Biomass and volatile profile responses of hydroponically grown Basil (Ocimum basilicum L.) to different nutrient solution regimes

Hydroponics is a soilless cultivation technique that has gained increasing interest in its capacity to address specific constraints of conventional agriculture, notably through the precise management of mineral nutrition and growing conditions. This study evaluated the effects of different nutrient solution regimes defined by nitrogen (N) and phosphorus (P) concentrations and covarying potassium (K) concentrations on basil (Ocimum basilicum L.) cultivated in a hydroponic nutrient film technique (NFT) system. Agronomic parameters and volatile organic compounds (VOCs) were analyzed at final harvest, while response surface methodology (RSM) was applied to identify model-predicted candidate nutrient combinations associated with the production of linalool, eugenol, and eucalyptol. Nine nutrient treatments varying in N and P concentrations were evaluated, while K was considered a dependent nutrient, with Hoagland nutrient solution (HNS) serving as the reference treatment. High N treatment produced the highest fresh aerial weights, reaching 430 g, 408 g, and 365 g, values close to those obtained with the HNS (471 g). Intermediate N treatments showed fresh weight values of 154 g, 254 g, and 210 g, whereas low N treatments resulted in fresh aerial weights below 100 g. Aerial dry weight, leaf number, leaf area, and plant height followed the same trend. In contrast, root parameters remained generally similar among treatments. An exploratory principal component analysis (PCA) suggested three compositional groupings: high N associated with eugenol, intermediate N with eucalyptol, and low N with linalool, with an intermediate profile for HNS. In-sample quadratic model fits yielded R 2 values of 0.64 for linalool, 0.82 for eugenol, and 0.73 for eucalyptol. The model-predicted candidate combinations of N, P, and associated K were 62.2, 17.1, and 80 mg L −1 for linalool; 105, 31, and 135 mg L −1 for eugenol; and 58.2, 31, and 93.5 mg L −1 for eucalyptol, respectively. These conditions increased the compound production indices, calculated as aerial fresh weight × relative GC peak area (%)/100, compared with the HNS reference treatment, from 8.46 to 17.43 for linalool, from 5.61 to 14.7 for eugenol, and from 1.29 to 4.10 for eucalyptol. These findings suggest that nutrient solution formulation may influence basil biomass and the relative abundance of selected VOC constituents. The results also highlight the potential of nutrient formulation as a strategy for modulating specific VOC profiles. In addition, RSM may serve as an exploratory approach for identifying nutrient formulations associated with variations in selected VOC compounds.

Basil (Ocimum basilicum L.), a member of the Lamiaceae family, is among the most widely cultivated aromatic and medicinal plants (Nadeem et al., 2020). It is highly valued for its culinary, pharmaceutical, cosmetic, and industrial applications, mainly due to its abundance of VOCs such as linalool, eugenol, and eucalyptol (Vieira and Simon, 2006). These compounds are of particular interest because they contribute to the characteristic aroma of basil and are associated with several biological activities, such as antimicrobial, insecticidal, and fungicidal properties (Azizah et al., 2023). These characteristics contribute to the economic importance of basil, whose global demand continues to increase (McGrath, 2024).

Hydroponic cultivation has gained increasing attention as a soilless production technique that can help address some constraints of conventional soil-based systems, particularly under increasing pressure on water resources, limited arable land availability, and the growing need for efficient and precisely controlled production methods (Singh et al., 2023). This approach is particularly suitable for high-value crops, especially aromatic and medicinal plants (Aires, 2018). By allowing controlled nutrient delivery and optimized growing conditions, hydroponics can improve resource-use efficiency, reduce water consumption, and support uniform crop production (Rajaseger et al., 2023). In this context, the nutrient solution plays a key role in determining crop responses, as its composition directly influences nutrient uptake, vegetative growth, biomass accumulation, and metabolic pathways in aromatic plants, including basil (Ren et al., 2022; Fathidarehnijeh et al., 2024). Despite its recognized importance, most studies on basil cultivated in hydroponic or soilless systems have primarily focused on growth, yield, mineral nutrition, and agronomic performance (Saha et al., 2016; Walters and Currey, 2018; Khater et al., 2021). A limited number of recent studies have shown that the composition and concentration of hydroponic nutrient solutions can influence basil quality and essential oil composition, including major VOCs such as linalool, eugenol, and eucalyptol (Ciriello et al., 2020; Hazrati et al., 2024). Furthermore, modeling approaches in basil research have mainly focused on growth, nutrient uptake or consumption, and extraction process optimization, while predictive approaches linking nutrient solution composition to specific VOC profiles remain limited (Tran et al., 2018; Aghamirzaei et al., 2024). RSM is a statistical modeling approach used to evaluate relationships between multiple experimental factors and measured responses and to identify factor combinations associated with desired outcomes. It has been increasingly applied in plant and agricultural research to investigate interactions among experimental factors and support process optimization (Mnayer et al., 2017; Zhang et al., 2024). To the best of our knowledge, no previous study has applied RSM to predict nutrient solution composition in relation to the VOC profile of hydroponically grown basil. Therefore, the present study aimed to evaluate the effects of nine N–P nutrient combinations on the agronomic performance and selected VOC constituents of hydroponically grown basil. In addition, RSM was applied as an exploratory modeling approach to characterize nutrient–response relationships and identify nutrient combinations associated with higher indices of linalool, eugenol, and eucalyptol production.

Study location and components

The study was conducted from 25 December 2025 to 26 February 2026 in a soilless cultivation greenhouse at TERRA Teaching and Research Center, Gembloux Agro-Bio-Tech, University of Liège (Liège, Belgium). Basil was grown under ten hydroponic nutrient regimes to evaluate their effects on agronomic traits and VOC profiles.

Hydroponic system configuration

The hydroponic experiment was conducted in a greenhouse where temperature and ventilation were managed by aerothermal heating units and automated windows controlled according to computer-defined setpoints (Figure 1A). Plant growth was supported entirely by artificial lighting under a fixed photoperiod of 16 h light and 8 h dark. Lighting was provided using VEGELED PANDORA PLUS LB302 horticultural floodlights (Pandora LB302, spectrum M1FR; reference LB302XHWX120M1FRH). The lamps had a power consumption of 300–325 W, a circular beam angle of 120°, and were dimmable through two 0–10 V channels. According to the manufacturer’s spectral data, the emitted spectrum consisted of 0.20% UV (380–399 nm), 15.2% blue (400–499 nm), 32.9% green (500–599 nm), 43.20% red (600–699 nm), and 8.50% far-red radiation (700–780 nm), with 91.30% of the emission falling within the PAR range (400–700 nm). This fixed lighting strategy was applied throughout the experimental cycle to reduce environmental variability and maintain comparable growth conditions among the experimental units. Greenhouse climatic conditions were monitored continuously during the study. Air temperature (T air) and relative humidity (RH) were measured using a Vaisala HUMICAP HMP110 sensor, whereas atmospheric carbon dioxide (CO 2) concentration was measured using a separate Vaisala GMP252 sensor. Both sensors were connected to a Campbell Scientific datalogger. The acquisition system enabled automated recording of microclimatic data every 15 min, allowing continuous monitoring of the environmental conditions surrounding the crop.

The cultivation system consisted of ten independent NFT units equipped with Goponic ® gutters (Agri Logic System, Normandy, France) (Figure 1B). Each unit included five horizontal growing channels, each measuring 301 cm in length, 18 cm in width, and 14 cm in height. This arrangement provided a cultivation surface of 3.59 m 2 per unit and contained 70 planting positions, separated by 14.5 cm spacing between plants. Basil plants were established in plastic pots containing rockwool cubes supplied in GRODAN ® 77-cell trays. This inert growing support was selected to maintain the plants in position while allowing the roots to remain in regular contact with the recirculating nutrient solution. Each NFT unit was connected to an individual 100 L-capacity reservoir placed below the growing table and filled with 80 L of demineralized water. During cultivation, the nutrient solution continuously circulated through the channels in a closed-loop system using a SYNCRA pump operated at a flow rate of approximately 10 L·min −1, ensuring the formation of a thin and stable nutrient film around the roots while providing sufficient oxygenation for optimal root development. The reservoirs were equipped with sensors enabling continuous monitoring of pH, electrical conductivity (EC), and nutrient solution temperature. The pH was measured using a Milwaukee MA913B/3 Epoxy Industrial pH Replacement Probe, while EC and nutrient solution temperature were recorded using a Milwaukee MA812/2 electrode connected to the MC310 EC Monitor. Among these parameters, only pH was automatically adjusted when necessary, using peristaltic dosing pumps which injected a 2.5% (v/v) sulfuric acid solution to maintain the target pH range (5.5 - 6.5). Before the beginning of the experiments, all sensors were calibrated according to the manufacturer’s recommendations to ensure the accuracy and reliability of the recorded data.

Plant material, experimental design, and plant growth measurements

Basil seeds (Ocimum basilicum L., cv. ‘Grand Vert’, seed lot no. 003218), obtained from Vilmorin (France), were germinated in greenhouse trays containing rockwool cubes, where the irrigation water was renewed every three days. Transplanting into the experimental greenhouse was carried out 21 days after sowing. The seedlings were then transferred into ten independent NFT systems, each containing 54 plants (Figure 2). Although each NFT unit provided 70 planting positions, eight positions at each end of the growing table (16 positions in total) were intentionally left unoccupied to minimize potential edge effects. The NFT units were randomly positioned within the greenhouse and remained in the same position throughout the experimental period. Before the main experiment, a preliminary homogeneity test was conducted by cultivating the same plant material under an identical nutrient solution across all NFT units. Comparable plant performance was observed among the units, supporting the homogeneity of the experimental setup.

Each NFT system represented a single experimental unit corresponding to one nutrient treatment. The individual plants grown within each NFT system were considered subsamples of that experimental unit. After transferring to the NFT systems, the nutrient solution was completely renewed weekly by replacing the entire reservoir water and adding the required salts to restore the initial nutrient concentrations. No additional water was added between weekly renewals. Nine nutrient treatments were applied, each to a separate NFT unit (Table 1), formulated with varying concentrations of N, P, and K, prepared from mineral salts, potassium nitrate (KNO 3), monopotassium phosphate (KH 2 PO 4), and calcium nitrate tetrahydrate (Ca(NO 3) 2 ·4H 2 O), and calcium hydroxide (Ca(OH) 2). Calcium hydroxide was used as a supplementary calcium source to compensate for variations in Ca supplied by Ca(NO 3) 2 and to maintain the final Ca concentration constant across all nutrient solutions. The treatments were established according to a 3 x 3 full factorial design with three levels of N and P. The K concentration varied in response to adjustments in N and P, depending on the amounts of KNO 3 and KH 2 PO 4 used. The final concentrations of N, P, and K in each nutrient solution were calculated based on the mass of mineral salts added and the volume of water used, according to the following equations (Equations 1–3):

where m represents the mass of each salt added (g), V is the volume of water used (L), the coefficients correspond to the proportion of N, P, or K contained in each mineral salt, and the factor 1000 was used to convert concentrations from g/L to mg/L.

A tenth NFT unit received a standard HNS and served as the reference treatment. To ensure the isolation of macronutrient effects, all other mineral elements were maintained at identical concentrations across all experimental units, in accordance with HNS recommendations. At the end of the nine-week cultivation period, all basil plants were harvested in the morning between 8:00 a.m. and 12:00 p.m. Before harvest, plant height was measured using a graduated measuring tape, leaf number was counted manually, and leaf area was estimated using the Canopeo online platform (https://canopeoapp.com/). Plants were harvested by cutting the stem at the junction with the roots. Fresh shoot and root biomass were immediately determined using an electronic precision balance (± 0.01 g).

For dry biomass determination and phytochemical analyses, shoots and roots were separated, placed in labeled aluminum containers, and dried in a forced-air oven at 40 °C for five days until a constant weight was achieved. Dry shoot and root biomass were then recorded. The dried plant material was pooled and ground into a homogeneous powder. The resulting composite sample was subsequently divided into three subsamples, each of which was subjected to hexane maceration and phytochemical analysis.

Extraction and volatile organic compound profile

Extraction was performed by hexane maceration according to the method described by Azzahra et al. (2025), with minor modifications adapted to this study. Briefly, the three powdered samples from each treatment were first passed through a 50 µm sieve to obtain a uniform particle size and improve solvent extraction efficiency. Subsequently, 1.0 g of powder was mixed with 10 mL of hexane at a ratio of 1:10 (w/v) in glass vials. The mixtures were homogenized and incubated at room temperature for 72 h in the dark to minimize the photodegradation of photosensitive compounds. After maceration, the extracts were successively filtered through Whatman No. 1 filter paper and a 0.22 µm syringe filter to obtain a clear solution suitable for chromatographic analysis. The filtrates were then transferred into airtight glass vials and stored at 4 °C in the dark. The analyses of the maceration extracts were performed using gas chromatography coupled with mass spectrometry (GC–MS), following the same methodology as that described by Elakrouch et al. (2026).

Mathematical modeling procedures

After VOC analysis, three VOC compounds showing marked variation among nutrient treatments, namely linalool, eugenol, and eucalyptol, were selected for further modelling. To integrate shoot biomass with GC–MS relative abundance, an exploratory compound index (CI) was calculated for each selected molecule using Equation 4:

CI: compound index.

RSM was then applied to investigate the relationship between nutrient supply and the production of the selected VOC compounds. For each molecule, a second-order polynomial regression model was fitted separately, using CI as the response variable (Y) and the coded concentrations of N (Nc) and P (Pc) as explanatory variables. The general form of the model was expressed as follows (Equation 5):

where Y represents the response variable, corresponding to the CI of each selected VOC compound; β 0 is the intercept; β 1 and β 2 represent the linear effects of N and P, respectively; β 11 and β 22 represent the quadratic effects of N and P; β 12 represents the interaction effect between N and P; Nc and Pc are the coded concentrations of N and P, respectively; and ϵ is the residual error term.

This modelling approach was used to explore the linear, quadratic, and interactive associations of N and P with the CI of linalool, eugenol, and eucalyptol, and to identify model-predicted candidate nutrient combinations associated with higher fitted CI values. Given the absence of independent replication at the NFT-unit level, the quadratic response-surface models were used as exploratory and descriptive tools rather than as confirmatory optimization models. The fitted surfaces were intended to characterize response patterns within the tested nutrient domain and to identify candidate nutrient formulations requiring subsequent independent experimental validation. Model performance was evaluated using the coefficient of determination (R 2) and root mean square error (RMSE).

Statistical analysis

All statistical analyses and RSM were performed using Python software (version 3.12). Descriptive statistics were used to characterize the cultivation conditions, including T air, water temperature (T wat), pH, EC, CO 2 concentration, and RH. Agronomic variables, including plant fresh weight, leaf number, leaf area, and related parameters, were expressed as mean calculated from individual plants within each NFT unit. These plants were considered subsamples of the corresponding treatment, and the resulting analyses were treated as descriptive and exploratory. For VOCs, the three subsamples were considered analytical replicates. VOC data were Pareto-scaled before PCA, which was used as an exploratory approach to visualize variation in volatile profiles among treatments. Potential outliers in the indices of linalool, eugenol, and eucalyptol were screened using standardized Z-scores, with |Z| > 3 used as the threshold for identifying potentially aberrant observations.

Detailed information on cultivation conditions during the experimental period (Tair, RH, CO2 concentration, Twat, EC, and pH) is presented in the Supplementary Appendix Figures A1, A2. All basil plants were cultivated under the same greenhouse conditions across the ten NFT units. Daytime temperature ranged approximately from 20 to 27 °C, with slightly lower values recorded at night. RH varied between approximately 22% and 65% and showed an inverse trend with Tair. CO 2 concentration remained relatively stable, with only minor fluctuations. In the hydroponic reservoirs, the monitored parameters were regularly controlled. The target pH range was 5.5–6.5; however, transient deviations outside this range were observed in some NFT units. Twat remained relatively stable at approximately 22 °C, whereas EC varied among nutrient solution regimes, reflecting the differences in nutrient solution composition among treatments.

Growth and agronomic performance

For descriptive purposes, the nine treatments were grouped into three N-level categories (S1–S3, S4–S6, and S7–S9), as treatments within each category displayed comparable agronomic responses and overall trends. These grouped means were used only to summarize the observed patterns. At harvest, agronomic parameters differed descriptively among the nutrient regimes (Figure 3). Plant height reached an average of 87.9 cm in S1 to S3, compared with 97.3 cm under the HNS reference regime. S4 to S6 showed intermediate values, with an average plant height of 81.1 cm, whereas S7 to S9 exhibited markedly lower values, averaging 38.7 cm. Fresh shoot weight was also higher in S1 to S3, with an average of 401 g plant −1, compared with 471 g plant −1 under the HNS reference regime. S4 to S6 presented intermediate values, averaging 206 g plant −1, whereas S7 to S9 recorded the lowest aerial biomass, with an average of 45.6 g plant −1. A similar trend was observed for shoot dry weight, with mean values of 36.1 g plant −1 in S1 to S3, 42.4 g plant −1 under the HNS reference regime, 18.5 g plant −1 in S4 to S6, and 4.13 g plant −1 in S7 to S9. Leaf number followed the same pattern. S1 to S3 exhibited an average of 757 leaves plant −1, close to the value observed under the HNS reference regime (820 leaves plant −1). S4 to S6 showed intermediate values, with an average of 438 leaves plant −1, whereas S7 to S9 recorded the lowest values, averaging 112 leaves plant −1. Leaf area reached an average of 3217 cm 2 plant −1 in S1 to S3, compared with 3373 cm 2 plant −1 under the HNS reference regime. S4 to S6 showed an intermediate mean leaf area of 2258 cm 2 plant −1, whereas S7 to S9 exhibited the lowest values, with an average of 662 cm 2 plant −1. Finally, root fresh and dry weights showed more limited variation among the different nutrient-regime groups. Mean root fresh weight values were 83.6 g plant −1 in S1 to S3, 64.8 g plant −1 under the HNS reference regime, 65.9 g plant −1 in S4 to S6, and 33.6 g plant −1 in S7 to S9. Mean root dry weight values were 4.1 g plant −1, 3.1 g plant −1, 3.1 g plant −1, and 1.8 g plant −1, respectively.

Volatile organic compound profile and compound index of linalool, eugenol, and eucalyptol

The VOC profiles of nutrient solution regimes were characterized through the identification and semi-quantification of volatile compounds (Figure 4). Exploratory PCA suggested a structuring of the treatment profiles according to their phytochemical composition. The first two principal components explained 67.1% of the total variance, with PC1 accounting for 45.7% and PC2 for 21.4%. The distribution of the nutrient regimes in the PCA space showed separation associated with the correlated N, P, and K concentrations. Nutrient regimes S1, S2, and S3, characterized by high N (with associated K) inputs and variable P concentrations, were positioned on the positive side of PC1. These regimes were mainly associated with eugenol, 3-hexenol, α-cubebene, 2-hexanone, and 3-hexanone. Conversely, S7, S8, and S9, characterized by the lowest N and K inputs and variable P concentrations, were located on the negative side of PC1 and were strongly associated with linalool, β -myrcene, spathulenol, methyleugenol, camphor, and α-cedrene. S4, S5, and S6, characterized by intermediate N and K inputs, occupied an intermediate position, mainly along the PC2 axis. These regimes were associated with eucalyptol, neophytadiene, limonene, and α-sabinene. The HNS reference treatment, corresponding to the highest concentrations of N, P, and K, was located close to the origin of the PCA, on the negative side of PC2, suggesting a volatile profile that did not cluster distinctly with any specific fertilization group. Other compounds, including (E)-β-ocimene, bisgermacrene, and epicubenol, were also represented in the PCA space but showed weaker discriminatory power among nutrient regimes.

Three VOCs showing marked variation in the PCA and each representative of one of the three main nutrient-regime groups, eugenol (high-input regimes), eucalyptol (intermediate-input regimes), and linalool (low-input regimes), were selected to relate their accumulation to biomass production through the calculation of a compound index (Table 2). For eugenol, the highest values were observed in S2 and S3, reaching 14.7 and 13.6, respectively, which were numerically higher than the HNS (5.6). For eucalyptol, the highest values were recorded in S4 and S6, with 3.1 and 2.9, compared with 1.2 in the HNS. Linalool also showed higher values than the HNS in several systems, particularly in S5, which exhibited the highest value (12.8) compared with HNS (8.4). Conversely, S7 and S8 displayed the lowest values for most compounds, at levels close to or below those observed in the HNS.

Mathematical modeling of nutrient conditions associated with linalool, eugenol, and eucalyptol production

The outlier assessment revealed no aberrant observations for the linalool, eugenol, or eucalyptol indices; therefore, the complete dataset was retained for model fitting. The quadratic regression models showed variable in-sample goodness of fit depending on the target VOCs (Figure 5A). The highest in-sample R 2 was obtained for eugenol (R 2 = 0.82), followed by eucalyptol (R 2 = 0.73) and linalool (R 2 = 0.64). The corresponding adjusted R 2 values were 0.77, 0.67, and 0.55, respectively, while the in-sample RMSE values were 1.84, 0.41, and 1.61. These statistics describe the in-sample goodness of fit and should not be interpreted as measures of out-of-sample predictive performance. In all three models, most experimental observations were distributed close to the 1:1 regression line. The response surface plots illustrated the differential effects of N and P concentrations on the accumulation of the three VOCs (Figure 5B). Distinct response profiles were observed for each compound. For linalool, the model-predicted surface showed a quadratic pattern, with higher predicted CI values at intermediate nutrient concentrations. The highest model-predicted CI within the experimental domain was obtained at 62.2 mg L −1 N and 17.1 mg L −1 P, associated with approximately 80 mg L −1 K. However, given the limited predictive performance of the linalool model, these conditions should be interpreted as the highest predicted linalool CI within the experimental domain rather than as a definitive nutritional optimum. Beyond these concentrations, the predicted CI progressively decreased. In contrast, eugenol accumulation increased continuously with increasing nutrient concentrations, particularly under high N and P levels. Unlike linalool and eucalyptol, the response surface did not show a closed dome-shaped profile, indicating that the maximum predicted value was located at the upper boundary of the tested nutrient range rather than representing a true internal optimum. Therefore, the conditions of 105 mg L −1 N and 31 mg L −1 P, corresponding to approximately 135 mg L −1 K, should be interpreted as the highest predicted eugenol CI within the experimental domain, rather than as a definitive nutritional optimum. For eucalyptol, the response surface also showed a quadratic behavior, although with a distribution different from that observed for linalool. The highest eucalyptol CI values were observed under intermediate N concentrations combined with relatively high P levels. The highest model-predicted CI within the tested experimental domain was estimated at 58.2 mg L −1 N and 31 mg L −1 P, associated with approximately 93.5 mg L −1 K. At higher N concentrations, the predicted eucalyptol index tended to decrease, highlighting the differential sensitivity of this compound to nutrient supply.

To further describe the potential effects associated with the model-selected nutrient conditions, the fitted compound index (CI) values were visually compared with the mean CI values observed in the HNS (Figure 6). The HNS values are presented as mean ± SD, whereas the model-fitted values are presented with 95% confidence intervals for the mean response. For all three compounds, the fitted values under the model-selected nutrient conditions were higher than the corresponding HNS means. Specifically, the linalool CI increased from a HNS mean of 8.46 to a fitted value of 17.4 (95% CI: 14.3–20.4), the eugenol CI increased from 5.61 to 14.7 (95% CI: 12.3–17.2), and the eucalyptol CI increased from 1.28 to 4.10 (95% CI: 3.45–4.75). These comparisons are descriptive and should not be interpreted as evidence of statistically significant differences or as experimentally validated improvements. Independent cultivation experiments would be required to validate the model-selected conditions.

Soilless cultivation systems are increasingly recognized for their capacity to precisely regulate crop-growing conditions, particularly nutrient availability, thereby promoting stable and homogeneous production (Appicciutoli et al., 2025). In parallel, aromatic plants have attracted growing scientific and industrial interest due to their richness in high-value bioactive compounds (Maggini et al., 2022). In this context, the modulation of phytochemical profiles through the adjustment of cultivation conditions, combined with modeling approaches, represents a promising strategy to guide the targeted production of compounds of interest for pharmaceutical, cosmetic, and nutraceutical applications. Despite the relevance of the findings, some limitations should be acknowledged. First, the study focused mainly on N and P, without explicitly considering the potential role of micronutrients in plant growth and physiological performance. In addition, K was not independently manipulated, as its concentration varied with the amounts of KNO 3 and KH 2 PO 4 used to adjust N and P levels. Consequently, K was correlated with N across the experimental nutrient regimes. In addition, nutrient concentrations in leaf tissues were not measured, preventing a direct assessment of plant nutritional status relative to VOC accumulation. Furthermore, the hexane maceration method used in this study is not specific to the volatile fraction and may also extract other lipophilic compounds. Basil was used as the only model crop, and the experiment was conducted over a single cultivation cycle without independent replication at the NFT-unit level. Therefore, although the models identified promising nutrient formulations, their performance was interpreted cautiously given the exploratory nature of the analysis and their in-sample fit to the data used for model development. The model-predicted nutrient combinations were therefore considered candidate formulations rather than definitive optima and require experimental validation across independent cultivation cycles before any practical recommendation.

Growth and agronomic performance

The agronomic response of basil demonstrated that plant performance does not depend solely on the total quantity of nutrients supplied, but also on the nutritional balance among mineral elements. The ability of nutrient regimes S1 to S3 to maintain aerial performances relatively close to those observed under the HNS suggests that the nutrient concentrations conventionally used in hydroponics may exceed the actual requirements of the plant under certain cultivation conditions. The progressive decline in growth observed across the nutrient regimes was associated with decreasing N concentrations and the concomitant changes in K concentration. Indeed, N is directly involved in chlorophyll synthesis, photosynthetic activity, protein production, and biomass accumulation (Mu and Chen, 2021; Wang et al., 2024). Previous studies in which N was independently varied support the importance of N availability for crop growth. Sapkota et al. (2019) (Sapkota et al., 2019) demonstrated, in two hydroponically cultivated lettuce cultivars, that fresh biomass reached 115 and 93.1 g plant −1 under complete nutrient solution conditions, compared with only 57.1 and 41.9 g plant −1 under N-deficient conditions. Likewise, Petropoulos et al. (2016) (Petropoulos et al., 2016) reported that increasing N concentration from 100 to 200 mg L −1 significantly improved biomass production in lettuce and rocket, with fresh weight increases reaching 113% and 107%, respectively. Under the present experimental conditions, differences associated with P appeared less pronounced than those observed along the N (with associated K) gradient. This interpretation is consistent with the observations of Germano et al. (2022), who showed that reducing P concentration from 1.0 to 0.1 mol m − ³ in hydroponically cultivated basil did not significantly affect shoot biomass, leaf area, or total dry matter. Regarding K, Bevan et al. (2021) similarly reported, in soilless cultivation of Cannabis sativa, that no significant effect on yield was observed across K concentrations ranging from 60 to 340 mg L −1.

The overall similarity observed in root parameters among systems may reflect different physiological strategies leading to comparable root development. Under balanced nutrient conditions, root growth likely supported optimal vegetative development, whereas under lower nutrient availability, plants may have increased root investment to improve nutrient acquisition. This response is consistent with the well-documented plasticity of root systems under nutrient limitation reported by Wu et al. (2005) (Wu et al., 2005) and Gruber et al. (2013) (Gruber et al., 2013), whereby P deficiency or low N availability can stimulate root elongation, lateral root formation, or root hair development in several species, including Fraxinus mandshurica, Arabidopsis thaliana, cotton, and wheat.

VOC profile and compound index of linalool, eugenol, and eucalyptol

The separation observed among nutrient regimes in the exploratory PCA highlights the structuring of VOC profiles associated with mineral nutrition. This structuring primarily followed the N gradient (with associated K), while nutrient regimes with different P concentrations but comparable N levels (with associated K) exhibited similar phytochemical profiles. These results are consistent with previous studies demonstrating the influence of mineral nutrient availability on basil VOCs. Nurzyńska-Wierdak et al. (2013), reported significant variations in VOCs in response to N nutrition across several sweet basil cultivars, with responses depending on genotype and fertilization level. Nutrient regimes characterized by low N inputs (with associated K) were associated with higher levels of linalool, β -myrcene, and related terpenoids. Linalool is produced through the monoterpene biosynthetic pathway from geranyl diphosphate (GPP), which is formed predominantly via the plastidial methylerythritol phosphate (MEP) pathway and subsequently converted by specific terpene synthases (Iijima et al., 2004). The relative increase in linalool under low N regimes (with associated K) suggests a shift in carbon allocation toward terpene biosynthesis, in accordance with the Carbon–Nutrient Balance (CNB) hypothesis (Bryant et al., 1983). Comparable trends were reported by Nurzyńska-Wierdak and Borowski (2013) (Nurzyńska-Wierdak et al., 2013), with linalool contents of 59.9%, 51.2%, and 50.4% under increasing levels of N nutrition. Regimes corresponding to intermediate N inputs (with associated K) were primarily associated with eucalyptol. The biosynthesis of this compound also relies on GPP derived from the MEP pathway but involves specific terpene synthases directing metabolic flux toward eucalyptol formation. The contrasting distribution of linalool and eucalyptol along the N gradient (with associated K) therefore suggests modulation of metabolic flux partitioning among different branches of monoterpene biosynthesis. Nutrient-dependent variations in terpene composition have also been described in Artemisia argyi (Wang et al., 2023). In contrast, nutrient regimes receiving high N inputs (with associated K) were more strongly associated with eugenol. Eugenol biosynthesis proceeds through the phenylpropanoid pathway, initiated from phenylalanine through the activity of phenylalanine ammonia-lyase and followed by several enzymatic steps leading to phenylpropanoid precursors and ultimately to eugenol (Dhar et al., 2020). The association of eugenol with high N regimes (with associated K) suggests modulation of phenylpropanoid metabolism in response to increased mineral nutrient availability. Comparable responses of eugenol to mineral nutrition have been reported in O. basilicum (Nguyen and Niemeyer, 2008) and O. gratissimum (Chandel et al., 2024). The HNS, despite receiving the full nutrient concentrations, was positioned near the PCA origin and did not cluster with any specific fertilization group. The position of the HNS in the PCA may reflect its distinct nutrient composition; however, the underlying physiological mechanism cannot be resolved from the present experiment.

Mathematical modeling for identifying candidate nutrient conditions for linalool, eugenol, and eucalyptol production

Within the fitted dataset, the performance coefficients indicate that the models captured part of the complex and non-linear responses of VOC biosynthesis to variations in N and P supply. These findings confirm the relevance of RSM for studying and optimizing plant secondary metabolites. Similar results were reported by Farjaminezhad and Garoosi (2021) in Azadirachta indica, where quadratic models applied to the accumulation and production of azadirachtin, mevalonic acid, and squalene showed high coefficients of determination (R 2 = 87.5–95.7%), high Precision values, and non-significant lack-of-fit tests, indicating good agreement between the models and the experimental data. Likewise, Alshammari et al. (2023) obtained high coefficients of determination (R 2 = 0.96–0.99), low RMSE values, and coefficients of variation below 10% for phenolic compounds in Ajwa dates, confirming the reliability of RSM models for optimizing plant metabolites.

The response surfaces obtained in the present study revealed distinct nutritional response patterns among the studied compounds, suggesting differential regulation of the metabolic pathways involved in their biosynthesis. This variability may be related to the role of N in the synthesis of enzymes and carbon precursors required for secondary metabolism, as well as the involvement of P in cellular energy transfer. This interpretation is consistent with Klimánková et al. (2008) who showed that, in five basil cultivars grown under conventional and organic systems, linalool, eugenol, and 1,8-cineole varied according to cultivar and production system, with relative abundances ranging from 16–32%, 9–22%, and 3–20%, respectively. These findings confirm that basil VOC exhibit compound-specific accumulation dynamics in response to cultivation conditions. In this context, identifying compound-specific model-predicted candidate nutrient conditions appears particularly relevant for directing production toward targeted phytochemical profiles. The fitted formulations predicted higher production of the three compounds of interest, highlighting the potential of nutritional modulation as a strategy to enhance the value of basil bioactive compounds. These results also emphasize the relevance of predictive modeling as a decision-support tool for managing soilless cultivation systems aimed at producing high-value metabolites. However, further experimental validation is required to confirm the stability and reproducibility of these responses under real cultivation conditions.

This single-cycle exploration study indicates that nutrient solution composition was associated with differences in vegetative growth and VOC accumulation in hydroponically grown basil. The N gradient (with associated K) was most closely associated with agronomic performance. However, as K co-varied with N by design, it was not possible to statistically separate their independent contributions. Similarly, the present design did not allow inference of an independent effect of P on plant growth. Furthermore, nutrient formulations generated distinct VOC profiles, with certain conditions favoring the accumulation of linalool, others promoting eugenol, and others enhancing eucalyptol accumulation. Notably, HNS was positioned near the PCA origin and did not cluster with any specific nutrient-regime group, suggesting a distinct overall volatile profile, although the physiological basis of this pattern could not be resolved from the present experiment. These findings highlight the close relationship between mineral nutrition and the regulation of secondary metabolism, particularly VOC accumulation, in basil. Quadratic models provided exploratory in-sample fits and identified candidate combinations of N, P, and K for each target compound. The nutrient formulations predicted by the models yielded higher fitted compound-index values of linalool, eugenol, and eucalyptol compared with the HNS reference regime, thereby demonstrating the potential of nutrient management as a strategy to guide phytochemical accumulation in hydroponic systems. Overall, these results indicate that nutrient solutions can be tailored not only to maximize plant growth, but also to modulate the qualitative and/or quantitative synthesis of secondary metabolites of commercial and functional interest. RSM therefore represents a valuable tool for the development of more efficient and targeted nutrient management strategies in hydroponic cultivation systems. Further studies should evaluate these optimized formulations under real cultivation conditions, across different environmental conditions, and over multiple production cycles.