Optimization, Metabolomic Analysis, Antioxidant Potential and Depigmenting Activity of Polyphenolic Compounds from Unmature Ajwa Date Seeds (Phoenix dactylifera L.) Using Ultrasonic-Assisted Extraction

This study sought to optimize the ultrasonic-assisted extraction of polyphenolic compounds from unmature Ajwa date seeds (UMS), conduct untargeted metabolite identification and assess antioxidant and depigmenting activities. Response surface methodology (RSM) utilizing the Box–Behnken design (BBD) and artificial neural network (ANN) modeling was applied to optimize extraction conditions, including the ethanol concentration, extraction temperature and time. The determined optimal conditions comprised the ethanol concentration (62.00%), extraction time (29.00 min), and extraction temperature (50 °C). Under these conditions, UMS exhibited total phenolic content (TPC) and total flavonoid content (TFC) values of 77.52 ± 1.55 mgGAE/g and 58.85 ± 1.12 mgCE/g, respectively, with low relative standard deviation (RSD%) and relative standard error (RSE%). High-resolution mass spectrometry analysis unveiled the presence of 104 secondary metabolites in UMS, encompassing phenols, flavonoids, sesquiterpenoids, lignans and fatty acids. Furthermore, UMS demonstrated robust antioxidant activities in various cell-free antioxidant assays, implicating engagement in both hydrogen atom transfer and single electron transfer mechanisms. Additionally, UMS effectively mitigated tert-butyl hydroperoxide (t-BHP)-induced cellular reactive oxygen species (ROS) generation in a concentration-dependent manner. Crucially, UMS showcased the ability to activate mitogen-activated protein kinases (MAPKs) and suppress key proteins including tyrosinase (Tyr), tyrosinase-related protein-1 and -2 (Trp-1 and -2) and microphthalmia-associated transcription factor (MITF), which associated melanin production in MNT-1 cell. In summary, this study not only optimized the extraction process for polyphenolic compounds from UMS but also elucidated its diverse secondary metabolite profile. The observed antioxidant and depigmenting activities underscore the promising applications of UMS in skincare formulations and pharmaceutical developments.


Introduction
Extraction, the pivotal initial step in the retrieval and purification of bioactive compounds from plant sources, often relies on conventional methods characterized by lengthy extraction times and limited effectiveness [1].To address these shortcomings, green extraction processes have been developed, offering an eco-friendly alternative.These processes are known to significantly reduce processing times, enhance heat and mass transfer rates, improve product quality, minimize solvent usage and promote the adoption of Generally Regarded as Safe (GRAS) solvents [2].The adoption of green extraction methods not only conserves energy but also minimizes adverse impacts on the environment and human health.Among these techniques, ultrasound-assisted extraction (UAE) stands out as a green and highly efficient approach, demonstrating superior recovery yields and the preservation of target compound activities, making it particularly valuable for the extraction of antioxidants [3].
The conventional one-factor-at-a-time approach to optimization often overlooks the intricate interactions among variables, failing to guarantee optimal conditions and necessitating numerous trials, thereby increasing time, costs, and resource consumption [1].To address this challenge, statistical methodologies such as the Box-Behnken design (BBD), a component of response surface methodology (RSM), have emerged, enabling the prediction of optimal extraction conditions and the comprehension of the relationships between extraction factors [4].RSM encompasses a range of statistical and mathematical techniques for optimizing processes influenced by multiple variables, facilitating the development of new products and the enhancement of existing ones.It elucidates how independent variables affect processes, either individually or collectively, offering a mathematical model to represent chemical or biological procedures and assess the impact of independent factors [5].Nonetheless, the forecast accuracy of RSM may be compromised in the presence of nonlinear relationships between variables [6].Artificial neural networks (ANNs), with their capacity for learning algorithms and modeling nonlinear systems, are increasingly adopted as predictive tools across various disciplines, including food technology [7].
Free-radical reactions in biology play a crucial role in tissue damage and pathological events in living organisms, especially in aerobic life where lipids with polyunsaturated fatty acids are susceptible to oxidation [8].Excess oxygen or insufficient reduction can generate reactive oxygen species (ROS) such as superoxide anions, hydroxyl radicals and hydrogen peroxide.Aerobic organisms have a natural antioxidant defense system, but if inadequate, ROSs may cause oxidative damage to macromolecules [9].Phytochemicals with intrinsic antioxidant activity have emerged as potential remedies for oxidative stress-induced disorders.Antioxidative phenolics in plant tissues, serving various roles from structural to defensive, are believed to contribute to their medicinal actions.These compounds, studied extensively for their positive effects on human health, can orchestrate cellular protective signaling cascades, making them valuable candidates for mitigating oxidative stress-induced disorders.Exploring the therapeutic potential of these natural compounds holds promise in understanding and preventing diseases associated with free-radical reactions [8,9].
The date palm, (Phoenix dactylifera L. Arecaceae family), is a globally popular and nutritionally significant fruit.Among its various cultivars, the Ajwa date stands out as one of the most esteemed and expensive varieties due to its ethnomedical associations with health-enhancing properties [13].Preclinical research has highlighted its diverse health-promoting attributes, including antioxidative, anti-inflammatory, anticancer, hepato-protective, antimicrobial, nephroprotective, antidiabetic and hyperlipidemic effects [14][15][16][17][18][19].Additionally, the fruit is a rich source of dietary fiber, minerals, organic acids and vitamins, contributing to its nutritional and therapeutic significance, with carbohydrates constituting over 70% of its composition.Furthermore, it contains a plethora of bioactive components, such as polyphenols, encompassing phenolic acids, flavonoids and lignans [20].Notably, the benefits of Ajwa dates extend beyond the fruit itself to its seeds, the often overlooked and underutilized byproducts of various date-related industries.These seeds, derived from technological or biological transformations of date fruits, are typically discarded or used as fertilizer or animal feed.However, they hold untapped potential as a source of high-value-added components, although limited research has explored their potential in pharmaceutical or nutraceutical product development [21].
This study unveils an innovative method to enhance the extraction of polyphenols from unmature Ajwa date seeds (UMS) by employing UAE, RSM-BBD and ANN.Through the systematic optimization of key parameters such as temperature, time and ethanol concentration, a robust model was developed to maximize polyphenol yield from UMS, representing a significant advancement in the field.Furthermore, the application of highresolution mass spectrometry facilitated the comprehensive identification of bioactive secondary metabolites within UMS, shedding new light on their potential pharmacological benefits.Additionally, the evaluation of antioxidant potential underscores the promising applications of UMS in both food and pharmaceutical industries.An important aspect of this study is the investigation into UMS's depigmentation properties using MNT-1 cells, accompanied by mechanistic studies.These findings not only emphasize UMS's potential in dermatology and skincare but also lay the groundwork for future applications in this field.

Sample Collection and Preparation
Unmature Ajwa date fruits (Kimri stage), harvested in Al-Madina Al-Munawara, Saudi Arabia, were collected in July, 2022 and scientifically verified at the National Herbarium and Genebank of Saudi Arabia, with a voucher specimen (No.NHG005) stored for reference.UAE was conducted at a fixed 25 kHz frequency using specialized equipment (Elma Schmidbauer GmbH, Singen, Germany).Unmatured date seeds (UMS) were meticulously cleaned, air dried (40 ± 1 • C for 2 days using the laboratory dry oven) and ground (average particle size: 300 µm of diameter).Dry powder samples (1.0 g) underwent triple extraction with 10 mL of solvent, following the design in Table 1.Additionally, heat and maceration extraction followed established methods [22].The extracted samples were filtered, concentrated in a rotary evaporator (Tokyo Rikakikai Co., Ltd., Tokyo, Japan) and lyophilized with a freeze dryer (Il-shin Biobase, Goyang, Republic of Korea).The resulting UMS extract was stored at −20 • C for subsequent experiments.The lyophilized extract was dissolved in a mixture of DMSO and dH 2 O (1:9) to create a stock solution of 100 µg/mL.Subsequently, serial dilution was performed using dH 2 O to prepare the experimental concentrations.

Measurement of Total Phenolic (TPC) and Total Flavonoid Content (TFC)
The TPC and TFC in UMS extracts were quantified using the Folin-Ciocalteu test and aluminum chloride colorimetric method, respectively [11].For TPC measurement, 2 µL of a sample (100 µg/mL) was combined with 10 µL of Folin-Ciocalteu's phenol reagent (Sigma-Aldrich, St. Louis, MO, USA).After 5 min, 100 µL of a 7% Na 2 CO 3 solution was introduced, followed by the addition of 90 µL of dH 2 O.The mixture was then incubated in the dark for 90 min at room temperature.Absorbance was subsequently measured at 750 nm.On the other hand, for TFC measurement, 2 µL of the sample was mixed with a solution comprising 100 µL of dH 2 O, 5 µL of 5% NaNO 2 , and 10 µL of 10% AlCl 3 •6H 2 O.After 10 min, 40 µL of NaOH (1 M) was added, and the absorbance was measured against the reagent blank at 506 nm.The experiments were conducted in triplicates for each trial.Utilizing regression equations derived from calibration curves, the TPC (y = 0.0512x + 0.0018; r 2 = 0.9835) and TFC (y = 0.014x + 0.0021; r 2 = 0.9994) were determined.TPC was expressed as gallic acid equivalent (mg)/dry weight sample (g), and TFC as catechin equivalent (mg)/dry weight sample (g).

Cell-Free Antioxidant Assays
The free radical scavenging activity of UMS was evaluated using established protocols for DPPH, ABTS, superoxide, hydroxyl and nitric oxide radicals [8,9].Percent inhibition was calculated using Equation (1) and IC 50 values were determined for each radical to assess UMS efficacy.

Radical scavenging activity(% inhibition
where A and B denote the absorbance of the control and sample, respectively.Each sample was examined three times. For DPPH and ABTS radical scavenging assays, various concentrations of the sample (2 µL) were mixed with 198 µL of DPPH (0.2 M in 50% methanol) and ABTS solutions (2.5 mM potassium persulfate and 7 mM ABTS).The mixtures stood for 10 min at room temperature in the dark, and absorbance readings were taken at 517 nm and 734 nm for the DPPH and ABTS assays, respectively.Ascorbic acid served as the reference antioxidant.
Superoxide radical generation was verified using the non-enzymatic PMS/NADH complex, where NBT is reduced to formazan.Samples (2 µL) were mixed with a superoxide radical generation mixture (73 µM NADH, 50 µM NBT and 15 µM PMS in PBS), incubated for 30 min, and absorbance was read at 562 nm [9].Gallic acid served as the reference antioxidant.
For hydroxyl radical scavenging activity, samples (5 µL) were added to the Fenton reaction mixture (3.6 mM deoxyribose, 0.1 mM EDTA, 0.1 mM ascorbic acid, 1 mM H 2 O 2 and 0.1 mM FeCl 3 in PBS).After a 1 h incubation at 37 • C, 1% TBA and 2.8% TCA were added, heated, and absorbance was measured at 532 nm.Quercetin was the standard.
For nitric oxide measurement, samples (10 µL) were mixed with sodium nitroprusside (10 mM) in PBS, incubated for 150 min, then reacted with Griss reagent (Sigma-Aldrich, St. Louis, MO, USA).Absorbance was measured at 546 nm using catechin as the reference compound.

Cell Culture and Intracellular ROS Generation Assay
RAW 264.7 macrophages (ATCC, Rockville, MD, USA) were maintained in DMEM supplemented with 10% FBS and 100 µg/mL each of streptomycin and penicillin, under standard culture conditions (37 • C, and 5% CO 2 ).Cells (5 × 10 5 /mL) were seeded in 96-well plates and incubated for 12 h.Subsequently, they were treated with UMS (6.25-100 µg/mL) for 24 h, both alone and in combination with t-BHP (oxidative stress inducer).Cellular toxicity was assessed using the MTT assay, while reactive oxygen species (ROS) generation induced by t-BHP was evaluated by the DCFH-DA method, as previously described [8].

Melanin production(% inhibition
where A and B represent the absorbance of non-treated cells and treated with UMS or arbutin (positive control), respectively [11].

Experimental Design of RSM for the Extraction Process
In this study, BBD was employed to optimize the extraction process of UMS for maximizing TPC and TFC.The independent extraction variables considered were ethanol concentration (X 1 ), extraction time (X 2 ) and temperature (X 3 ), while the response variables of interest were TPC (Y 1 ) and TFC (Y 2 ).The relationships between these variables were modeled using a second-order polynomial Equation (3): where Y represents the response variable; X i and X j are the coded independent variables; β 0 is the constant coefficient; and β i , β ii and β ij are the coefficients for linear, quadratic and interaction effects, respectively.The outcomes of these interactions were visualized through three-dimensional (3D) surface plots.

Artificial Neural Network (ANN) Modeling
A multilayer perceptron (MLP) neural network was used to establish a link between independent variables (X 1 , X 2 and X 3 ) and response variables Y 1 and Y 2 using a backpropagation feed-forward ANN model [23].The dataset was divided into training (70%), validation (15%) and testing (15%) sets.Training was conducted using a hit and trial technique to minimize the mean square error (MSE) calculated from Equation (4).Two different types of neural networks, feed-forward and cascade feed-forward, were utilized, employing the Broyden-Fletcher-Goldfarb-Shanno (BFGS) and Levenberg-Marquardt backpropagation (trainlm) algorithms.The ANN model's architecture consisted of three layers, as depicted in Supplementary Figure S1, with the output generated through nonlinear activation functions in the hidden layer.Several statistical parameters, including R 2 , RMSE, AAD and SEP, were computed using specific Equations ( 5)-( 8), to evaluate and compare the predictive performance of the ANN and RSM.This approach allowed for a robust assessment of the nonlinear relationships between input and output variables.
where Y p is the predicted response; Y e is the observed response; Y m is the average response variable; n is the number of experiments

Validation of the Model
To ascertain the optimal extraction parameters for UMS, a combination of response surface and Derringer's desirability function was employed.Each response was transformed into a unique desirability function, ranging from 0 to 1 based on their relative desirability, from lowest to highest.These component functions were then integrated into a total desirability function using Equation ( 9) [1].
To assess the agreement between observed and expected UMS outcomes, calculations of the RSD and RSE were performed using Equations ( 10) and (11), respectively.According to the criteria set forth, data were considered consistent with predictions when RSD and RSE values fell below 10% and 5%, respectively [24,25].

RSD(%) =
Standard deviation between predicted and actual values Mean values between predicted and actual values × 100 ( 10)

Analysis of Chemical Compounds by ESI-MS/MS
In this study, negative-mode ESI-MS experiments were conducted using the Q-Exactive™ Orbitrap Mass Spectrometer (Thermo Fisher Scientific Inc., San Jose, CA, USA).Sample infusion into the ESI source was accomplished directly at a rate of 20 µL/min utilizing a 500-µL syringe (Hamilton Company Inc., based in Reno, NV, USA), coupled with a syringe pump (Harvard in Holliston, MA, USA, model 11).Data acquisition was conducted through Xcalibur 3.1 with Foundation 3.1.(San Jose, CA, USA).The ESI-MS conditions featured a mass resolution of 140,000, a mass spectra range of m/z 50-1000, sheath gas at a flow rate of 5 and auxiliary gas at 0. Additionally, a spray voltage of 4.20 kV, capillary temperature at 320 • C and automatic gain control set at 5 E 6 were applied.MS/MS studies employed three stepwise normalized collision energies (10, 20 and 30) [9].The identification of m/z peaks was achieved by comparing calculated (exact) masses of deprotonated (M-H) adducts with m/z values and ESI-MS/MS fragmentation patterns from in-house and online databases such as FooDB [26], METLIN [27] and CFM-ID 4.0 [28].The chemical structure was drawn using ChemDraw Professional 15.0 (PerkinElmer, Waltham, MA, USA).

Statistical Analysis
Experimental data underwent robust analysis employing Design Expert 11 (Stat-Ease, Minneapolis, MN, USA) and GraphPad Prism 9 (GraphPad Software 9.0.2,San Diego, CA, USA).MATLAB's Neural Network Toolbox TM (MATLAB R2020a, MathWorks, Natick, MA, USA) facilitated artificial neural network (ANN) analysis.Results, presented as mean ± standard deviation from a minimum of three independent experiments (n = 3), were rigorously scrutinized for statistical significance at p < 0.001, < 0.01 and < 0.05.Design Expert 11 enabled RSM analysis.GraphPad Prism 9 conducted one-way analysis of variance and Tukey's multiple comparison test for all biological activities, deeming p < 0.05 statistically significant.

Single Factor Analysis
In this study, experiments were conducted to assess the individual effects of three key factors-ethanol concentration, extraction time and temperature-on the extraction yields of TPC and TFC from UMS.
As shown in Figure 1A, the choice of solvent concentration is pivotal.Ethanol, particularly at 75% concentration, is demonstrated to be the optimal choice due to its compatibility with the solubility of polyphenolic compounds.Notably, aqueous ethanol is favored for its low toxicity and cost-effectiveness, which enhance its efficiency in polyphenolic extraction, as corroborated by recent studies [4].Figure 1B highlights the impact of the ultrasonic extraction time.Extended times (10-30 min) are beneficial for increased polyphenolic yield, but prolonged extraction negatively affects yields.This decline is attributed to structural degradation and solvent loss through vaporization, diminishing mass transfer efficiency [29].Figure 1C emphasizes the role of temperature in polyphenolic compound extraction.Elevated temperatures (30 • C to 50 • C) enhance yields by disrupting cellular structures and reducing viscosity, consistent with prior research [30].However, temperatures above 50 • C result in reduced yields due to decreased acoustic cavitation intensity, diminished extraction efficiency and potential thermosensitive compound degradation [31].yphenolic yield, but prolonged extraction negatively affects yields.This decline is tributed to structural degradation and solvent loss through vaporization, diminish mass transfer efficiency [29].Figure 1C

Fitting of the RSM and ANN Models
Table 1 presents the outcomes of 17 extraction scenarios, summarizing the exp mental parameters and results.The yields of TPC and TFC in UMS extracts exhibite range from 60.23 ± 0.79 to 76.65 ± 0.49 mgGAE/g and 37.62 ± 1.10 to 59.40 ± 0.89 mgCE respectively.The peak TPC and TFC values were achieved at the central point of the sign (X1: 60% EtOH; X2: 30 min; and X3: 50 °C).Both RSM and ANN predictions w closely aligned with experimental results, with minimal discrepancies.Quadratic poly mial equations (Eq.( 12) and ( 13

Fitting of the RSM and ANN Models
Table 1 presents the outcomes of 17 extraction scenarios, summarizing the experimental parameters and results.The yields of TPC and TFC in UMS extracts exhibited a range from 60.23 ± 0.79 to 76.65 ± 0.49 mgGAE/g and 37.62 ± 1.10 to 59.40 ± 0.89 mgCE/g, respectively.The peak TPC and TFC values were achieved at the central point of the design (X 1 : 60% EtOH; X 2 : 30 min; and X 3 : 50 • C).Both RSM and ANN predictions were closely aligned with experimental results, with minimal discrepancies.Quadratic polynomial equations (Equations ( 12) and ( 13)) were utilized to model the response variables (TPC and TFC, respectively), accounting for their variation concerning the extraction factors.
As described in Table 2, ANOVA was utilized to evaluate the statistical significance of second-order quadratic model equations.Significance levels were determined by p-values, classifying terms as significant (p < 0.05), highly significant (p < 0.01) or exceptionally significant (p < 0.001).Conversely, terms with p-values above 0.05 were considered nonsignificant.The statistical significance of the model equations, as determined by the F-test and p-values, underscores their reliability.The high F-values (for 107.80 TPC and 46.54 for TFC) with p-values less than 0.0001 confirm the models' significance.Furthermore, lack of fit tests yielded non-significant p-values (for 0.1772 TPC and 0.3516 for TFC), affirming the appropriateness of the second-order polynomial models for predicting total polyphenolic extraction yields [1].The determination coefficient (R 2 ) demonstrates the model's adequacy, with values of 0.9928 for TPC and 0.9836 for TFC, indicating that over 99% of the variation in total polyphenolic yield can be explained by the model.High adjusted (R 2 adj: 0.9836 and 0.9624 for TPC and TFC, respectively) and predicted determination coefficients (R 2 pred: 0.9192 and 0.8503 for TPC and TFC, respectively) further confirm the correlation between the observed and predicted data.Furthermore, low coefficient of variation (C.V.) values (1.21 for TPC and 3.40 for TFC) and high Adeq.precision values (25.0494 for TPC and 16.9654 for TFC) suggest a high degree of precision and reliability in the experimental data.
In addition, to visualize interactions between independent variables, 3D surfaces and contour plots were generated using multiple linear regression equations.These graphical representations (Figure 2A,B) help elucidate the main and cross-product effects of independent variables on the response variables, enhancing the understanding of the underlying processes.This comprehensive statistical analysis supports the robustness and validity of the model for predicting total polyphenolic extraction yields.The three-dimensional response surface analysis in RSM examined the relationship between TPC, TFC and extraction parameters.Contour plot shapes conveyed the significance of mutual interactions.Elliptical contours implied negligible interactions, while circular contours indicated significant interactions [22,25].In Figure 2A,B, all response surfaces were convex, affirming the appropriate selection of variable ranges for ultrasonic time, temperature and ethanol concentration, highlighting their collective impact on TPC and TFC extraction yields.There is a growing body of evidence supporting the superiority and sophistication of artificial neural network (ANN) modeling over RSM, making ANNs a promising alternative for intricate nonlinear multivariate modeling in various fields, particularly in biological processes.ANNs are inspired by the human CNS and its intricate network of interconnected neurons, enabling complex computations in response to input data [32,33].In this study, experimental data were subjected to ANN modeling for validation.The predicted values from the ANN model closely matched the observed values (Table 1), affirming the model's appropriateness.
The ANN model effectively captured the nonlinear relationships between extraction parameters (X1, X2 and X3) and response variables (Y1 and Y2), as evidenced in Figure 3A  There is a growing body of evidence supporting the superiority and sophistication of artificial neural network (ANN) modeling over RSM, making ANNs a promising alternative for intricate nonlinear multivariate modeling in various fields, particularly in biological processes.ANNs are inspired by the human CNS and its intricate network of interconnected neurons, enabling complex computations in response to input data [32,33].In this study, experimental data were subjected to ANN modeling for validation.The predicted values from the ANN model closely matched the observed values (Table 1), affirming the model's appropriateness.
The ANN model effectively captured the nonlinear relationships between extraction parameters (X 1 , X 2 and X 3 ) and response variables (Y 1 and Y 2 ), as evidenced in Figure 3A-D

Comparison of the Prediction Abilities of the RSM and ANN Models
The comparative evaluation of RSM and ANN models for predicting TPC and TFC in UMS revealed the superior performance of the ANN model (Table 3).With higher R 2 values (0.9963 for TPC and 0.9912 for TFC) than RSM (0.9928 for TPC and 0.9836 for TFC), the ANN model exhibited an enhanced predictive capability.Lower AAD and RMSE values indicated a better fit.The ANN model also exhibited lower SEP values (0.0171 for TPC and 0.0326 for TFC), further emphasizing its accuracy.The ANN model's flexibility in approximating nonlinear systems surpassed the RSM model, being constrained to secondorder polynomial regression.Additionally, the ANN model's efficiency in handling multiple responses in a single run outperformed the RSM model, often requiring multiple runs for multi-response optimization.Dadgar et al. emphasized the ANN model's superiority in accuracy, precision and fitting experimental data to target responses, establishing it as a more effective tool in this context [34].

Comparison of the Prediction Abilities of the RSM and ANN Models
The comparative evaluation of RSM and ANN models for predicting TPC and TFC in UMS revealed the superior performance of the ANN model (Table 3).With higher R 2 values (0.9963 for TPC and 0.9912 for TFC) than RSM (0.9928 for TPC and 0.9836 for TFC), the ANN model exhibited an enhanced predictive capability.Lower AAD and RMSE values indicated a better fit.The ANN model also exhibited lower SEP values (0.0171 for TPC and 0.0326 for TFC), further emphasizing its accuracy.The ANN model's flexibility in approximating nonlinear systems surpassed the RSM model, being constrained to second-order polynomial regression.Additionally, the ANN model's efficiency in handling multiple responses in a single run outperformed the RSM model, often requiring multiple runs for multi-response optimization.Dadgar et al. emphasized the ANN model's superiority in accuracy, precision and fitting experimental data to target responses, establishing it as a more effective tool in this context [34].

Model Validation and Comparison with Other Extraction Methods
The simultaneous optimization of TPC and TFC in UMS extracts was achieved using Derringer's desirability function [34].The maximum overall desirability (D = 0.953) was attained under specific conditions: ethanol concentration (X 1 , 61.43 ∼ = 61.00%),extraction time (X 2 : 29.60 ∼ = 30.00min) and extraction temperature (X 3 , 50.24 ∼ = 50.00• C).The contour plot as a function of ethanol concentration, extraction time and temperature at the optimum condition are presented in the Supplementary Materials, Figure S2A-C.As described in Figure 4A,B, the predictive model was validated with triplicate experiments, resulting in TPC and TFC values of 77.52 ± 1.55 mgGAE/g and 58.85 ± 1.12 mgCAE/g, respectively, with low RSD% (1.80 for TPC and 1.71 for TFC) and RSE% (2.59 for TPC and 2.45 for TFC).

Model Validation and Comparison with Other Extraction Methods
The simultaneous optimization of TPC and TFC in UMS extracts was achieved using Derringer's desirability function [34]  A comparative study was performed to evaluate the efficiency of the optimized extract (UOP) obtained through UAE in comparison to traditional methods.As shown in Figure 4C, UOP exhibited significantly higher TFC values compared to other extracts.Remarkably, UOP showed no significant difference in TPC compared to HEW (head assisted extraction coupled with 75% EtOH), while surpassing other methods.Similarly, UOP demonstrated superior 2,2-diphenyl-1-picrylhydrazyl (DPPH) and ABTS radical scavenging activity compared to all other extracts, while UE and HEW displayed no significant difference in their scavenging capacities (Figure 4D).These findings underscore the remarkable efficiency of UAE, which not only improves polyphenol yield and antioxidant activity but also minimizes extraction time.This improved efficiency may be attributed to A comparative study was performed to evaluate the efficiency of the optimized extract (UOP) obtained through UAE in comparison to traditional methods.As shown in Figure 4C, UOP exhibited significantly higher TFC values compared to other extracts.Remarkably, UOP showed no significant difference in TPC compared to HEW (head assisted extraction coupled with 75% EtOH), while surpassing other methods.Similarly, UOP demonstrated superior 2,2-diphenyl-1-picrylhydrazyl (DPPH) and ABTS radical scavenging activity compared to all other extracts, while UE and HEW displayed no significant difference in their scavenging capacities (Figure 4D).These findings underscore the remarkable efficiency of UAE, which not only improves polyphenol yield and antioxidant activity but also minimizes extraction time.This improved efficiency may be attributed to various factors affecting ultrasonic energy transmission, including gas diffusion, gas-liquid phase transitions and chemical reactions [29].

Identification of Secondary Metabolites in UMS by High-Resolution Mass Spectroscopy
In the optimized UMS extracts, secondary metabolites were analyzed using ESI-MS/MS in the negative ionization mode.A total of 104 compounds were successfully identified, relying on the precursor ion mass, characteristic fragmentation patterns and comparisons with the reference standards, literature and online databases (Table 4).The significance of these findings was assessed based on confidence levels: Level 1 for compounds confirmed with reference standards, Level 2 for probable identifications supported by MS n data from the literature, and Level 3 for tentative candidates [35].EF: elemental formula; OM: observed mass; CM: calculated mass; CL: confidence level.
In Table 4, compounds 1-13 were identified as phenolic compounds in UMS based on mass fragmentation patterns.Notably, the phenolic acids in UMS were often in glucosidic form or conjugated with quinic and shikimic acid.While compounds 1-12 had been reported in various date cultivars in previous studies [1,17,22,24], compound 13, with a monoisotopic mass [M-H] − of m/z 747.1895 and the molecular formula C 38 H 36 O 16 , was newly identified as 3-O-feruloyl-7-O-acyl-feruloyl-4-O-caffeoyl-quinic acid in the Ajwa date.

Antioxidant Effect of UMS
The assessment of antioxidant activities in phytochemicals necessitates a comprehensive understanding of various molecular mechanisms.In this investigation, multiple methodologies were employed to evaluate the antioxidant potential of UMS.The DPPH, ABTS, superoxide, hydroxyl and NO radical scavenging assays, as well as CUPRAC and FRAP assays, were utilized.As depicted in Figure 5A,B, UMS demonstrated a dosedependent and significant DPPH and ABTS radical scavenging potential, exhibiting IC 50 values of 42.62 ± 0.27 µg/mL and 34.42 ± 1.44 µg/mL, respectively.In comparison, the positive control, ascorbic acid, exhibited an IC 50 value of 13.79 ± 0.87 µg/mL and 21.44 ± 0.27 µg/mL in the DPPH and ABTS assays, respectively, indicating that UMS not only engage in hydrogen atom transfer but also operates through a single electron transfer mechanism.The evaluation of superoxide and hydroxyl radical scavenging abilities employed the PMS-NADH superoxide-generating system and Fenton reaction, respectively [8]. Figure 5C,D illustrate UMS's robust potential to scavenge superoxide and hydroxyl radicals, with IC 50 values of 36.84 ± 1.02 µg/mL and 42.32 ± 0.59 µg/mL, respectively.In contrast, the positive controls, gallic acid and quercetin, had IC 50 values of 14.12 ± 0.77 µg/mL and 11.21 ± 1.06 µg/mL for superoxide and hydroxyl radical scavenging, respectively.These results suggest that UMS employ a single electron transfer mechanism for its antioxidant activity.Moreover, UMS exhibited dose-dependent NO radical scavenging activity with an IC 50 value of 41.77 ± 1.64 µg/mL, while the positive control catechin had an IC 50 value of 8.47 ± 0.39 µg/mL (Figure 5E).Furthermore, CUPRAC and FRAP assays were performed, revealing UMS's notable reduction capability with values of 189.42 ± 12.98 and 158.32 ± 12.05 µmol AAE/g at 50 µg/mL, respectively (Figure 5F).5F).Furthermore, t-BHP, a well-known short-chain lipid peroxide analogue that is frequently used to produce oxidative stress in cells, was employed to examine cellular responses to oxidative stress in both cellular and tissue settings in order to evaluate the potential of UMS in attenuating cellular oxidative stress.The induction of oxidative stress by t-BHP resulted in notable cell death.However, pretreatment with UMS and gallic acid effectively mitigated cellular toxicity at non-toxic concentrations (Figure 5G).Additionally, as depicted in Figure 5H, UMS demonstrated a dose-dependent reduction in the generation of cellular ROSs, further highlighting its antioxidative properties.Furthermore, t-BHP, a well-known short-chain lipid peroxide analogue that is frequently used to produce oxidative stress in cells, was employed to examine cellular responses to oxidative stress in both cellular and tissue settings in order to evaluate the potential of UMS in attenuating cellular oxidative stress.The induction of oxidative stress by t-BHP resulted in notable cell death.However, pretreatment with UMS and gallic acid effectively mitigated cellular toxicity at non-toxic concentrations (Figure 5G).Additionally, as depicted in Figure 5H, UMS demonstrated a dose-dependent reduction in the generation of cellular ROSs, further highlighting its antioxidative properties.
This study corroborates existing evidence indicating that date seed extracts, particularly from Ajwa seeds, exhibit high radical scavenging activity in various cell-free antioxidant methods.Ajwa seeds surpass date flesh in antioxidant properties, positioning them as a promising natural source of antioxidants.The attributed effectiveness of Ajwa seeds in addressing various conditions is likely linked to polyphenolic compounds functioning as reducing agents, free radical scavengers and hydrogen donors.The antioxidant properties and polyphenolic composition of date seeds are influenced by factors such as genetic diversity, soil conditions, maturity stages, storage conditions and extraction methods.Furthermore, UMS, containing cinnamic acid, benzoic acid hydroxylated derivatives and various flavonoids, is implicated in its antioxidant properties, offering protection against oxidative-stress-induced diseases such as inflammation, hyperlipidemia and diabetes.

Depigmenting Effect of UMS on Hyperpigmented Melanocyte (MNT-1) Cells
Melanogenesis, the process of melanin production, plays a crucial role in skin pigmentation and can be implicated in various skin conditions [12].Mushroom tyrosinase is a well-established enzyme used to test compounds that inhibit melanogenesis [51] and the results indicated that UMS significantly suppressed mushroom tyrosinase activity in a concentration-dependent manner, with a lower IC 50 value (48.60 ± 1.02 µg/mL) compared to the positive control, arbutin (IC 50 = 131.03± 2.01 µg/mL) (Figure S4).
Furthermore, the study examined the impact of UMS extract on melanin levels in MNT-1 cells and revealed a dose-dependent reduction (Figure 6A) without causing cytotoxicity (Figure S5).This reduction in melanin content was associated with the downregulation of key melanogenesis-related proteins, including Tyr, Trp-1, -2 and MITF, as confirmed through western blot analysis (Figure 6B).The inhibition of MITF expression is particularly significant, as MITF is a master regulator of melanogenesis.The presence of flavonoids and procyanidins in the UMS extract aligns with the observed depigmenting effects.Flavonoids (luteolin, taxifolin, quercetin catechin, epigallocatehin, procyanidinA2 and B2) have been previously linked to anti-melanogenic activity in the cosmetic industry [52,53], and the mass spectrometric analysis confirmed the abundance of these compounds in the UMS extract.Moreover, the study delved into the underlying mechanisms of UMS's depigmenting effects.The study demonstrated that the UMS extract stimulated the p38 and ERK signaling pathways in MNT-1 cells, which play roles in regulating melanogenesis.In contrast, UMS fail to trigger JNK activation in MNT-1 cells (Figure 6C).Moreover, in order to confirm the involvement of the p38 and ERK signaling pathways in mediating UMS's depigmenting effects, specific inhibitors for p38 and ERK were administered either alone or in combination with UMS.The findings indicated that inhibiting p38 and ERK selectively successfully reversed the depigmenting effects of UMS.This association linked the activation of these pathways to the reduction of MITF expression and the inhibition of Tyr expression (Figure 6D,E).This supports earlier research that demonstrated that polyphenolic-rich plant extracts achieve depigmentation through similar MAPKs' signaling-mediated MITF downregulation [11,54].

Conclusions
This pioneering study investigated the optimization of UAE conditions for extracting bioactive compounds from unmature Ajwa date seeds using both RSM and ANN modeling.High-resolution mass spectrometry analysis identified phenolic acids, flavonoids, lignans and fatty acids in the extract.The ANN model outperformed the RSM model, exhibiting higher accuracy and reliability, with the optimal conditions determined to be 61%

Conclusions
This pioneering study investigated the optimization of UAE conditions for extracting bioactive compounds from unmature Ajwa date seeds using both RSM and ANN modeling.High-resolution mass spectrometry analysis identified phenolic acids, flavonoids, lignans and fatty acids in the extract.The ANN model outperformed the RSM model, exhibiting higher accuracy and reliability, with the optimal conditions determined to be 61% ethanol, 29 min of extraction time and an extraction temperature of 50 • C, while the extract/solvent ratio was fixed to 1:10 (g/mL).Under these conditions, the extract yielded maximum TPC of 77.52 ± 1.55 mg GAE/g and TFC of 58.85 ± 1.12 mg CE/g.Furthermore, UMS showed a potent antioxidant activity in various cell-free antioxidant assays and the mitigation of t-BHP induced cellular ROS generation.The date industry generates thousands of discarded byproducts, such as date pomace and seeds, that are rich with bioactive compounds.New aspects of using these byproducts to produce high-nutritional-value food products have recently attracted interest.Additionally, this study highlighted the extract's potential as an anti-melanogenic agent, showing its ability to inhibit mushroom tyrosinase, reduce melanin levels and modulate melanogenesis-related proteins.The activation of p38 and ERK signaling pathways further supports its potential for pigmentation-related skin care products.This research underscores the significance of Ajwa date seeds as a source of bioactive compounds and encourages its further exploration in dermatology and cosmetics, including isolating bioactive markers and in vivo testing for therapeutic applications.

Supplementary Materials:
The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/antiox13020238/s1, Figure S1: The ANN models architecture; Figure S2: The contour plot as a function of ethanol concentration, extraction time and temperature at optimum condition; Figure S3: Mushroom tyrosinase inhibition activity of UMS optimized extract; Figure S4: Effect of UMS optimized extract on MNT-1 cell viability; Table S1: List of antibodies used in this study; Table S2: Independent process variables with experimental ranges and levels for ultrasound assistant extraction of UMS [47].
emphasizes the role of temperature in polyp nolic compound extraction.Elevated temperatures (30 °C to 50 °C) enhance yields by d rupting cellular structures and reducing viscosity, consistent with prior research [3 However, temperatures above 50 °C result in reduced yields due to decreased acou cavitation intensity, diminished extraction efficiency and potential thermosensitive co pound degradation [31].

Figure 1 .
Figure 1.Effect of (A) ethanol concentration; (B) extraction time; and (C) extraction temperature UMS extraction on ethanol concentration, time and temperature single factor on the yield of T and TFC of UMS extract.Different letters represent statistical significance (p < 0.05) between e group.(a, b, c and d for TPC; x, y and z for TFC).

Figure 1 .
Figure 1.Effect of (A) ethanol concentration; (B) extraction time; and (C) extraction temperature on UMS extraction on ethanol concentration, time and temperature single factor on the yield of TPC and TFC of UMS extract.Different letters represent statistical significance (p < 0.05) between each group.(a, b, c and d for TPC; x, y and z for TFC).
-D.Like RSM, the fitness and significance of the ANN model relied on various parameters, including R 2 values for training, validation, testing, overall error reduction and the number of epochs to prevent overfitting or underfitting.Notably, the best validation performance for TPC occurred at epoch 6, and for TFC at epoch 4 (Figure3A,B).Furthermore, high R 2 values exceeding 0.97224 for TPC and 0.99998 for TFC (Figure3C,D) underscore the model's precision and its potential to accurately represent the complex relationships between the variables.

Figure 2 .
Figure 2. The three-dimensional (3D) response surface plots of UMS extraction on ethanol concentration, time and temperature for TPC (A) and TFC (B).
. Like RSM, the fitness and significance of the ANN model relied on various parameters, including R 2 values for training, validation, testing, overall error reduction and the number of epochs to prevent overfitting or underfitting.Notably, the best validation performance for TPC occurred at epoch 6, and for TFC at epoch 4 (Figure3A,B).Furthermore, high R 2 values exceeding 0.97224 for TPC and 0.99998 for TFC (Figure3C,D) underscore the model's precision and its potential to accurately represent the complex relationships between the variables.

Figure 3 .
Figure 3. Evaluation of ANN model performance.Assessing the validation of the constructed ANN model for (A) TPC and (B) TFC.Illustration of the optimal multilayer perceptron (MLP) architecture used in training, testing and validating the regression analysis to minimize errors for ANN model development for (C) TPC and (D) TFC.

Figure 3 .
Figure 3. Evaluation of ANN model performance.Assessing the validation of the constructed ANN model for (A) TPC and (B) TFC.Illustration of the optimal multilayer perceptron (MLP) architecture used in training, testing and validating the regression analysis to minimize errors for ANN model development for (C) TPC and (D) TFC.
. The maximum overall desirability (D = 0.953) was attained under specific conditions: ethanol concentration (X1, 61.43 ≅ 61.00%), extraction time (X2: 29.60 ≅ 30.00 min) and extraction temperature (X3, 50.24 ≅ 50.00 °C).The contour plot as a function of ethanol concentration, extraction time and temperature at the optimum condition are presented in the Supplementary Materials, Figure S2A-C.As described in Figure 4A,B, the predictive model was validated with triplicate experiments, resulting in TPC and TFC values of 77.52 ± 1.55 mgGAE/g and 58.85 ± 1.12 mgCAE/g, respectively, with low RSD% (1.80 for TPC and 1.71 for TFC) and RSE% (2.59 for TPC and 2.45 for TFC).

Figure 4 .
Figure 4. Model validation and comparative study of TPC and TFC of various extraction methods.(A) TPC and TFC value of optimized condition.(B) Relative standard deviation (RSD) and relative standard error (RSE) value of optimized condition.(C) Effect of various extraction techniques on the yield of TPC and TFC of UMS extract.(D) Effect of various extraction techniques on the DPPH and ABTS radical scavenging effects of UMS extract.Different letters represent statistical significance (p < 0.05) between each group.(a, b, c and d for TPC; m, n, o, and p for TFC).U: ultrasonic assisted extraction; H: heat assisted extraction; M: maceration extract; OP: optimized condition; E: 100% EtOH; W: 100% aqueous; EW: 75% EtOH; AAE: ascorbic acid equivalent.

Figure 4 .
Figure 4. Model validation and comparative study of TPC and TFC of various extraction methods.(A) TPC and TFC value of optimized condition.(B) Relative standard deviation (RSD) and relative standard error (RSE) value of optimized condition.(C) Effect of various extraction techniques on the yield of TPC and TFC of UMS extract.(D) Effect of various extraction techniques on the DPPH and ABTS radical scavenging effects of UMS extract.Different letters represent statistical significance (p < 0.05) between each group.(a, b, c and d for TPC; m, n, o, and p for TFC).U: ultrasonic assisted extraction; H: heat assisted extraction; M: maceration extract; OP: optimized condition; E: 100% EtOH; W: 100% aqueous; EW: 75% EtOH; AAE: ascorbic acid equivalent.

Author
Contributions: F.A.: methodology, formal analysis, investigation, writing-original draft.M.N.: investigation, formal analysis.M.B.A.: conceptualization, investigation, formal analysis, writing-review and editing.S.K.: conceptualization, supervision, writing-review and editing.S.-H.L.: conceptualization, supervision, funding acquisition, writing-review and editing.All authors have read and agreed to the published version of the manuscript.Funding: This study was supported by the National Research Foundation of Korea (NRF), funded by the Ministry of Science and ICT (2020R1A2C2011495, 2021R1IA1A01058062 and RS-2023-00278670).Institutional Review Board Statement: Not applicable.Informed Consent Statement: Not applicable.Data Availability Statement: Data will be available upon request.

Table 1 .
Box-Behnken design (BBD) for the independent variables and corresponding response value (experimental).

Table 3 .
Comparison of the prediction ability of RSM and ANN.

Table 3 .
Comparison of the prediction ability of RSM and ANN.

Table 4 .
List of possible identified compounds of UMS using ESI-MS/MS in the negative ionization mode ([M-H] − ).