3.1. Preliminary Single Factor Experiments and Selection of Limiting Parameters
Several extraction parameters have been described to affect the extraction efficiency of polyphenols from various plant matrices [
35]. Here, using a single-factor experiment approach, the influence of 5 independent parameters (aqEtOH concentration, extraction duration, US frequency, extraction temperature and liquid to solid ratio) on the SILM extraction yield from the mature fruit of
S. marianum were evaluated. The objective of these preliminary experiments being to identify the limiting extraction parameters.
The choice of the solvent is an important parameter to fix during the development of an extraction method. Several organic solvents, such as methanol, ethanol (EtOH) or acetone, are commonly used for the extraction of plant polyphenols [
36]. Here, considering our objective to propose these extracts for future cosmetic applications, and in consideration of the development of a green chemistry extraction method, EtOH was retained. Indeed, EtOH is a solvent that is less toxic to humans and more environmentally friendly when compared to other organic solvents (e.g., methanol) [
35]. Moreover, its extraction capacity can be modulated by the addition of water, thus making it an ideal solvent for the extraction of a wide range of polyphenols with low to high polarity. Interestingly, these two universal solvents (i.e., EtOH and water) have commonly been used for various food and/or cosmetic applications [
35,
36].
Here, the extraction capacity of 5 aqueous EtOH (aqEtOH) solutions at different concentrations (0%, 25%, 50%, 75%, and 100% (v/v) of EtOH in water) were assayed (
Figure 2a). For these preliminary experiments evaluating the impact of aqEtOH concentrations, the other extraction parameters were arbitrary fixed to: 25:1 mL/g DW liquid to solid (L/S) ratio, 30 min for the extraction duration, 30 kHz for the US frequency and 45 °C for the extraction temperature. In our hands, aqEtOH concentration appeared to impact significantly the SILM extraction yield from milk thistle fruits. An optimal extraction yield was obtained for an aqEtOH concentration of 50% (v/v). Extreme values for aqEtOH concentrations (i.e., 0 and 100% (v/v)) resulted in 4- to 10-fold decreases in the SILM content, respectively, whereas aqEtOH concentrations of 25 and 75% (v/v) resulted in intermediate SILM contents.
The US frequency is also known to potentially impact the extraction efficiency. This parameter acts through the modulation of the cavitation effect and the diffusion coefficient of the targeted compounds into the extraction solvent. This could result in a greater solubilization of the target compound in the considered extraction solvent, and to a higher extraction efficiency [
36]. Increasing the US frequency could result in a lower extraction duration, and therefore a lower energy consumption [
37]. However, application of high US frequencies could alter or destroy the native structure, thus reducing both the extraction yield and the biological activity of the targeted compound(s) [
38]. Consequently, US frequency has to be considered carefully during the development of an UAE method. In our hands, the impact of 4 different US frequencies (0, 15, 30 and 45 kHz) on the SILM extraction yield was evaluated (
Figure 2b). For this purpose, the other extraction parameters were arbitrary fixed to: 50% (v/v) aqEtOH concentration, 25:1 mL/g DW L/S ratio, 30 min for the extraction duration and 45 °C extraction temperature. We noted a significant impact of US frequency, with the highest extraction yield obtained using 30 kHz US frequency. The absence (0 kHz) or lower application (15 kHz) of US frequency resulted in a lower extraction efficiency, whereas the highest applied US frequency (45 kHz) led to a decrease in the SILM extraction yield, possibly because of the reported destructive effect high US frequencies [
38].
To reduce energy consumption in a context of a green chemistry approach, optimizing extraction duration is essential [
37]. Increasing extraction duration will not necessarily result in a gain in terms of extraction yield, since a prolonged US exposure can lead to the deterioration of the compounds [
38]. Here, we considered 6 extraction durations (0, 15, 30, 45, 60 and 90 min) with the other parameters arbitrarily fixed to: 50% (v/v) aqEtOH concentration, 25:1 mL/g DW L/S ratio, 30 kHz US frequency duration and 45 °C extraction temperature. A gradual increase in SILM extraction yield from milk thistle fruit as a function of the extraction duration was first observed. The maximum extraction efficiency was reached after 45 min, followed by a significant decrease with 60- and 90-min extraction time (
Figure 2c). This observation is in agreement with other studies reporting on the degradation of antioxidant phenolic compounds with prolonged US treatments [
35,
37,
38].
Different extraction temperatures (30, 40, 50, 60 and 70 °C) were next evaluated (
Figure 2d). The other parameters were fixed to: aqEtOH concentration 50% (v/v), L/S ratio 25:1 mL/g DW, extraction time 45 min and US frequency 30 kHz. Using these conditions, the extraction temperature was not identified as a limiting parameter. According to the hot spot theory, the cavitation bubbles are considered as a microreactor generating a local environment in the surrounding liquid after their collapse with high temperature (ca. 4500 °K) and pressure (ca. 1000 atm) [
36]. This theory could explain the low impact of few dozen temperature degrees on the SILM extraction yield. This parameter was not considered as a limiting parameter, and was not further optimized. An extraction temperature of 45 °C was used hereafter.
Finally, 3 liquid to solid (L/S) ratios (10:1, 25:1 and 50:1, in mL of aqEtOH (50% (v/v)) per gram of DW material) were evaluated (using fixed aqEtOH concentration of 50% (v/v), extraction duration of 30 min, US frequency of 30 kHz and extraction temperature of 45 °C) (
Figure 2e). Only slight and non-significant differences in SILM extraction yields were observed. This parameter was not considered as s limiting parameter, and was not further optimized. Slightly better results were obtained with a L/S ratio of 25:1, and therefore this was used hereafter.
3.2. Development of a Multifactorial Approach
From the preliminary experiments, the significant impacts of aqEtOH concentration, extraction duration and US frequency were evidenced (
Figure 2). These parameters were therefore selected for further optimization. To take into account the possible interactive influence of these parameters, experimental factorial design (design of experiment, DOE) coupled with statistical analysis was employed. This strategy is known to be more effective, precise and rapid for integrating a large number of extraction conditions and for evidencing possible interactions between independent variables as compared with single factor approaches [
39]. Taking into account the preliminary experiments, the 3 influencing variables were: aqEtOH concentration (variable X
1, ranging from 25 to 75% (v/v)), US frequency (variable X
2, ranging from 15 to 45 kHz) and extraction duration (variable X
3, ranging from 20 to 60 min). Their coded levels and experimental values are presented in
Table 1. According to the results obtained during preliminary experiments for L/S ratio and extraction temperature, these parameters were fixed to 25:1 mL/g DW and 45 °C, respectively.
Full factorial design was used to optimize this extraction process considering its high reproducibility as a consequence of the real measurement of a large number of experimental conditions compared to other DOE approaches [
40]. The 27 different bath conditions (run ID) were determined and randomized (run order) in silico. The corresponding independent process variables of each batch condition are presented in
Table 2. Each batch condition was assayed in independent triplicates. The separation was based on the method described by Drouet et al. [
14], here further optimized (see Materials and Methods,
Section 2.3), allowing a high resolutive separation of the different peaks as shown in
Figure 3. The extraction yield results for SILM and each individual constituent of this mixture are presented in
Table 2 and
Table S1, respectively.
The SILM content extracted from mature fruit of
S. marianum ranged from 1.80 (run ID#13) to 17.98 (run ID#26) mg/g DW (
Table 2).
The individual composition of the SILM of each extract was also determined (
Table S1), with results given from the most to the less abundant component:
- -
SILB was detected under each extraction condition and its contents ranged from 1.29 (run ID#1) to 7.52 (run ID#26) mg/g DW;
- -
the detected SILD contents ranged from 0.40 (run ID# 13) to 4.21 (run ID#20) mg/g DW, whereas SILD was not detected for one extraction condition (run ID#16). This run ID#16 presented low aqEtOH concentration (25% (v/v)) and high US frequency (45 kHz);
- -
the detected ISILA contents ranged from 0.45 (run ID#3) to 2.49 (run ID#26) mg/g DW, whereas ISILA was not detected for 9 extraction conditions (run ID#1, #4, #7, #10, #13, #16, #19, #22 and #25). All these run IDs presented the lowest aqEtOH concentration (i.e., 25% (v/v)) as X
1 condition (i.e., X
1 = −1,
Table 1);
- -
the detected SILC contents ranged from 0.01 (run ID#19) to 1.52 (run ID#26) mg/g DW, whereas SILC was not detected for 4 extraction conditions (run ID#7, #16, #22 and #25). Runs #7, #16 and #25 presented low aqEtOH concentration (25% (v/v), X
1 = −1,
Table 1) and high US frequency of 45 kHz (X
2 = +1,
Table 1). Run ID#22 presented the same low aqEtOH concentration and intermediate US frequency of 30 kHz (X
2 = 0,
Table 1) but during a prolonged period of time (X
3 = +1 (i.e., 60 min),
Table 1);
- -
the detected SILA contents ranged from 0.01 (run ID#13) to 1.09 (run ID#26) mg/g DW, whereas SILA was not detected for 3 extraction conditions (run ID#4, #16 and #24). These run IDs presented the same low aqEtOH concentration (i.e., 25% (v/v), X
1 = −1,
Table 1), whereas run ID#16 and #24 presented high US frequency (X
2 = +1,
Table 1) and prolonged extraction duration of 40 and 60 min, respectively (X
3 = 0 or +1, respectively,
Table 1);
- -
TAX was detected under each extraction conditions and its content ranged from 0.16 (run ID#3) to 0.68 (run ID#20);
- -
and finally, the detected ISILB contents ranged from 0.03 (run ID#9) to 0.55 (run ID#26) mg/g DW, whereas ISILB was not detected for 3 extraction conditions (run ID#1–4, #6, #7, #10, #12–16, #19, #22, #24, #25 and #27). Here, we therefore observed that the use of aqEtOH concentration of 25% (v/v) failed to extract ISILB. These results may also be related to the low accumulation of ISILB in the mature fruit of the considered milk thistle cultivar.
To sum up these results, the hypothesis of low extraction yields of SILM and its constituents as a consequence of their lower solubility in extraction solvent with high polarity (i.e., 25% (v/v) aqEtOH concentration, X
1 = −1,
Table 1) and/or of drastic/destructive extraction conditions (i.e., high and/or prolonged US treatment) can be made.
A model of the SILM extraction yield as a function of the 3 different variables was obtained by multiple regression analysis (
Table 3). Using the conditions described in
Table 1 and
Table 2, the SILM extraction yield (Y
SILM) as a function of the 3 different variables (X
1: aqEtOH concentration, X
2: US frequency and X
3: extraction duration) in the form of a polynomial equation was: Y
SILM = 13.52 + 2.29X
1 + 0.78X
2 + 1.96X
3 − 7.45X
12 − 0.86X
22 − 0.25X
32 + 0.32X
1X
2 + 0.55X
1X
3 − 0.11X
2X
3 (
Table 3).
The statistical analysis (
Table 3) evidenced the significant impact on the SILM extraction efficiency from mature fruit of
S. marianum of the linear coefficients X
1 (aqEtOH concentration) and X
2 (extraction time) and the quadratic coefficients X
12. On the contrary, the other linear X
3 (US frequency), quadratic X
22 and X
32 as well as interaction coefficients were not significant (
p > 0.05). Therefore, aqEtOH concentration (X
1), as well as extraction duration (X
3), appeared to be the most influential parameters for this extraction process over US frequency (X
2) for SILM extraction. The same trend was observed for the individual constituents of the SILM, with the exception of ISILB for which the quadratic coefficients X
12 was the sole significant coefficient (
Table S2).
In addition to all these significant coefficients, SILB extraction was also significantly impacted by the US frequency (linear coefficients X2). SILB was therefore the only compound for which extraction was significantly influenced by this US frequency variable. Nevertheless, we have to keep in mind, here, that during the DOE under all the extraction conditions, US were applied at 3 different frequencies that appeared to be in the best range in preliminary experiments. From these preliminary experiments, it clearly appeared that the absence of US treatment drastically reduced extraction efficiency. Therefore, here we can conclude that US frequency X3 variable did not significantly influenced the SILM extraction yield in the selected range of values for this variable, whereas the absence of US had clearly resulted in a less efficient extraction process during the preliminary experiments.
Results of the analysis of variance (ANOVA) and the fit for the models obtained for SILM and its constituents are listed in
Table 4 and
Table S3, respectively. The high F-value (14.73) and the low
p-value (
p < 0.0001) indicated that the model was highly significant and could predict the SILM content as a function of the variable values with a great precision (
Table 4). The same trend was recorded for each individual constituent of the SILM (
Table S3), with a lower but still significant precision for ISILB. This was also confirmed by the low and non-significant lack of fit values. The model precision in the prediction of the experimental values is evidenced by the predicted vs. experimental plot presented in
Figure S1. A determination coefficient
R2 of 0.891 (with adjusted value of 0.833) for SILM extraction model, and ranging from 0.810 for TAX to 0.946 both for SILC and SILD extraction models were obtained. ISILB extraction models presented the lowest
R2 value of 0.589 (
Table 4 and
Table S3). The coefficient value (CV) indicated the adequacy between the model and experimental values.
To understand the complexity of the models, 3D plots were drawn for SILM (
Figure 4) and each individual constituent (
Figures S2–S8).
The linear coefficients of the second-order polynomial equation for X
1 aqEtOH concentration, X
2 US frequency and X
3 extraction duration, as well as the interaction coefficients X
1X
2 (aqEtOH concentration x US frequency) and X
1X
3 (aqEtOH concentration x extraction duration), were all positives, indicating that the increase of these parameters results in a favourable action on the SILM extraction yield. However, their low values, in association with the negative values recorded for their quadratic coefficients (X
12, X
22 and X
32, respectively), as well as for the interaction coefficient between aqEtOH concentration and US frequency (X
2X
3), indicate that the extraction of SILM reaches a maximum value before decreasing for high values of these parameters. We can observe these tendencies on the 3D plots with first a positive action on SILM extraction yield with increased aqEtOH concentrations combined with higher US frequency and/or prolonged extraction duration (
Figure 4a,b). However, the highest aqEtOH concentration, on the one hand, as well as prolonged extraction duration at high US frequency, on the other hand, resulted in a marked decline of SILM extraction yields (
Figure 4). Ethanol/water mixtures represent commonly used eco-friendly solvents able to extract a wide range of polyphenols; however, optimal aqEtOH concentration is highly dependent of the nature of the considered polyphenol [
36]. Applying high US frequency during a prolonged period of time is known to be potentially destructive and to induce polyphenols oxidation, especially when water is used as solvent [
27,
36,
38,
41]. This could lead to a significant decrease in extraction yield, but also to the loss of the biological activities of the extract [
27,
36,
38,
41].
Here, according to the adjusted second order polynomial equation, optimal conditions were: 54.5% (v/v) aqEtOH as solvent, 36.6 kHz for the US frequency and 60 min as extraction time (with a fixed extraction temperature of 45 °C and liquid to solid ratio of 25:1 mL/g DW). Adjusted to the material, an US frequency of 30 kHz was used. Under these optimized conditions, SYLM extraction yield from the mature fruit of AJQ S. marianum cultivar reached 20.28 ± 0.41 mg/g DW.
3.4. Evaluation of the Biological Activities of the Extracts Relevant to Cosmetics
To evaluate the influence of the extraction process, to ensure that the biological activity is retained during this process, and to correlate the biological activity with the phytochemical profiles of the extracts, we next determined the antioxidant and anti-aging potential relevant to cosmetics of each of the 27 run IDs. Indeed, SILM and its flavonolignans have received a recent interest for their potent antioxidant and anti-aging activities relevant to cosmetic [
14,
15,
16,
17,
18,
19].
CUPRAC assay has been reported to effectively evidence the antioxidant activity of milk thistle extracts [
20]. Here, the antioxidant activity evaluated by the CUPRAC assay ranged from 51.33 (run ID#10—SILM content of 3.21 mg/g DW) to 183.80 (run ID#26—SILM content of 17.98 mg/g DW) µM AEAC (
Figure 5,
Table S4). Oxidative stress has been associated with aging and could lead to the formation of advanced glycation end products (AGEs) [
42]. Here, the strong inhibition of AGEs formation confirmed the antioxidant capacity of these extract evidenced by the CUPRAC assay. The inhibition of AGEs formation ranged from 6.64 (run ID#13—SILM content of 1.80 mg/g DW) to 74.31 (run ID#26—SILM content of 17.98 mg/g DW) % of inhibition (
Figure 5,
Table S4). A strong significant correlation was observed between SILM content and CUPRAC antioxidant activity (PCC = 0.862) as well as between SILM content and AGEs inhibitory action (PCC = 0.997) (
Table 6).
All the SILM constituents were also correlated with these antioxidant activities (
Table 6). The SILM content and composition of wild ecotypes of
S. marianum from Pakistan have been correlated with their antioxidant activity measured by CUPRAC assay [
14]. Natural antioxidants have attracted growing interest over the last decade because of their possible use as alternative to the potentially harmful, synthetic antioxidant such as butylated hydroxyanisole (BHA) or butylated hydroxytoluene (BHT) in different formulations [
43,
44,
45]. Recently, a SILB palmitate derivative has been synthesized and displayed a pronounced anti-lipoperoxidant activity, inhibiting the formation of conjugated diene production in two different lipophilic media (bulk oil and o/w emulsion) subjected to accelerated storage test [
45]. Here, this antioxidant action in vitro is further confirmed by the CUPRAC assay correlated with the SILM and SILB contents. In addition, oxidative stress has been associated with aging and age-related diseases [
46], in particular leading to the formation of AGE [
47]. The ability of natural compounds to inhibit their formation have therefore attracted increasing attention in cosmetics. The high inhibition of AGE formation also correlated with the SILM content, in particular with SILA and SILB which is of special interest for future applications.
The next step was to evaluate the inhibitory activity of the extracts toward collagenase and elastase. Indeed, the potent inhibitory action of SILM and its flavonolignans toward these enzymes has been recently evidenced [
15]. A strong inhibitory effect was observed for collagenase, whereas it was less marked for elastase (
Figure 5,
Table S4). Collagenase inhibition ranged from 4.21 (run ID#16—SILM content of 2.49 mg/g DW) to 49.13 (run ID#26—SILM content of 17.98 mg/g DW) % of inhibition, while, elastase inhibition ranged from 6.84 (run ID#13—SILM content of 1.80 mg/g DW) to 22.93 (run ID#26—SILM content of 17.98 mg/g DW) % of inhibition. A strong and significant correlation linking these enzymatic inhibitory actions with SILM content was measured (
Table 6). Elastase and collagenase enzymes act on the remodelling and/or degradation of the extracellular matrix components in the dermis, thus potentially leading to skin alterations such as skin tonus decrease, formation of deep wrinkles and resilience losses [
48,
49,
50]. Our results confirmed the potential of SILM and its constituent as inhibitor of collagenase, and to a less extend of elastase. Future works focusing on the inhibition mechanism rationalization of each flavonolignans would be of particular interest for future applications.