ). Growth parameters, including plant height, shoot biomass, and root dry weight, experienced moderate decreases in H160, in contrast to the significant declines observed in H20 under salinity ( Supplementary Figures S7C-F ). Ion profiling showed H160 roots retained 40% higher K⁺ content with a lower Na⁺/K⁺ ratio than H20, while its leaves maintained 40% lower Na⁺ accumulation and 45% reduced Na⁺/K⁺ ratio ( Supplementary Figures S7 G-I ), demonstrating coordinated regulation of Na⁺ exclusion in leaves and K⁺ retention in roots. These ion balance results indicate that H160 achieves salt tolerance by coordinating Na⁺ exclusion in leaves and K⁺ retention in roots, maintaining cellular ionic homeostasis critical for osmotic balance and enzyme function under salinity. In contrast, H20 failures to restrict Na⁺ accumulation in leaves and preserves root K⁺ levels, which leads to disrupted ion homeostasis and results in severe growth inhibition and wilted phenotype.

Given these physiological disparities, we performed qRT-PCR validation at 0 h, 24 h, and 48 h post-treatment. Results demonstrated that Gbar_D02G014580 and Gbar_D02G014610 showed higher expression in H20 at 48 h, while Gbar_D02G014590 exhibited elevated expression in H20 at 24 h ( Figures 6A-C ). Conversely, Gbar_D02G014670 expression in H160 significantly surpassed that in H20 at both 24 h and 48 h, with the most pronounced difference observed at 48 h ( Figure 6D ). In addition, there was no significant difference in Gbar_D02G014700 at each time ( Figure 6E ). These findings implicate Gbar_D02G014590, Gbar_D02G014610, Gbar_D02G014670, and Gbar_D02G014700 as salt-responsive candidate genes. Functional annotation of Arabidopsis homologs revealed that Gbar_D02G014670 (AT2G01850, GbXTH27) encodes a xyloglucan endotransglucosylase/hydrolase (XTH), which was critical for cell wall remodeling, while Gbar_D02G014590 (AT1G16020, GbCCZ1A) encodes a vacuolar fusion protein, as a component of the MON1–CCZ1 complex, CCZ1A regulates post-Golgi vesicle transport to ensure targeted transport of storage proteins to protein storage vacuoles. CCZ1A dysfunction leads to seed development defects (Pan et al., 2021). Gbar_D02G014610 (AT1G24706, GbTHO2) is a core component of the THO/TREX complex, which was essential for miRNA biogenesis. Gbar_D02G014700 (AT1G14710) encodes a hydroxyproline-rich glycoprotein family protein. Gbar_D02G014580 (AT1G68020, GbTPS6) encodes a trehalose-6-phosphatase catalyzing trehalose-6-phosphate (T6P) biosynthesis. T6P regulates sucrose biosynthesis, source-sink allocation, and developmental signaling in plants (Fichtner and Lunn, 2021).

VIGS of GbXTH27 in the G. barbadense standard line 3–79 under salt stress (200 mM NaCl) were performed to verify its gene function. VIGS resulted in white-leaf phenotype ( Supplementary Figure S8A ) and significant transcript reduction ( Supplementary Figure S8B ). After 7 days of salt stress treatment, silenced plants (pTRV2: GbXTH27) displayed exacerbated wilting in cotyledons and true leaves compared to controls (TRV:00) ( Figures 7A, B ). Under salt stress conditions, TRV: GbXTH27 showed significantly lower values in plant height, shoot fresh weight, shoot dry weight, and root dry weight than TRV:00 controls ( Supplementary Figures S9A-D ). Additionally, Na⁺ content in the roots, shoots, and leaves of TRV: GbXTH27 was significantly higher than that in TRV:00 ( Supplementary Figure S9E ), while K⁺ content in the shoots and leaves of TRV: GbXTH27 was significantly lower than in TRV:00 ( Supplementary Figure S9F ). Consequently, the Na⁺/K⁺ ratio in the roots, shoots, and leaves of TRV: GbXTH27 was significantly higher than in TRV:00 ( Supplementary Figure S9G ), indicating that TRV: GbXTH27 experienced more severe salt stress. After 7 days of NaCl treatment, the leaf DAB staining area of TRV: GbXTH27 plants was significantly larger than that of TRV:00 controls, with deeper staining intensity in TRV: GbXTH27( Figure 7C ). This indicates that silencing GbXTH27 leads to a significant increase in reactive oxygen species (ROS) accumulation in leaves, significantly reducing the salt stress resistance of cotton seedlings. Physiological assays revealed diminished SOD activity and Pro content, alongside elevated MDA levels in silenced plants ( Figures 7D-F ), indicating compromised antioxidant capacity, membrane integrity, and osmotic adjustment. These results confirm GbXTH27 as a key regulator of salt tolerance. The peak SNP (Gbar_D02_45674375) within QTL-SALT98 serves as a molecular marker for breeding salt-tolerant cotton cultivars.