The effect of various extraction techniques on the quality of sage (Salvia officinalis L.) essential oil, expressed by chemical composition, thermal properties and biological activity

Highlights • Sage essential oils was isolated by different extraction techniques.• Principal compound was viridiflorol followed by camphor, thujenes, and verticiol.• Samples were assessed for antioxidant, antimicrobial, and cytotoxic activities.• Obtained essential oils were investigated for thermal behavior by TGA analysis.• ANN model was developed, for the anticipation of antioxidant activity.


Introduction
Medicinal herbs are known and used for the treatment of various diseases and disorders since ancient times. Such extensive application of these plants is because of the presence of the various chemical compounds which are beneficial for human health. Awareness about the existence of such compounds in the plants dramatically increased the popularity of tea as a drink all around the globe (Vidović et al., 2013). The beverage itself is preparing using almost every part of the plant including fruits, leaves, flowers, and sometimes seeds and stems, while preparation is quite simple and includes pouring the hot water over the plant material allowing the steeping process for 5-10 min (Chan et al., 2010;Piljac-Žegarac et al., 2013). It is worth mentioning that preparation relies on medical purposes as well as on tradition. Thus, tea may be made by multiple infusions, using water heated at different temperatures, or with milk, honey, lemon, and/or sugar (Belščak et al., 2011). After steeping for a certain period in hot water, biologically active compounds such as terpenes and polyphenols dissolved in the water. Their presence is very significant because they are very potent antioxidant agents which are able to neutralize reactive oxygen (ROS) (Aoshima et al., 2007;Atoui et al., 2005). Besides the above-mentioned antioxidant activity, compounds in plants possess a wide range of another biological potency such as antimicrobial, anti-carcinogenic, antiviral, anti-inflammatory, anti-allergic, immune-stimulatory, and estrogenic activities (Čestic et al., 2016).
Sage (Salvia officinalis L.) is one of about 900 species from the genus Salvia. It is a perennial, evergreen shrub native to Mediterranean and Middle East areas (Radivojac et al., 2020;Russo et al., 2013). This plant is well known as a medicinal plant, used for the treatment of several diseases and disorders, such as depression, obesity, diabetes, lupus, dementia, cancer, and heart diseases (Hamidpour et al., 2014). According to the many research, sage has antimicrobial action against various Gram-positive (Streptococcus sp. Bacillus sp., etc.) and Gramnegative (Escherichia coli, Klebsiella pneumonia, Pseudomonas sp., etc.) bacteria, and fungi (Císarová et al., 2018).
The objective of this study is to investigate the influence of the extraction techniques on the chemical profile, thermal behaviour, and biological activity of sage essential oil. For such purposes, two extraction techniques (hydrodistillation and microwave-assisted hydrodistillation) were used for the isolation of essential oils. Obtained samples were tested for antioxidant, antimicrobial, and cytotoxic activities by different in vitro assays. Furthermore, another objective of this study was to investigate the potential to anticipate the antioxidant activity of sage essential oils, depending on the content of bioactive compounds obtained in samples. This investigation was investigated according to Yoon's interpretation method (Yoon et al., 1993), performed using the developed artificial neural network model.

Plant material
For this study, leaves of the sage (Salvia officinalis L.) was selected for isolation of the essential oil. Plant material was acquired from the Institute for Medical Plant Research "Dr. Josif Pancic" (Belgrade, Republic of Serbia) in 2020. Plant material was kept in the paper bags in shade and at a constant temperature. The leaves were grounded in the blender prior to the essential oil isolation. The mean particle size (0.6034 mm) was determined by sieve set (CISA Cedaceria Industrial, Barcelona, Spain).

Isolation of the essential oil
Isolation of the essential oil (EO) from the sage plant material was done by using previously described classical hydrodistillation and microwave-assisted hydrodistillation (Radivojac et al., 2020). Hydrodistillation was done at 200 and 400 W (D 200 W and D 400 W), while microwave-assisted hydrodistillation was performed at 200 (MWD 200 W), 400 (MWD 400 W), 600 (MWD 600 W), and 800 W (MWD 800 W).

Chemical profile
Chemical profile of EO samples was done by using Thermo Fisher Focus GC coupled with Polaris Q mass spectrometer according to the previously reported (Šojić et al., 2019) and fully described method (Micić et al., 2021). Dissolved samples were injected into the GC through TriPlus autosampler (2 µL) into the TR WAX-MS column (30 m × 0.25 mm, 0.25 µm). Compounds were identified by comparing their mass spectra with mass spectra in NIST 2011 database and MS spectra obtained after injecting the standards. Results were expressed as relative percentage (%) and as milligram of analyzed compound per gram of EO (mg/g).

Thermal analysis
TA Instruments TGA Q500 Thermogravimetric Analyzer (Delaware, USA) was used for the thermal analysis of sage EOs. Samples (10.0 ± 0.5 mg) were heated from ambient temperature to 200 • C, with heating rate of 5 • C/min, under nitrogen purge flow of 60 mL/min. TA Advantage Universal analysis 2000 software was used to process all thermograms.

Antioxidant activity
Antioxidant activity of the obtained EOs was determined by using 5 different in vitro assays (DPPH, CUPRAC, FRAP, ABTS, HRSA, and TBARS). The DPPH assay is based on the scavenging of stable DPPH radical and was conducted according to the previously described method (Espín et al., 2000) with slight modification and adaptation for analysis of the EOs . CUPRAC (cupric ion reducing antioxidant capacity) relies on the reduction of Cu (II) ions to Cu (I) ion which is followed by the color change into orange-yellow. Measurements were done according to the previously described method (Özyürek et al., 2011). FRAP (ferric antioxidant power reduction) is essentially the electron transfer process (Cerretani & Bendini, 2010). The ABTS test is the reaction of the conversion of the ABTS + into neutral ABTS which is followed by discoloration (Zhen et al., 2016). HRSA (hydroxy radical scavenging assay) is the scavenging of the hydroxy radicals formed during the Fenton reaction. The test was conducted according to the previously described method (Sundararajan et al., 2018). TBARS (thiobarbituric acid reactive substances) assay measures the amount of malonaldehyde presented in the sample and generated during the lipid peroxidation. Assay was conducted according to the previously reported method (Falowo et al., 2019). All results were expresses as IC 50 values (µg/mL) indicating the amount of the sample required to neutralize 50% of radical species.

Cytotoxic activity
Cytotoxic activity was assessed according to the previously reported MTT assay (Radojković et al., 2016). Samples were tested against 4 cell lines: cells derived from cervical cancer cells (HeLa), cells derived from human colon cancer (LS-174), adenocarcinomic human alveolar basal epithelial cells (A549), and normal cell line . Results are expressed as IC 50 values in µg/mL.

Statistical analysis
All experiments were run in triplicate. The presented results are formatted as: means ± standard deviation (±SD, n = 3). The posthoc Tukey's HSD test was implemented to test the differences between means of samples. The estimations were done using Statistica 2010 software. Principal Component Analysis (PCA) was used for evaluation of the influence of bioactive compounds content, antioxidant activity, microbiological data and cytotoxic activity of samples obtained in various sage samples.

ANN modelling
A multi-layer perceptron model (MLP), employing three layers was used for artificial neural network (ANN) modelling, in order to explore the antioxidant activity of sage samples, according to the bioactive compounds content. The experimental database was normalized to enhance the conduct of the ANN model. The Broyden-Fletcher-Goldfarb-Shanno (BFGS) algorithm was engaged in solving nonlinear problems throughout the network modelling (Kollo & von Rosen, 2005). A sequence of various topologies (more than 100,000) were investigated in the course of ANN modelling, altering the number of neurons in the hidden layer (from 5 to 20), arbitrarily setting initial weights and biases (Ochoa-Martínez & Ayala-Aponte, 2007).

The accuracy of the model
The quantitative study of the erected ANN model's exactness was conducted according to the basic statistical tests, such as the coefficient of determination (r 2 ), reduced chi-square (χ 2 ), mean bias error (MBE), root mean square error (RMSE), mean percentage error (MPE), average absolute relative deviation (AARD) and sum of squared errors (SSE) (Aćimović et al., 2020).

Global sensitivity analysis
The Yoon's interpretation method was exploited to ascertain the relative influence of bioactive compounds content on antioxidant activity of sage samples. This method was established using the weight coefficients of the developed ANN model (Yoon et al., 1993).

Table1
Chemical profile of sage essential oil samples.  (Table 1S, Supplementary material) demonstrated similar situation like in the case of relative content. High content of camphor (55.01-144.09 mg/g), eucalyptol (5.29-79.26 mg/g), α-thujone (8.97-85.48 mg/g), β-thujone (2.89-29.70%), and borneol (45.80-92.45 mg/g) were noticed. The same trends were noticed in the case of the influence of the extraction parameters. All ketones achieved maximal yield in D 200 W, the yield of eucalyptol reached its maximum in MWD 800 W, while borneol was in MWD 200 W. When comparing the yields of alcohols borneol, and linalool with the yields of their less polar derivatives borneol acetate and linalool acetate, different behavior might be noticed. Borneol and bornyl acetate were achieved their maximal yield in the same sample (MWD 200 W), while situation was slightly different in the case of linalool and linalyl acetate. The highest yield of linalool was in sample MWD 200 W, while linalyl acetate was not found in samples obtained by hydrodistillation, while the highest yield achieved in MWD 800 W. Although linalool is hydrocarbon alcohol it achieved its maximal yield at same condition as borneol. On the other hand, structural differences obviously showed much higher influence in the case of their acetate derivatives.
In order to thoroughly describe the structure of the exploratory data that would provide a better perception of dissimilarities between different sage samples, according to the bioactive compounds content, PCA was used, and the results are displayed in Fig. 1. The first two PCs explained 77.16% of the total variance in the experimental data. According to results, sample MWD 400 W was characterized by the increased amount of bioactive compounds such as: bornyl acetate, menthol, isothujol, myrthenol, spathulenol, cis-carveol, phenylethyl alcohol, caryophyllene oxide, globulol, ledol, viridiflorol, aromadendrene oxide, thymol, verticiol, and valencene, while sample MWD 800 W was described by the augmented concentration of following bioactive compounds: α-pinene, camphene, β-pinene, limonene, eucalyptol, pcymene, α-thujone, β-thujone, cis-linalool oxide, linalyl acetate, transβ-caryophyllene, terpinen-4-ol, alloaromadendrene, carane, and pcymen-8-ol. It might be seen that mentioned compounds in MWD 400 W sample are oxygenated terpenes, while majority of above-mentioned compounds in sample MWD 800 W are monoterpenes. Samples D 200 W and D 400 W were classified according to the increased concentration of bioactive compounds like: camphor, caryophyllenyl alcohol, widdrol, o-cymen-5-ol and ledene oxide-(II). Linalyl acetate was not detected in MWD samples, while β-pinene was found in MWD-800 W.

Thermal analysis
The evaporation process of tested sage EOs was analyzed using thermogravimetry, in non-isothermal conditions. The obtained thermogravimetric (TG) and differential thermogravimetric (dTG) curves are shown in Fig. 2.
The evaporation process of all samples started at room temperature and ended at about 153 • C for samples D 200 W and MWD 800 W, about 158 • C for samples D 400 W and MWD 200 W, and about 167 • C for samples MWD 400 W and MWD 600 W. It can be seen that the evaporation rates significantly differed between the samples from the TG curves. The MWD 800 and MWD 400 samples were two extremes; the MWD 800 had the highest evaporation rate, and the MWD 400 the lowest. The evaporation rates of the rest of the samples were between these two extreme cases. Two peaks and one shoulder at the end of the  Table 1. Fig. 2. a) Thermogravimetric (TG) and b) differential thermogravimetric (dTG) curves of the evaporation process of sage essential oils obtained under a different extraction conditions. Heat rate 5 • C/min, nitrogen purge flow 60 mL/min. evaporation process were detected on all dTG curves, except for the MVD 400 sample. One peak with a larger shoulder on the left and a smaller shoulder on the right side was detected on the dTG curve of the MWD 400 sample. These facts indicate that the evaporation process of analyzed EOs is a complex one. This could be expected, since essential oils are mixtures of a large number of components with different volatility. Almost 50 components were detected in these EOs (Table 1S, Supplementary material). Two peaks on dTG curves point out that the evaporation process took place in two steps. The first one, where more volatile components evaporated, and the second one, where residual, less volatile components, evaporated. The boiling point of components can be taken as one of the measures of their volatility. The most prevalent components (make up greater than 80% of the total share in the samples, Table 1)  The MWD 600 W sample had approximately the equal ratio, while the MWD 400 W had a significantly higher share of LV components. This ratio of MV and LV components in the samples can serve to explain the shape of the obtained TG and DTG curves. In the case of the samples with a higher or equal share of MV components compared to LV components (all samples, except MWD 400 W), the effect of evaporation of MV components was dominant at lower temperatures. As a consequence, the first peak on the dTG curves arose. As the temperature increased, LV components accumulated in the samples. At higher temperatures, the effect of their evaporation became dominant, resulting in the appearance of a second peak on the dTG curves. In the case of MWD 400 W sample, due to the significantly higher share of LV components, their evaporation effect at lower temperatures was not negligible as in the other samples. As a result of the overlapping effects of the evaporation of MV and LV components, the shoulder appeared on the dTG curve at lower temperatures. At higher temperatures, a peak on dTG curve stood out, as a consequence of evaporation of residual LV components in the sample. Also, a large proportion of LV components was the cause of the lowest evaporation rate of MWD 400 W sample.

Biological activity of essential oil samples
Results of antioxidant, antimicrobial, and cytotoxic activity assessments are given in Tables 2S, 3S, and 4S, respectively. To thoroughly describe the structure of the exploratory data that would provide a better perception of dissimilarities between different sage samples, according to the antioxidant activity, microbiological data and cytotoxic activity of samples, PCA was used, and the results are displayed in Fig. 3. The first two PCs explained 86.36% of the total variance in the experimental data. All antioxidant activity assays (Table 2S) showed that MWD 400 W was the most potent sample. On the other hand, D 200 W was the least potent sample in the most cases. Sample MWD 400 W showed the highest content of viridiflorol, caryophyllene oxide, aromadendrene oxide, thymol, carvacrol, and verticiol. Some of these compounds or their synergistic activity may be the explanation for such potency of this sample. Antimicrobial activity (Table 3S)  Previous studies showed that biological activity is closely related to the chemical composition of the tested samples (Table 1 and 1S, chromatograms are given in Figures 1-6S, Supplementary material). Bosnić et al. (2006) have determined the 1,8-cineole compound appears to be more effective against gram-positive strains (S. aureus and B. subtillis) (Bosnić et al., 2006). In other words, the presence or absence of the compounds with a certain functional group may significantly impact the activity (Riabov et al., 2020). Thus, compounds such as α-pinene, 1,8cineole (eucalyptol), p-cymene, menthol, and terpinene-4-ol showed insignificant activity against DPPH radicals. This study also showed that α-terpineol and α-phellandrene showed moderate activity, while γ-terpinene, pulegone, myrcene, citral, and carvone showed significant activity (Wojtunik et al., 2014). Research also showed that the presence of the double bond in the structure enhances the activity, while molecules with conjugated double bonds neutralize radicals very quickly (Riabov et al., 2020). The higher potency of the 1,4-cyclohexadiene system against free radicals is the reason why γ-terpinene is more potent than α-terpinene. However, the aromatic system decreases the activity, but the hydroxyl functional group in fact increases the antioxidant activity of the molecules. The relationship between the structure and activity might be clearly seen in the case of the comparison of the activity of pcymene, thymol, menthone, menthol, and pulegone (Wojtunik et al., 2014).
On the other hand, the ability to make hydrogen bonds is proven to be very important for antimicrobial activity (Griffin et al., 1999). Griffin et al. (1999) also reported that geometric isomerism did not affect the activity by testing nerol and geraniol. They also reported that the position of the functional group does not influence the activity. However, the presence/absence of the hydroxyl functional group in the structure is of great importance to antimicrobial activity (Griffin et al., 1999). Previous studies showed that alcohol was more active than appropriate aldehyde/ketone (Trombetta et al., 2005;Zengin & Baysal, 2014). Besides the hydroxyl functional group, the presence and position of the double bond are also important for the activity (Riabov et al., 2020).
Thus, terpinene-4-ol, in this case, was more potent than α-terpineol, which was not the case for antioxidant activity. The explanation for the higher potency of terpinene-4-ol is the position of the OH functional group, which makes this molecule more suitable for making hydrogen bonds (Griffin et al., 1999). Since sage is rich in essential oils, its antimicrobial potential has a wide variability depending on the sensitivity of microorganisms and the efficiency of the tested compounds. It was also stated that hydrophobicity/lipophobicity also had an important role in determining antimicrobial potency. They divide into a lipid bilayer of the cell membrane, making it more permeable, causing leakage of vital cell contents (Beheshti-Rouy et al., 2015). Differences in these properties make these molecules easily permeable through the cell membrane, thus interfering with the fluidity and permeability of the membrane and leading to the cell lysis (Andrade-Ochoa et al., 2015;Zengin & Baysal, 2014).

ANN model
The obtained optimal neural network model could be applied to adequately anticipate the antioxidant activity of sage samples, on the basis of bioactive compounds content. The optimal number of neurons in the hidden layer was 9 (network MLP 47-9-6), while the highest r 2 values during the training cycle were 0.998).
The obtained ANN model for the prediction of outputs was complex (492 weights-biases coefficients) due to the high nonlinear relation between variables in the database. The goodness of fit tests results between experimental and ANN model were shown in Table 5S.
The ANN predicted values were in a close viscinity to the measured values in most cases, in terms of r2 values (Erbay & Icier, 2009;Turányi & Tomlin, 2014). The SSE values obtained with the ANN model was of the same order of magnitude as experimental errors for output variables presented in many references (Doumpos & Zopounidis, 2011). The ANN model had an insignificant lack of fit tests, which means the model satisfactorily predicted output variables.

Global sensitivity analysis-Yoon's interpretation method
According to the Fig. 4, α-thujone and menthone content showed the stronger positive influence on DPPH, CUPRAC, FRAP, ABTS and HRSA, while the content of verticiol and valencene content exerted the negative influence on DPPH, CUPRAC, FRAP, ABTS and HRSA. The positive influence on TBARS was observed for α-thujone, menthone, camphor and carvyl acetate content.
These conclusions are in line with PCA results, in which the highest content of α-thujone and menthone (codes 9 and 11, respectively, according to Fig. 1) were obtained for samples D 400 W and D 200 W, while the most pronounced antioxidant capacity was observed for samples D 200 W and D 400 W (Fig. 3).

Conclusion
This study showed that procedure of the EOs preparation may significantly influence the chemical composition, thermal behaviour, and biological activity of the essential oils. Although same compound was the principal in all samples (viridiflorol), content and presence/ absence of the minor compounds was significantly different. Besides viridiflorol, several other compounds such as 1,8-cineole (eucalyptol), αand β-thujones, camphor, borneol, and verticiol were also found in significant amounts. Diversities in the composition induced the diversities in the biological activity, where MWD 400 W was the most potent antioxidant agent. On the other hand, in the case of antimicrobial and cytotoxic activities situation was slightly different. Structural diversities influenced in the way that D 200 W and MWD 400 W were the most potent antibacterial agents depending on the microbial strain. In the case of cytotoxic activity, samples prepared by classical hydrodistillation (D 200 W and D 400 W) were the most potent cytotoxic agents.
The artificial neural network model expressed quite a good fit for anticipating antioxidant assays according to the bioactive content, (r 2 values during training cycle for these variables were: 0.998). According to the Yoon interpretation method, α-thujone and menthone content expressed the highest positive effect on DPPH, CUPRAC, FRAP, ABTS and HRSA, while the content of verticiol and valencene content showed the stronger negative influence on DPPH, CUPRAC, FRAP, ABTS and HRSA. The positive influence on TBARS was observed for α-thujone, menthone, camphor and carvyl acetate content.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.