In the year 2022, five distinct wheat cultivars—Cangmai 6002 (C6002), Cangmai 6005 (C6005), Jimai 22 (JM22), Shiluan 02-1 (SL02-1), and Malannuomai NO.1 (N01)—were collected from two locations in Cangzhou, Hebei Province, China, namely Xi Huayuan (XHY) and Zhong Jie village (ZJ), which are 50 km apart in a straight line, with basically the same climate conditions but different longitude and latitude and soil salinity and alkalinity (Table 1). Among these cultivars, C6002 and C6005 are medium gluten wheat varieties with moderate tolerance to salinity-alkalinity, whereas JM22, SL02-1 and N01 are categorized as medium-gluten, strong-gluten, and waxy wheat varieties, respectively, all of which are sensitive to salinity-alkalinity stress, and are the main cultivated varieties in the local area (Supplementary Table 1). The five wheat varieties were assigned randomly to plots in each location following a randomized complete block design, with each plot covering an area of 666.67 m² (1 mu).

After the wheat had fully matured, they were collected using an Austrian Wintersteiger classic combine harvester. From each sample, approximately 5 kilograms were randomly selected and labeled. Then, three separate 100-gram portions were drawn at random from each 5-kilogram sample. A total of 30 samples were obtained (2 fields × 5 genotypes × 3 replicates), and all were processed using a versatile grinder (model ZK-100B, manufactured by Zhongke Haoyu Technology Co., Ltd., China). After processing, the samples were placed in sealed bags and kept at -20°C for subsequent use.

The extraction and analysis of VOCs was conducted according to the published experimental scheme, with some improvements (22). The headspace solid-phase microextraction (HS-SPME) technique was employed to extract VOCs. Wheat wholemeal samples (2 g) were placed in 20 mL vials, equilibrated at 60°C for 40 min, and extracted with HS-SPME for 80 min at 60°C. The extracted fiber was 50/30 μm divinylbenzene/carboxenTM/polydimethylsiloxane (DVB/CAR/PDMS) from Supelco (Bellefonte, PA, United States).

Analysis of the VOCs was performed using GC-MS (TRACE1300, ISQ9000, United States), employing a DB-Wax capillary column (30 m × 0.25 mm × 0.25 μm). Post-extraction, the fiber was desorbed into the GC injector port for 2 min at 250°C in splitless mode. Helium at a flow rate of 1 mL/min was used as the carrier gas. The oven temperature was initially set at 40°C for 5 min, then ramped at 5°C/min to 220°C for 10 min. The MS was operated in electron ionization mode (EI) at 70 eV, with a multiplier voltage of 1247 V. The ion source temperature was maintained at 200°C, and the interface temperature at 250°C. Full scan mode was employed to detect VOCs across a mass range from m/z 35 to 600. Each sample underwent triplicate analysis.

Compounds identification relied on comparison of their mass spectra with those in the NIST 20 library. VOCs with Match Factor greater than 750. Additionally, retention index was calculated using Kovats retention index based on a series of straight-chain alkanes (C7-C27) under the aforementioned chromatographic conditions.

To confirm GC-MS identifications (Table 2), the standards (reference compounds) of 2-pentyl-furan (3777-69-3, ≥ 98%), 1-hexanol (111-27-3, ≥ 99%), 1-octen-3-ol (3391-86-4, ≥ 98.04%) and limonene (138-86-3, 1,000 μg/mL) were selected. These standards were diluted and mixed with high-quality chromatographic grade methanol (99.9%). A stock solution containing a concentration of 50 ppm for the mixed standards was prepared. The sources of 2-pentyl-furan, 1-hexanol, 1-octen-3-ol and methanol were Shanghai Macklin Biochemical Co., Ltd., China, while the source of Limonene was Tan-Mo Technology Co., Ltd., China.

An aliquot of 1 mL from the above-prepared mixed standard sample stock solution was extracted into a headspace bottle with a volume of 20 mL. Then it was analyzed by HS-SPME-GC-MS using the same conditions as the previous tests. The reliability of VOCs identification results in this study can be considered satisfactory, if the retention times for the aforementioned reference compounds in both standard and test samples were essentially identical. The confirmation results are presented in Supplementary Figure 1.

The method was adapted from to Jia et al. (23) with minor modifications. This study investigated interactions between sixteen significant human olfactory receptors (OR51E1, OR52D1, OR1D2, OR7C1, OR8B3, OR6F1, OR13C3, OR10A6, OR5P3, OR3A1, OR14A2, OR9G1, OR11H6, OR2W1, OR11H4, and OR4Q3) and impotent VOCs. Protein structures for the olfactory receptors were sourced from UniProt and AlphaFold. SDF files for each small molecule ligand were obtained from the PubChem database and converted to PDB format using PyMOL 2.3.0. Molecular docking was performed using AutoDockTools (ADT) v1.5.7 and AutoDock Vina v1.2.0, prioritizing receptor-ligand complexes prioritized based on low binding energy rankings. The results were further analyzed using PyMOL and LigPlot for interaction visualization.

The ROAV was a very useful tool for assessing the contribution of each aroma compound to the overall flavor profile (24). The larger the ROAV, the greater the contribution to the overall flavor of the sample.

The ROAVs. of each component were calculated to analyze its contribution of each component to the overall wheat aroma. The formula was as follows:

ROAV = 100 × C%_A/C%_max × T_max/T_A

Where, C%_A and T_A were the relative percentage content and odor threshold concentration (Table 3) of the compound, respectively. C%_max and T_max were the relative percentage content and odor threshold concentration (Table 3) of the component that contributes the most to the overall flavor of the sample, respectively.

All experiments were conducted in triplicates and the data are presented as the mean ± standard deviation. The data treatment was carried out using the SPSS 23.0 package software (SPSS Inc., Chicago, IL, United States). T-tests were performed to analyze significant differences in VOCs of wheat varieties cultivated in various cultivation sites. The bar graph and heat map were created using Origin 2024 software (OriginLab Corporation, Northampton, MA). Principal component analysis (PCA) and linear discriminant analysis (LDA) were utilized for data pattern identification and analysis.

The composition and contents of VOCs in five wheat varieties cultivated in SA and control soils were compared by GC-MS. A total of 75 VOCs were identified and classified into 11 chemical classes namely: alcohols, hydrocarbons, terpenes, aromatic compounds, acids, ethers, esters, aldehydes, ketones and heterocyclic compounds. Calculate the proportion of the peak area of each type of substance in all substances in each sample (Table 2). Among these classes, alcohols, alkanes, terpenes, aromatic compounds, acids, esters, aldehydes and the remaining substances accounted for 30.05–36.13%, 22.59–26.55%, 0.81–16.41%, 9.39–15.73%, 5.56–9.08%, 3.11–3.79%, 2.26–4.63%, and 5.53–12.52% of the total VOCs in the wheat varieties, respectively.

Paired t-tests demonstrated significant alterations in the volatile profiles of saline-alkali (SA) soil-cultivated wheat versus control samples (Figure 1A). Under SA cultivation, seven compounds exhibited significantly elevated peak areas (Supplementary Table 2): β-pinene showing the most pronounced increase (525.91%, p < 0.05), followed by limonene (121.15%, p < 0.01), 1-(1,5-dimethyl-4-hexenyl)-4-methyl-benzene (35.28%, p < 0.05), lavender lactone (35.71%, p < 0.05), 1,6-dimethyl-naphthalene (38.82%, p < 0.01), 1,8-dimethyl-naphthalene (13.78%, p < 0.05), and tetradecane (4.72%, p < 0.05). In contrast, the levels of (E,E)-3,5-octadien-2-one and butanoic acid were significantly decreased (Supplementary Table 2) by 16.05% (p < 0.05) and 38.71% (p < 0.01), respectively. It was demonstrated SA stress induces metabolic reprograming that favors the biosynthesis of terpenoid and aromatic compounds.

To better characterize VOCs changes across five wheat varieties cultivated in soils with different SA levels, the relative contents of key VOCs were measured, and the results are shown in Figure 1Bβ-pinene and limonene exhibited marked increases (Supplementary Table 3) by 474.09% (p < 0.05) and 109.74% (p < 0.01), respectively, confirming their roles as SA stress-responsive biomarkers. Conversely, tridecane, (E,E)-3,5-octadien-2-one and butanoic acid exhibited significant reductions (Supplementary Table 3): 10.17% (p < 0.05), and 42.95% (p < 0.01), respectively.

Based on the pairing results of peak areas and relative contents, β-pinene, limonene, (E,E)-3,5-octadien-2-one and butanoic acid could be regarded as important indicator compounds for wheat cultivation in SA soil. Given that wheat cultivated on SA soil has a relatively strong aroma, these compounds might also be the ones that contribute to the wheat aroma.

To further investigate whether these four VOCs act as critical compounds for wheat cultivated in saline-alkali land, molecular docking was employed. In the docking simulations, a negative binding energy indicates spontaneous ligand-receptor binding, with its magnitude correlating directly to binding affinity. More negative values reflect stronger intermolecular interactions and higher probability of stable complex formation (25). Based on established criteria: binding energy < -4.25 kcal/mol suggests detectable binding activity, implying potential ligand-receptor recognition; binding energy < -5.0 kcal/mol signifies high binding affinity, likely corresponding to biologically relevant interactions (26). For this study, compounds meeting the threshold of binding energy < -4.25 kcal/mol were prioritized for further analysis. The binding energy heat map between the important compounds and the olfactory receptors (Figure 2) revealed that binding energy range of butanoic acid with all 16 olfactory receptors was -4.79 to -3.82 kcal/mol, and its affinity is relatively weak (> -5 kcal/mol), whereas limonene, β-pinene and (E,E)-3,5-octadien-2-one exhibited moderate to strong binding affinity, with binding energies of -6.02 to -4.64 kcal/mol, -6.19 to -4.39 kcal/mol, and -6.17 to -4.29 kcal/mol, respectively. These results align with prior evidence that such binding energies modulate receptor conformational dynamics and enhance agonist efficacy in flavor perception (27).

Molecular docking studies have demonstrated that aroma compounds enhance food flavor by binding to potential active sites of ORs (28). Among the sixteen ORs examined, OR11H4 exhibited significantly stronger binding affinity (Figure 2), and was therefore selected to elucidate mutual binding interactions with three key compounds. As shown in Figure 3, fourteen interacting amino acid binding residues (Leu217, Ser213, Tyr269, Tyr83, Tyr166, Tyr288, Val265, Val216, Phe114, Phe115, Phe261, Gly118, Thr119, and Cys122) were identified in the OR11H4 binding porket. Notably, Leu217, Ser213, Tyr269, Phe115, and Thr119 served as common binding sites, suggesting their suggesting their critical role as key interaction residues (23). Furthermore, limonene formed hydrophobic interactions with Leu217, Ser213, Tyr269, Tyr83, Val265, Val216, Phe114, Phe115, Gly118, and Thr119; β-pinene bound via hydrophobic interactions to Leu217, Ser213, Tyr269, Tyr166, Tyr288, Val265, Phe115, Phe261, Gly118, and Thr119; while (E,E)-3,5-octadien-2-one established hydrogen bonds with Tyr269 and Tyr288, alongside hydrophobic interactions involving Leu217, Ser213, Tyr83, Tyr166, Val216, Phe114, Phe115, Thr119, and Cys122.

In summary, Leu217, Ser213, Tyr269, Phe115, and Thr119 are critical binding residues, with hydrophobic interactions and hydrogen bonds acting as primary driving forces in ligand-receptor recognition.

The characteristic “wheat aroma” is influenced not only by the concentration of key odorants but also by their sensory threshold. Relative odor activity values (ROAVs) could be utilized for assessing the contribution of individual components toward the overall flavor (24). Based on established criteria (29), VOCs with ROAVs. > 1 dominate the overall flavor, while those with ROAVs between 0.1 and 1 exert modifying effects; VOCs with ROAVs. < 0.1 contribute negligibly. To identify key aroma compounds in the SA soil-cultivated wheat, ROAVs were calculated for limonene, β-pinene, and (E,E)-3,5-octadien-2-one. Consistently, ROAVs exceeded 1 across five wheat varieties grown in soils with distinct salinity-alkalinity gradients (Table 3), indicating their significant contribution to the characteristic flavor of SA-cultivated wheat.

Based on this premise, the three VOCs were considered as key flavor compounds that potentially contribute significantly to the overall flavor perception of “wheat aroma.” The proportion range of each component can be determined by calculating the percentage of peak area for each key flavor compound among all key flavor compounds, as presented in Supplementary Table 4.

Principal component analysis (PCA), an unsupervised dimensionality reduction technique, extracts orthogonal components that maximize variance retention while approximating original variables. This method effectively identifies inter-group similarities and intra-group differences within complex datasets (30). PCA conducted on the three potential key aroma compounds revealed that the first three principal components (PC1-PC3) accounted for 67.45, 22.09, and 10.46% of total variance, respectively, with a cumulative variance contribution rate was 100% (Figure 4A). This confirms PC1-PC3 comprehensively capture the dominant variation patterns in the aroma profile. Loadings analysis further identified limonene, β-pinene and (E,E)-3,5-octadien-2-one could serve as reliable indicators for distinguishing XHY and ZJ cultivated wheat.

Complementarily, linear discriminant analysis (LDA)—a supervised technique optimizing inter-class separation while minimizing intra-class variance (31)—was employed to validate the classification efficacy of these three VOCs. Notably, LDA achieved 90.0% cross-validated accuracy in distinguishing wheat grown in distinct salinity-alkalinity soils. Projection plots (Figure 4B) demonstrated complete spatial segregation between XHY and ZJ samples along the first three discriminant functions (LD1-LD3), which collectively explained 100% of total variance. LD1 dominated class separation (67.16% variance), with limonene exhibiting the highest standardized coefficient (discriminant coefficient = 0.73), indicating its pronounced sensitivity to salinity gradients.

Synthesizing PCA and LDA outcomes, limonene, β-pinene, and (E,E)-3,5-octadien-2-one are established as signature flavor compounds defining the aroma signature of SA-cultivated wheat, and may also be the characteristic biomarkers of wheat aroma.