Valorisation of Maclura pomifera Pruning Leaves by Green Crosscurrent Extraction: Process Optimisation, Chemical Profiling, and Preliminary Cytotoxicity Assessment

Abstract

Urban tree-pruning residues are an underused bioactive resource. This study evaluated a short three-stage crosscurrent process for valorising Maclura pomifera pruning leaves with water and aqueous ethanol. Each stage contacted the preceding residue with fresh solvent and was optimised independently by response surface methodology for temperature, time and, for hydroethanol, ethanol concentration. Responses were total phenolic content (TPC), total flavonoid content (TFC), ABTS radical scavenging, potassium ferricyanide reducing antioxidant power (PFRAP) and silver nanoparticle antioxidant capacity (SNPAC). Extracts from all three optimised stages were pooled per solvent and assessed by UPLC–ESI–MS/MS and for cytotoxicity in four mammalian cell lines. Cumulative TPC reached 21.49 (aqueous) and 23.39 (hydroethanolic) mg gallic acid equivalents/g dry weight, and ABTS reached 52.59 and 49.58 mg ascorbic acid equivalents/g dry weight. Two aqueous stages recovered over 90% of cumulative TPC and ABTS in 15 min. The ethanol-rich third stage contributed 39.2% of cumulative TFC and 28.8% of SNPAC. Hydroethanol favoured flavonoids, whereas water phenolic acids and simple aromatics. Both pooled extracts gave higher IC₅₀ values than ascorbic acid, with the hydroethanolic extract being more cytotoxic. Crosscurrent extraction with water or hydroethanol recovers phenolics rapidly, and staging the ethanol concentration yields compositionally distinct fractions whose suitability for specific applications requires further toxicological evaluation.

1. Introduction

Urban trees provide shade, improve air quality, and support climate resilience [1], but they require regular pruning to control growth and maintain safety, generating recurrent streams of leaves, branches, and other residues [2,3]. These materials are generally discarded [4] or used for low-value energy recovery [5,6,7], despite the possibility of containing extractable organic compounds with potential applications in food, cosmetic, pharmaceutical, agricultural, and biotechnological industries [8,9,10,11].

Maclura pomifera (Osage orange, Moraceae) is widely planted as an ornamental and windbreak species for its robust growth and rot-resistant wood [12,13,14], and like other urban trees it requires regular pruning that generates abundant, potentially low-cost foliage. Phytochemical research on the species has focused almost exclusively on the fruit, a rich source of prenylated isoflavones (osajin, pomiferin) with antioxidant, anti-inflammatory, and anticancer activities [15,16,17,18]; by comparison, the chemical composition and valorisation potential of the leaves have received considerably less attention. Developing processes that can efficiently recover and characterise bioactive constituents from M. pomifera pruning leaves would therefore support their use as a value-added urban biomass resource.

Phenolic compounds and flavonoids are among the principal classes of secondary metabolites investigated in plant biomass valorisation, valued in particular for their antioxidant properties [19,20,21]. Increased interest in natural antioxidants, driven by safety concerns over synthetic additives and the shift towards bio-based ingredients, has encouraged the use of agricultural and forestry residues as sources of phenolic-rich extracts for food, nutraceutical, cosmetic, and biomedical applications [20,22,23,24,25,26]. Phenolic-rich extracts are also of interest beyond direct antioxidant use: their reducing and capping constituents can support plant-mediated nanoparticle synthesis [27,28], a property this study exploits analytically, using silver nanoparticle formation as one of three antioxidant-capacity assays.

Valorising this class of compounds requires more than a yield or activity value; the chemical composition and safety profiles of the resulting extracts must also be characterised to determine possible downstream applications. Liquid chromatography coupled to mass spectrometry can resolve the relative distribution of phenolic acids, flavonoids, and other constituents across extracts obtained under different solvent conditions [29,30], while cell-based cytotoxicity assays offer a fast, ethically preferable first screen ahead of any in vivo testing [31,32,33,34,35,36,37,38]. Ascorbic acid, a reference antioxidant that is safe at nutritional levels but cytotoxic at high concentrations [39,40,41,42], provides a practical benchmark for interpreting extract activity against non-specific cytotoxicity. Several studies describe the bioactivity and cytotoxicity of M. pomifera fruit extracts [15,16,17,18,43], but no comparable data exist for the leaves, which currently limits their incorporation into value chains that require a preliminary safety assessment.

The extraction process is a central factor in converting plant residues into higher-value products. Conventional techniques such as maceration, digestion, decoction, and Soxhlet extraction require long extraction times and, in several cases, sustained heating, which increases energy demand and risks degrading thermolabile compounds [44,45]. Single-stage solid–liquid extraction is also inherently inefficient; as solutes accumulate in the liquid phase, and the concentration gradient that drives extraction declines, added time or solvent brings diminishing returns [46,47,48]. Multistage crosscurrent extraction offers a process intensification alternative, contacting the solid with fresh solvent across short successive stages whose operating conditions can be set independently. Renewing the solvent restores the concentration gradient, and stage-wise adjustment of extraction conditions can selectively recover chemically distinct fractions, giving both higher cumulative recovery and a degree of compositional fractionation. When sustainable solvents such as water or aqueous ethanol are used at moderate temperatures, this approach aligns with green extraction principles [49,50,51,52]. Despite these advantages, crosscurrent extraction remains rarely applied to urban pruning residues, and its potential for M. pomifera leaves in particular, how successive aqueous and hydroethanolic stages affect cumulative recovery, antioxidant response, and extract composition, has not been examined.

Research on the valorisation of lignocellulosic biomass has advanced considerably through green-solvent process intensification, particularly in the deconstruction of the structural matrix. Deep eutectic solvents, including microwave-responsive and ternary formulations, achieve rapid and selective delignification under mild conditions and enable downstream conversion of the carbohydrate fraction to fuels and platform chemicals [53,54]. These approaches target the structural polymers of the biomass. The present work addresses the complementary fraction: the extractable secondary metabolites that can be recovered before any structural processing, leaving a delignifiable residue. Recovering this fraction first is a prerequisite for cascading valorisation, in which the highest-value constituents are isolated ahead of bulk conversion.

The present study aimed to develop and evaluate a green crosscurrent extraction process for M. pomifera pruning leaves, integrating process optimisation, chemical profiling, and preliminary cytotoxicity assessment. Three successive extraction stages were independently optimised for aqueous and hydroethanolic systems using response surface methodology. Extraction temperature, extraction time, and, for the hydroethanolic system, ethanol concentration were evaluated using total phenolic content, total flavonoid content, and three complementary antioxidant responses, 2,2′-azino-bis(3-ethylbenzothiazoline-6-sulfonic acid (ABTS) radical-scavenging capacity, potassium ferricyanide reducing antioxidant power (PFRAP), and silver nanoparticle antioxidant capacity (SNPAC), as process-performance indicators. The optimised stage extracts were subsequently combined and characterised by ultra-performance liquid chromatography coupled to electrospray ionisation tandem mass spectrometry (UPLC–ESI–MS/MS), and their preliminary cytotoxicity was evaluated using BHK21, IBRS2, MDBK, and Vero cell lines. The specific contribution of this work is threefold. First, each extraction stage was optimised independently rather than applying a single set of conditions repeatedly, which allows the operating point to be adapted to a feedstock whose composition changes as extraction proceeds. Second, this reveals a behaviour not captured by conventional single-stage optimisation: the ethanol-rich final stage functions as a targeted fractionation step, contributing a disproportionate share of flavonoid and silver-nanoparticle antioxidant capacity while adding little to bulk phenolic recovery. Multistage extraction is therefore shown to act as a means of compositional fractionation rather than simply of yield improvement. Third, the study provides the first extraction, chemical profiling and preliminary cytotoxicity data for M. pomifera leaves, a fraction of an already widely planted urban species for which phytochemical work has focused almost exclusively on the fruit.

2. Materials and Methods

2.1. Plant Material

M. pomifera pruning waste was collected in September from the alleys and gardens of the National Polytechnic School of Algiers (Algiers, Algeria). Leaves were separated from branches, rinsed thoroughly with tap water, and air-dried at room temperature in the dark for ten days. The dried leaves were milled and sieved to a particle size between 74 and 149 µm, then stored in airtight containers until use.

The moisture content of the plant material was determined gravimetrically by drying to constant weight. Fresh leaves contained 59.7 ± 1.2% w/w moisture (105 °C, n = 3). The residual moisture content of the milled, air-dried powder used in all extractions was 4.1 ± 0.2% (105 °C, n = 3).

2.2. Solvents and Reagents

Absolute ethanol (≥99.8%), Folin–Ciocalteu reagent, gallic acid, sodium carbonate, aluminium chloride, potassium acetate, quercetin, ABTS, potassium persulfate, ferric chloride, potassium ferricyanide, ascorbic acid, and silver nitrate were obtained from Sigma-Aldrich (St. Louis, MO, USA). Dulbecco’s modified Eagle medium (DMEM), foetal bovine serum (FBS), penicillin–streptomycin, and trypsin–EDTA were supplied by Gibco (Thermo Fisher Scientific, Waltham, MA, USA). Resazurin was purchased from Sigma-Aldrich. All chemicals were of analytical grade or higher; ultrapure water was produced using a Milli-Q system (Merck Millipore, Darmstadt, Germany).

2.3. Experimental Design and Optimisation

Response surface methodology (RSM) was used to optimise extraction parameters for both aqueous and hydroethanolic systems across a three-stage process. A face-centred central composite design evaluated the effects of extraction time, temperature, and (for the hydroethanolic system) ethanol concentration on TPC, TFC, and antioxidant activity. The factor ranges were identical at every stage and for both solvent systems, and the composition of each design is summarised in Table 1. The hydroethanolic design comprised 16 runs per stage (8 factorial, 6 axial and 2 centre points). The aqueous design comprised 10 runs at stage 1 (4 factorial, 4 axial and 2 centre points) and 13 runs at stages 2 and 3, in which five centre-point replicates were run; the additional replicates provide a more robust pure error estimate for the lower-magnitude responses obtained at these stages.

Table 1. Face-centred central composite designs used for the aqueous and hydroethanolic extraction systems. (a) Factors and coded levels. (b) Design composition.

(a)
Factor		Code	Unit	−1	0	+1
Extraction temperature		X1	°C	15	37.5	60
Extraction time		X2	min	1	5.5	10
Ethanol concentration		X3	% v/v	30	50	70
(b)
System	Stage	Factors	Factorial Points	Axial Points	Centre Points	Total Runs
Aqueous	1	2	4	4	2	10
Aqueous	2, 3	2	4	4	5	13
Hydroethanolic	1, 2, 3	3	8	6	2	16

Full design matrices with all measured responses are given in Table S2.

The factor ranges were established from preliminary trials on the same feedstock, combined with process and equipment constraints, and are summarised in Table 1. The temperature range (15–60 °C) was bounded below by 15 °C, held using the refrigerated function of the incubator shaker as a controlled room-temperature reference requiring no heating, and above by 60 °C, which is both the maximum operating temperature of the incubator (MaxQ 4000R, Thermo Fisher Scientific, Waltham, MA, USA) and the upper end of the range generally reported as optimal for polyphenol recovery [55,56]. Phenolic compounds are reported to remain substantially stable at this temperature over short contact times, with thermal degradation becoming significant only after prolonged exposure [57]. The extraction-time range (1–10 min) was determined from preliminary trials, in which the readily extractable fraction was released within the first minutes of contact and longer contact gave only marginal increments. The ethanol range (30–70% v/v) was bounded above at 70% v/v, the highest level at which chlorophyll co-extraction remained acceptable in preliminary trials. Higher ethanol contents recovered chlorophyll in amounts that add non-phenolic mass to the extract, which is undesirable for downstream applications [58]; co-extraction remained acceptable at 70% v/v, which is why this level was retained as the upper bound. A lower bound below 30% v/v was unnecessary because the parallel aqueous system already provides the 0% v/v condition, so the two designs together span a polarity ladder of 0, 30, 50 and 70% v/v ethanol.

Regression and analysis of variance were performed in SAS JMP (Version 19.03; SAS Institute Inc., Cary, NC, USA) to assess model significance, factor effects, interactions, quadratic terms and lack of fit for all responses. Significance was assessed at α = 0.05. Full second-order models were fitted and reduced through backward elimination to remove non-significant terms (p > 0.05) with hierarchy preserved. Lack of fit was tested against the pure error estimated from the replicated centre points. Residuals were examined as externally studentised residuals; where lack of fit was significant, the influence of the run with the largest residual was assessed by refitting the model without it. Predictive performance was assessed by leave-one-out cross-validation, in which each model was refitted with one run omitted in turn, the predicted residual error sum of squares (PRESS) accumulated over all runs, and the predicted R² were calculated.

Optimal conditions for each stage were identified with SAS JMP (SAS Institute Inc., Cary, NC, USA) by maximising a desirability function applied to the reduced response surface models. Each response was assigned a maximisation goal and rescaled to an individual desirability, d, on a 0–1 scale increasing monotonically with the response, with the low and high anchor points set at the minimum and maximum response values observed in the design for the stage concerned. The responses optimised together were combined into a composite desirability, D, taken as the geometric mean of the individual desirabilities. All responses were given equal importance, so the geometric mean was unweighted. The variables entering the optimisation were the design factors alone: extraction temperature (X₁), extraction time (X₂) and, for the hydroethanolic system, ethanol concentration (X₃). No covariates or blocking terms were included in any model or in the optimisation. D was maximised numerically within the experimental domain of each design. For the aqueous system, a single composite desirability was maximised simultaneously across all five responses (TPC, TFC, ABTS, PFRAP and SNPAC). For the hydroethanolic system, the responses divided into two groups with opposing dependence on ethanol concentration: TPC, ABTS and PFRAP favoured water-rich compositions, whereas TFC and SNPAC favoured ethanol-rich compositions. A single composite desirability across all five responses would therefore have selected an intermediate ethanol concentration that optimises neither group. Two desirability functions were consequently evaluated at each hydroethanolic stage, one maximising TPC, ABTS and PFRAP and one maximising TFC and SNPAC. The first was used to select the conditions for stages 1 and 2, and the second for stage 3.

The conditions identified by numerical optimisation were validated experimentally. Each optimal condition set was applied in triplicate and the measured responses were compared with the model predictions.

2.4. Extraction Procedure

All runs used 5.0 g of milled leaf powder and 50 mL of solvent (1:10 w/v), pre-equilibrated at the target temperature. The ratio refers to the air-dried powder as weighed. Extractions were performed in 250 mL closed flasks with orbital shaking in an incubator shaker (MaxQ 4000R, Thermo Fisher Scientific, Waltham, MA, USA) at 150 rpm. After the specified time, mixtures were vacuum-filtered (Whatman No. 4, 20–25 µm) and the filtrates stored at −20 °C; the solid cake from individual design runs was not retained.

The feedstock for each subsequent stage was prepared separately (Figure 1). A bulk extraction was carried out under the optimal conditions identified for the preceding stage, and the recovered cake was dried at 45 °C overnight and homogenised. This bulk residue was then used for all runs of the following stage, so that every run within a design started from an identical feedstock and differences between runs could be attributed to the factor settings alone. Inter-stage drying was therefore a requirement of the experimental design rather than of the process itself. In continuous operation the cake would pass directly to the next stage without drying. Some residue was lost during vacuum filtration of the bulk preparations. These losses were not quantified; the purpose of the bulk step was to generate a homogeneous feedstock for the following design rather than to establish a mass balance. These losses do not affect the response surface models because all runs within a stage used the same homogeneous residue. Responses at each stage are expressed per gram of the material fed to that stage. Each design run was performed in triplicate, as were the validation experiments carried out under the optimised conditions. Results were expressed as means ± standard deviations, and response surface models were fitted to the mean of the three replicates for each run.

Figure 1. Crosscurrent extraction procedure and downstream analysis.

For UPLC-ESI-MS/MS and cytotoxicity assays, extracts from the optimal conditions of all three stages were combined and concentrated under reduced pressure (30 mbar) at 45 °C (Hei-VAP, Heidolph, Schwabach, Germany). Solvent was not recovered at any point in the process. Extracts were concentrated under reduced pressure and the distillate discarded, solvent recovery being outside the objectives of the present study.

After completion of the optimised three-stage process, a single additional extraction was carried out on the final residue under deliberately exhaustive conditions (60 °C, 60 min) to assess how much extractable phenolic material remained. This fourth stage was not optimised and was not part of the experimental design; water was used for the aqueous system and 70% v/v ethanol, corresponding to the stage 3 optimum, for the hydroethanolic system. Only TPC was determined.

2.5. Analytical Methods

2.5.1. Total Phenolic Content

TPC was determined by the Folin–Ciocalteu method [59], with slight modification: 0.5 mL diluted extract was mixed with 2.5 mL of ultrapure water and 2.5 mL of Folin–Ciocalteu reagent, incubated for 5 min in the dark, and then treated with 0.75 mL 20% (w/v) sodium carbonate and 1 mL ultrapure water. After 20 min, absorbance was read at 750 nm (Uviline 9000, SECOMAM, Alès, France). Results were expressed as mg gallic acid equivalents (GAE) per g dry weight (DW) using a gallic acid calibration curve that ranged from 10 to 100 µg/mL. Cumulative values are the sum of the three stage-wise yields.

2.5.2. Total Flavonoid Content

TFC was determined by the aluminium chloride method [60]: 0.5 mL extract, 2 mL water, and 0.15 mL 5% sodium nitrite were mixed and incubated 5 min; 0.15 mL 10% AlCl₃ was added, followed after 1 min by 1 mL 1 M NaOH and 1.2 mL ultrapure water. After 5 min, absorbance was read at 510 nm. Results were expressed as mg quercetin equivalents (QE) per g DW using a calibration curve that ranged from 10 to 100 µg/mL.

2.5.3. Antioxidant Assays

Because antioxidant activity depends on the underlying reaction mechanism, radical-scavenging, reducing-power, and metal-based assays can rank the same extracts differently, and no single assay fully captures this behaviour; combining complementary responses gives a more reliable picture of an extract’s antioxidant potential [19,61,62]. Three complementary assays probed radical-scavenging and reducing mechanisms [63,64]. For ABTS, a 7 mM ABTS/2.45 mM potassium persulfate solution was incubated in the dark for 12–16 h, diluted with phosphate buffer (0.1 M, pH 7.4) to A₇₃₅ = 0.700 ± 0.05, and then mixed with 100 µL of extract (2.9 mL of solution); absorbance was read at 735 nm after 10 min [65]. For PFRAP, 2.5 mL phosphate buffer, 2.5 mL of 1% (w/v) potassium ferricyanide, and 1 mL of extract were incubated at 50 °C for 15 min; the reaction was stopped with 2.5 mL 10% (w/v) trichloroacetic acid, centrifuged (3000 rpm, 10 min), and 2.5 mL supernatant was mixed with 2.5 mL of water and 0.5 mL of 0.1% FeCl₃ before reading at 700 nm [61]. For SNPAC, 1 mM AgNO₃ was heated for 10 min, and then 10% (v/v) trisodium citrate (1% w/v) was added under stirring to form a colloidal solution; 100 µL of extract was added to 2.9 mL of SNPAC solution and absorbance was read at 423 nm after 10 min [66]. Ascorbic acid served as the standard for all three assays; results were expressed as mg ascorbic acid equivalents (AAE) per g DW. A calibration curve was constructed for each assay in the following ranges: 20–120 µg/mL for ABTS, 50–400 µg/mL for PFRAP, and 20–200 µg/mL for SNPAC. The calibration curves along with their corresponding R² are provided in Figure S1.

2.5.4. UPLC-ESI-MS/MS

UPLC-ESI-MS/MS analyses were performed on a Shimadzu LCMS-8040 ultra-high-sensitivity system equipped with a Nexera XR LC-20AD binary pump (Shimadzu LCMS-8040, Shimadzu Corporation, Kyoto, Japan). The electrospray ionisation conditions were as follows: collision-induced dissociation (CID) gas, 230 kPa; conversion dynode, −6.00 kV; desolvation line temperature, 250 °C; nebulising gas flow, 3.0 L/min; heat block temperature, 400 °C; and drying gas flow, 10.0 L/min. Separation was carried out on a Restek Ultra C18 column (3 µm, 150 × 4.6 mm) (Restek Corporation, Bellefonte, PA, USA) with a mobile phase consisting of water with 0.1% formic acid (solvent A) and methanol (solvent B) at 0.20 mL/min. The gradient programme was: 0–0.2 min, 98% A; 0.2–2.5 min, 98–25% A; 2.5–4.0 min, 25–0% A; 4.0–7.0 min, 0% A; 7.0–7.1 min, 0–98% A; 7.1–12.0 min, 98% A. The injection volume was 5 µL. Identification of phenolic compounds was performed in multiple reaction monitoring (MRM) mode using LabSolutions-LCMS software (Version 5.97; Shimadzu Corporation, Kyoto, Japan) to obtain the phenolic profiles of the identified constituents. Multiple reaction monitoring transitions were established in-house by injection of standards on the same instrument, with the precursor ion and the quantifier and qualifier product ions selected by automated optimisation for each compound. Compounds were identified by matching the precursor ion and both product ions to those of the corresponding standard acquired under identical ionisation conditions. Relative peak areas are reported for comparison between the two extracts and are semi-quantitative: they were not converted to concentrations by external calibration, and differences in ionisation efficiency between compounds mean that areas are comparable for a given compound across extracts but not between different compounds.

2.5.5. Cytotoxicity Assays

Cytotoxicity was assessed as a preliminary screening. Its purpose was to compare the two solvent systems and a reference antioxidant over a common concentration range. The assay was performed on four different mammalian cell lines: Vero (African green monkey kidney epithelium), BHK21 (baby hamster kidney fibroblast), MDBK (bovine kidney epithelium) and IBRS2 (porcine kidney epithelium) (National Institute of Veterinary Medicine, Algiers, Algeria), all continuous non-tumour lines of renal origin. They were selected on the basis of availability and were used to screen for non-specific cytotoxicity, not as models of a particular target tissue.

Combined optimal-condition extracts were evaporated at a bath temperature of 45 °C (30 mbar, Hei-VAP, Heidolph, Schwabach, Germany), then resuspended in phosphate-buffered saline (PBS, pH 7.4), and sterile-filtered (0.2 µm). The cell lines were maintained in DMEM with 10% FBS and 1% penicillin–streptomycin at 37 °C, 5% CO₂. Cells were seeded at 1 × 10⁴ cells/well in 96-well plates, allowed to adhere for 24 h, and then exposed to extracts (0.2–6.5 mg/mL, quadruplicate) for 24 h. Resazurin (20 µL, 0.15 mg/mL) was added and plates incubated 1–4 h at 37 °C before reading fluorescence (560 nm excitation/590 nm emission, Fluoroskan) [67]. Residual-ethanol controls (70% ethanol evaporated under the same conditions and redissolved in PBS) were included alongside PBS-only negative controls. Ascorbic acid served as the reference compound over the same concentration range. Half-maximal inhibitory concentration (IC₅₀) values were calculated by fitting a four-parameter logistic model on the SkanIt software (Version 5.0.0.42; Thermo Fisher Scientific, Waltham, MA, USA).

2.6. Multivariate Analysis

Multivariate statistical analyses were employed to elucidate the relationships between the extraction stages, phenolic and flavonoid contents, and antioxidant capacities. Pearson correlation analysis was conducted to quantify the linear associations between TPC, TFC, and the three complementary antioxidant responses (ABTS, PFRAP, and SNPAC) across all three stages for both solvent systems. To visualise the clustering of extraction stages and identify the dominant variables driving sample variance, Principal Component Analysis (PCA) was executed. Prior to PCA, all analytical data were standardised (mean-centred and scaled by unit variance) to ensure that variables with disparate units and magnitudes contributed equally to the model. The PCA biplots were generated to map the distribution of the sequential extraction stages against the variable loading vectors. All multivariate calculations, data standardisation, and visualisations were executed in a Python 3.14.7 environment utilising the pandas 2.3.3, scikit-learn 1.9.0, matplotlib 3.11.1, and seaborn 0.13.2 libraries.

3. Results and Discussion

3.1. Model Evaluation and Fitting

Second-order polynomial models were fitted to TPC, TFC and the three antioxidant responses across the three extraction stages for both solvent systems. All models were significant after removal of non-significant terms, with p-values ≤ 0.0063, with R² between 0.893 and 0.989 and adjusted R² between 0.872 and 0.982 (Table 2 and Table S1). Predicted R², obtained by leave-one-out cross-validation, ranged from 0.614 to 0.953 and was positive for every model, indicating that each model reproduces observations withheld from its estimation better than the response mean does. Lack-of-fit tests were non-significant (p > 0.05) for 28 of the 30 models, indicating that these second-order models adequately describe the relationship between extraction variables and responses within the tested domain and can support optimisation and interpretation of factor effects. Lack of fit was significant for two aqueous stage 2 models, TFC (p = 0.037) and PFRAP (p = 0.022). For both, the pure error estimated from the five centre-point replicates was the lowest of the five stage 2 responses (coefficients of variation of 1.9% and 1.8%, against 3.1–5.7% for TPC, ABTS and SNPAC), which makes the test sensitive to the deviation of a single run from the fitted surface. Residual analysis identified one such run in each design, at 60 °C/1 min for TFC and at 37.5 °C/10 min for PFRAP (externally studentised residuals of −6.0 and −4.7). When the models were refitted without the outlier run, neither model showed significant lack of fit (p = 0.66 and 0.18), and the stage 2 optimum was unchanged. As no experimental cause was found that would justify discarding either observation, both models were kept as fitted to all runs and were used with caution. They are not interpreted individually in Section 3.2 and enter the analysis only through the aqueous stage 2 desirability function, in which the fitted surfaces of all five responses reach their maximum at the same vertex of the design (60 °C, 10 min), so that the selected optimum does not depend on them.

Table 2. Goodness-of-fit statistics (R², R² adjusted, predicted R², lack-of-fit p-value) for response surface models describing aqueous and hydroethanolic extraction of M. pomifera leaves (TPC, TFC, ABTS, PFRAP, SNPAC) across three stages.

Response	Solvent	Stage 1 (R2/Adj./Pred./LoF p)	Stage 2 (R2/Adj./Pred./LoF p)	Stage 3 (R2/Adj./Pred./LoF p)
TPC	Aqueous	0.97/0.96/0.91/0.66	0.96/0.94/0.86/0.28	0.98/0.96/0.95/0.89
TPC	Hydroethanolic	0.98/0.97/0.95/0.32	0.95/0.92/0.87/0.42	0.92/0.90/0.85/0.38
TFC	Aqueous	0.96/0.91/0.81/0.87	0.98/0.97/0.86/0.04 a	0.98/0.97/0.93/0.40
TFC	Hydroethanolic	0.97/0.93/0.61/0.46	0.99/0.98/0.94/0.09	0.97/0.96/0.91/0.09
ABTS	Aqueous	0.98/0.96/0.84/0.30	0.89/0.87/0.82/0.68	0.94/0.93/0.90/0.93
ABTS	Hydroethanolic	0.90/0.87/0.79/0.13	0.97/0.95/0.90/0.25	0.98/0.97/0.93/0.42
PFRAP	Aqueous	0.98/0.96/0.83/0.53	0.93/0.91/0.83/0.02 a	0.97/0.96/0.92/0.42
PFRAP	Hydroethanolic	0.98/0.96/0.92/0.48	0.98/0.97/0.94/0.54	0.97/0.95/0.88/0.13
SNPAC	Aqueous	0.97/0.95/0.90/0.78	0.96/0.95/0.82/0.18	0.94/0.91/0.84/0.65
SNPAC	Hydroethanolic	0.94/0.91/0.85/0.71	0.98/0.96/0.87/0.58	0.99/0.98/0.93/0.08

ᵃ Significant lack of fit (p < 0.05).

Terms that were not significant at α = 0.05 were removed from the models, and their estimates and p-values in the full second-order fits are reported alongside the reduced models in Tables S3 and S4.

3.2. Factor Effects and Response Surface Morphology

The main significant regression coefficients are summarised in Table 3. Three-dimensional response surfaces illustrate the most influential factors for aqueous (Figure 2) and hydroethanolic (Figure 3) extraction. PFRAP and TFC broadly followed the trends of ABTS and SNPAC, respectively (Pearson |r| ≥ 0.8 in most cases, Section 3.3), and are discussed only briefly. The regression coefficients of the reduced models are given in Table 3. Complete statistics for every term, in both the full second-order models and the reduced models, are reported in Tables S3 and S4 for the aqueous and hydroethanolic systems, respectively; the corresponding coded polynomial equations are given in Table S5 and the complete design and response data in Table S2.

Table 3. Regression coefficients of the reduced response surface models, in coded units, for (a) the aqueous and (b) the hydroethanolic system at each extraction stage.

(a) Aqueous System
Stage	Response	Intercept	Temperature	Time		Temperature × Time		Temperature2		Time2
1	TPC	15.680 ***	1.488 ***	0.362 *		—		0.628 *		—
	TFC	7.927 ***	0.365	1.102 **		−0.825 **		−0.699 *		−0.439
	ABTS	39.583 ***	2.433 ***	0.370		−0.590 *		—		−0.676
	PFRAP	17.371 ***	0.785 ***	0.135		−0.305 *		−0.398 *		−0.358 *
	SNPAC	2.223 ***	0.112 **	0.147 ***		—		−0.191 **		−0.126 **
2	TPC	3.421 ***	0.725 ***	0.450 ***		0.207 *		−0.250 *		—
	TFC	2.239 ***	0.572 ***	0.305 ***		0.185 **		—		−0.164 *
	ABTS	7.441 ***	1.303 ***	0.620 **		—		—		—
	PFRAP	4.423 ***	0.803 ***	0.317 **		—		−0.306 *		—
	SNPAC	0.609 ***	0.128 ***	0.077 ***		0.037 *		—		−0.055 **
3	TPC	0.720 ***	0.250 ***	0.143 ***		0.102 **		0.093 **		—
	TFC	0.553 ***	0.360 ***	0.148 ***		0.072 *		0.134 ***		—
	ABTS	2.103 ***	0.670 ***	0.323 ***		—		0.310 **		—
	PFRAP	0.610 ***	0.268 ***	0.083 ***		—		0.065 *		—
	SNPAC	0.275 ***	0.087 ***	0.047 ***		0.022		—		—
(b) Hydroethanolic System
Stage	Response	Intercept	Temperature	Time	Ethanol	Temperature × Time	Temperature × Ethanol	Time × Ethanol	Temperature2	Time2	Ethanol2
1	TPC	13.626 ***	0.739 ***	−0.349 *	−3.483 ***	—	—	—	−0.897 **	−0.837 *	—
	TFC	7.783 ***	0.872 ***	0.589 ***	0.699 ***	0.372 *	0.412 **	0.427 **	−0.479 *	—	0.545 *
	ABTS	30.506 ***	3.684 ***	1.321 *	−4.849 ***	—	—	—	—	—	—
	PFRAP	18.128 ***	0.685 **	1.145 ***	−2.472 ***	—	0.659 **	—	−1.124 **	—	−2.059 ***
	SNPAC	2.925 ***	0.141 **	0.058	0.422 ***	—	—	—	—	−0.236 **	0.234 **
2	TPC	4.634 ***	0.504 ***	—	−0.353 ***	—	−0.404 ***	—	0.688 ***	—	−0.277 *
	TFC	3.448 ***	0.817 ***	0.658 ***	1.763 ***	—	0.316 **	0.464 ***	—	—	0.961 ***
	ABTS	4.577 ***	1.924 ***	0.776 ***	−0.050	0.759 ***	−0.419 *	—	—	—	−0.479
	PFRAP	2.312 ***	0.751 ***	0.378 ***	0.013	0.216 ***	—	—	—	—	0.165 *
	SNPAC	0.977 ***	0.233 ***	0.133 ***	0.365 ***	0.083 *	0.138 **	0.087 *	—	—	0.194 **
3	TPC	1.016 ***	0.358 ***	0.159 ***	0.094 *	—	—	—	—	—	—
	TFC	1.477 ***	0.841 ***	0.552 **	1.781 ***	—	1.114 ***	—	—	—	1.364 ***
	ABTS	3.130 ***	0.900 ***	0.380 ***	0.304 ***	0.140 *	—	—	−0.564 ***	—	—
	PFRAP	0.778 ***	0.308 ***	0.121 ***	0.205 ***	—	0.128 **	—	−0.130 *	—	0.185 **
	SNPAC	0.413 ***	0.201 ***	0.072 **	0.367 ***	−0.058 *	0.245 ***	—	—	—	0.276 ***

Coefficients are given in coded units. Significance of each coefficient: * p < 0.05, ** p < 0.01, *** p < 0.001. — indicates a term that was not retained after model reduction. Full statistics for the complete second-order models and for the reduced models are given in Tables S3 and S4.

Figure 2. Response surface plots for aqueous extraction: combined effects of extraction time and temperature on (a,b) TPC in stages 1 and 2, (c) ABTS radical-scavenging activity in stage 1, and (d) SNPAC in stage 1. Surface colour encodes the fitted response value, from low (blue) to high (red).

Figure 3. Response surface plots for hydroethanolic extraction: effects of ethanol concentration and temperature on (a,b) TPC in stages 1 (time = 5 min) and 3 (time = 10 min), (c,e) ABTS radical-scavenging activity in stages 1 (time = 5 min) and 3 (time = 10 min), and (d,f) SNPAC in stages 1 (time = 5 min) and 3 (time = 10 min). Surface colour encodes the fitted response value, from low (blue) to high (red).

3.2.1. Temperature and Extraction Time

Temperature was the dominant factor across all stages and both solvents, with strong positive linear effects on TPC, ABTS, and SNPAC and significant quadratic terms in several models (Table 3). Elevated temperature enhances polyphenol diffusion, solubility, and matrix permeability, but high temperatures can also lead to thermal degradation of the more labile constituents above a threshold that depends on the plant matrix and on the duration of exposure [55,68,69,70,71,72,73].

In the aqueous system, stage 1 TPC increased with both time and temperature, with a positive quadratic temperature term, favouring the recovery of water-soluble phenolics at higher temperatures and a moderately longer time (Figure 2a). ABTS, however, showed a negative quadratic time term and a negative time × temperature interaction (Figure 2c), indicating that prolonged extraction at high temperature reduced radical-scavenging activity even as TPC continued to rise, suggesting that ABTS-active constituents degrade faster than the bulk phenolic pool. SNPAC displayed a convex surface with negative quadratic terms for both factors (Figure 2d), peaking at intermediate conditions, which is consistent with the higher thermal sensitivity typically reported for flavonoids, which correlate strongly with SNPAC in this study [55,56,74,75,76,77]. In stages 2 and 3, significant time × temperature interactions for TPC (Figure 2b), ABTS, and SNPAC (Table 3) indicate that once the readily extractable fraction is depleted in stage 1, further recovery depends on the combined effect of thermal energy and longer contact time to mobilise less accessible phenolics [78,79,80].

In the hydroethanolic system, time had a weaker effect on stage 1 TPC, while temperature showed a strong positive effect on ABTS and a marked negative quadratic effect on TPC (Figure 3a,c), suggesting rapid extraction of ethanol-soluble phenolics with limited additional time dependence within the 1–10 min window. Unlike the aqueous system, the hydroethanolic stage 3 ABTS response showed a positive time × temperature interaction (Table 3; Figure 3e), with no evidence of thermal degradation, which is consistent with reports that, for some plant matrices, hydroalcoholic mixtures stabilise thermally labile flavonoids and phenolic acids more effectively than water under comparable conditions [71,81,82]. Overall, water-based extraction showed signs of ABTS-active compound degradation at high temperatures and long times, whereas the hydroethanolic system maintained or improved ABTS activity under the same conditions.

3.2.2. Solvent Composition

Ethanol concentration strongly influenced the hydroethanolic system. In stage 1, TPC and ABTS declined with increasing ethanol, reaching a maximum value at 30% v/v (Figure 3a,c), indicating that the most readily extractable phenolics in M. pomifera leaves are predominantly hydrophilic, consistent with trends reported for other plant matrices [83,84,85]. SNPAC showed the opposite pattern, increasing with ethanol content and achieving the highest recovery at 70% v/v (Figure 3d), suggesting that constituents responsible for silver-ion reduction and nanoparticle stabilisation are less polar [86,87,88].

By stage 3, the TPC–ethanol relationship reversed to a positive linear dependence, while SNPAC showed a curvilinear increase with ethanol (Figure 3b,f), and ABTS also favoured higher ethanol content (Figure 3e). This shift indicates that once the most polar fraction is depleted in earlier stages, the residual pool is enriched in less polar phenolics and flavonoids that are better solubilised at higher ethanol contents. This pattern is reported in other multistage extraction studies, where fractions extracted using solvents with high ethanol content favoured flavonoid recovery [89].

Together, the time–temperature and solvent–composition patterns support a compositional partitioning between polar phenolics recovered early in water-rich conditions and less polar constituents recovered later under ethanol-rich conditions, contributing more strongly to TFC and SNPAC. This suggests that multistage extraction can act as a simple fractionation approach, enriching early aqueous stages in more polar phenolics and later ethanol-rich stages in flavonoid-rich, high-SNPAC fractions.

3.3. Relationships Between Phenolic Content and Antioxidant Responses

Pearson correlations and principal component analysis (PCA) were used to examine relationships between phenolic contents (TPC, TFC) and antioxidant responses (ABTS, PFRAP, SNPAC) across extraction stages. Correlation heatmaps (Figure 4) summarise pairwise relationships across all three stages in both solvent systems, and PCA biplots on standardised data (Figure 5) visualise how these relationships structure the overall response space and separate the extraction stages.

Figure 4. Correlation heatmaps for aqueous (stages 1, 2, 3) and hydroethanolic (stages 1, 2, 3) extraction, showing pairwise Pearson coefficients among TPC, TFC, ABTS, PFRAP, and SNPAC.

Figure 5. PCA biplots of the aqueous and hydroethanolic extracts, coloured by stage (standardised variables).

Across both solvents and all three stages, TPC correlated strongly with ABTS (r = 0.84–0.95), consistent with the well-documented association between phenolic content and radical-scavenging capacity in mixed hydrogen atom transfer (HAT) and electron-transfer assays [62,90,91,92]. TFC and SNPAC showed a comparable strong positive association that strengthened progressively from stage 1 (r = 0.78–0.79) through stage 2 to stage 3 (r = 0.97–0.98), while the TPC–SNPAC relationship diverged sharply between solvents in stage 1 (aqueous r = 0.44; hydroethanolic r = −0.73), narrowed in stage 2, and converged towards strongly positive values by stage 3. This pattern indicates that flavonoids, rather than total phenolics broadly, are the principal contributors to silver-ion reduction and nanoparticle stabilisation [93,94].

The PCA biplots corroborate this structure. In both solvent systems, samples separated along PC1 by extraction stage, reflecting the progressive depletion of readily extractable solutes described in Section 3.2, with stage 2 samples consistently occupying an intermediate position between stages 1 and 3. Loading vectors for TPC, ABTS, and PFRAP clustered together in both systems, consistent with their consistently strong pairwise correlations, while TFC and SNPAC formed a distinct cluster, particularly in the hydroethanolic system—visually confirming the flavonoid-driven SNPAC association identified above. The weak TPC–TFC coupling evident in stage 1 (r = 0.34 aqueous; r = −0.32 hydroethanolic) converged to near-uniform positive correlations (r > 0.94 for most pairs) by stage 3 in both systems, indicating that the residual extractable fraction behaves as a compositionally homogeneous pool released in parallel.

Overall, TPC is the strongest correlate of the ABTS and PFRAP responses, while TFC is the strongest correlate of SNPAC, with the two pools becoming increasingly co-extracted as the multistage process progresses.

These correlations describe the chemical reactivity of the extracts under defined assay conditions. ABTS, PFRAP and SNPAC each probe a different reaction mechanism and none reproduces the conditions of a biological system, so the associations reported here should not be read as evidence of antioxidant efficacy in vivo [64,65].

3.4. Optimisation and Stage-Wise Contributions

Optimal conditions identified as described in Section 2.3 were validated experimentally (Table 4). Relative deviations between predicted and measured responses remained below 12% for all response and stage combinations, and below 1% for five responses, including TPC and ABTS in the aqueous system at stage 3.

Table 4. Experimental validation of optimised extraction conditions: measured responses and relative prediction error at each stage.

Solvent	Stage	Optimal Conditions	Response	Predicted (95% CI)	Experimental (Mean ± SD)	Error (%)
Aqueous	1	60 °C, 5.4 min	TPC (mg GAE/g DW)	17.79 ± 0.39	16.15 ± 0.42	9.2
		D = 0.84	TFC (mg QE/g DW)	7.59 ± 0.72	7.43 ± 0.30	2.1
			ABTS (mg AAE/g DW)	42.02 ± 0.71	40.52 ± 0.83	3.6
			PFRAP (mg AAE/g DW)	17.76 ± 0.30	17.27 ± 0.40	2.8
			SNPAC (mg AAE/g DW)	2.14 ± 0.08	2.35 ± 0.12	9.8
	2	60 °C, 10 min	TPC (mg GAE/g DW)	4.55 ± 0.31	4.02 ± 0.25	11.7
		D = 0.93	TFC (mg QE/g DW)	3.14 ± 0.18	2.85 ± 0.58	9.1
			ABTS (mg AAE/g DW)	9.36 ± 0.55	8.66 ± 0.40	7.5
			PFRAP (mg AAE/g DW)	5.24 ± 0.32	4.89 ± 0.27	6.6
			SNPAC (mg AAE/g DW)	0.80 ± 0.05	0.76 ± 0.09	4.5
	3	60 °C, 10 min	TPC (mg GAE/g DW)	1.31 ± 0.09	1.32 ± 0.09	0.8
		D = 0.91	TFC (mg QE/g DW)	1.27 ± 0.09	1.27 ± 0.05	0.2
			ABTS (mg AAE/g DW)	3.41 ± 0.25	3.41 ± 0.10	0.1
			PFRAP (mg AAE/g DW)	1.03 ± 0.07	1.04 ± 0.02	1.3
			SNPAC (mg AAE/g DW)	0.43 ± 0.04	0.43 ± 0.03	0.3
Hydroethanolic	1	54 °C, 5 min, 30% EtOH	TPC (mg GAE/g DW)	17.20 ± 0.62	15.21 ± 0.42	11.6
		D = 0.93	TFC (mg QE/g DW)	7.66 ± 0.46	7.92 ± 0.31	3.4
			ABTS (mg AAE/g DW)	37.91 ± 1.94	36.89 ± 0.72	2.7
			PFRAP (mg AAE/g DW)	17.83 ± 0.72	18.81 ± 0.64	5.5
			SNPAC (mg AAE/g DW)	2.83 ± 0.18	2.72 ± 0.21	3.9
	2	60 °C, 10 min, 30% EtOH	TPC (mg GAE/g DW)	6.30 ± 0.28	6.46 ± 0.12	2.5
		D = 0.96	TFC (mg QE/g DW)	3.34 ± 0.44	3.68 ± 0.32	10.2
			ABTS (mg AAE/g DW)	8.03 ± 0.75	8.25 ± 0.67	2.8
			PFRAP (mg AAE/g DW)	3.81 ± 0.20	3.86 ± 0.33	1.3
			SNPAC (mg AAE/g DW)	1.03 ± 0.17	1.01 ± 0.19	1.9
	3	60 °C, 10 min, 70% EtOH	TPC (mg GAE/g DW)	1.63 ± 0.14	1.72 ± 0.16	5.7
		D = 0.91	TFC (mg QE/g DW)	7.13 ± 0.64	7.49 ± 0.43	5.1
			ABTS (mg AAE/g DW)	4.29 ± 0.27	4.44 ± 0.04	3.5
			PFRAP (mg AAE/g DW)	1.59 ± 0.13	1.70 ± 0.16	6.6
			SNPAC (mg AAE/g DW)	1.52 ± 0.12	1.51 ± 0.01	0.4

Predicted values are given with a 95% confidence interval. Experimental values are the mean ± standard deviation of three independent extractions carried out at the stated conditions. The relative error is the deviation of the measured value from the predicted value, expressed as a percentage. D denotes the combined desirability value.

In the aqueous system, stage 1 was optimised at 60 °C/5.4 min (TPC 16.15 ± 0.42 mg GAE/g DW, ABTS 40.52 ± 0.83 mg AAE/g DW); stages 2–3 were both optimised at 60 °C/10 min. Cumulative yields reached 21.49 mg GAE/g DW (TPC), 52.59 mg AAE/g DW (ABTS), and 23.20 mg AAE/g DW (PFRAP). In the hydroethanolic system, stage 1 was optimised at 54 °C/5 min/30% ethanol (TPC 15.21 ± 0.42 mg GAE/g DW, ABTS 36.89 ± 0.72 mg AAE/g DW); stages 2–3 were optimised at 60 °C/10 min, with ethanol at 30% (stage 2) and 70% (stage 3). Cumulative yields reached 23.39 mg GAE/g DW (TPC), 19.09 mg QE/g DW (TFC), and 49.58 mg AAE/g DW (ABTS), notably higher in TFC than the aqueous system, consistent with Section 3.2’s flavonoid enrichment in ethanol.

Stage 1 TPC yields with pure water (16.15 mg GAE/g DW) and 30% ethanol (15.21 mg GAE/g DW) were of comparable magnitude, as were the corresponding ABTS and PFRAP values, and optimal time/temperature conditions were similar across both solvents at every stage. This indicates that water alone can recover the most polar phenolic fraction as effectively as dilute ethanol. Water is the cheaper and more readily available solvent, although establishing an environmental advantage would require the recovery and energy consumption of each to be quantified.

Stage-wise contribution analysis (Figure 6) showed that stage 1 accounted for approximately 65–77% of cumulative recovery across most responses, stage 2 for approximately 16–28%, and stage 3 for a marginal 4.5–12% additional gain, except for TFC and SNPAC in the hydroethanolic system, where stage 3 contributed a substantially larger share (39.2% and 28.8%, respectively). This diminishing-return pattern reflects the progressive depletion of easily accessible solutes typical of multistage extraction [95,96,97].

Figure 6. Relative contribution of stages 1, 2, and 3 to the cumulative recovery of phenolics and antioxidant activities (TPC, TFC, ABTS, PFRAP, SNPAC) in (a) aqueous and (b) hydroethanolic multistage extraction of M. pomifera leaves. Error bars represent the propagated standard deviation (n = 3 independent extractions) at each stage.

On a solvent-demand basis, the full three-stage process consumes 30 mL of solvent per gram of feed material, comparable to the solid-to-liquid ratios commonly applied in single-stage conventional maceration rather than lower. The advantage of the crosscurrent configuration therefore is not solvent volume reduction but the reduction in contact time and operating temperature, the ability to mitigate thermal degradation, and the option of stopping after two stages. The appropriate number of stages therefore differs between the two solvent systems. For the aqueous process, limiting the extraction to two stages reduces solvent demand to 20 mL/g while retaining more than 90% of the cumulative TPC and ABTS within roughly 15 min of contact time, improving solvent intensity from 1.40 to 0.99 mL per mg GAE recovered. For the hydroethanolic system, the pronounced stage 3 contribution to TFC and SNPAC implies that the ethanol-rich final stage acts as a targeted fractionation step for recovering a flavonoid-enriched, high-SNPAC fraction rather than simply exhausting residual bulk phenolics, and it is retained on that basis. Distributing the contact time across successive short stages also limits the loss of radical-scavenging activity observed when high temperature is combined with prolonged contact within a single stage (Section 3.2.1).

To assess how much extractable material remained after three stages, a single unoptimised extraction was carried out on the final residue under deliberately exhaustive conditions (60 °C, 60 min), using water and, for the hydroethanolic system, 70% v/v ethanol matching the stage 3 optimum. This recovered only approximately 0.25 and 0.21 mg GAE/g DW respectively, corresponding to 1.1% and 0.8% of the cumulative totals (Table S7). Although this fourth stage was not optimised, the extraction time was substantially longer than any design run. The small quantity recovered suggests that the three-stage process captures approximately 99% of the phenolic material extractable under these conditions, and indicates that additional stages would not be productive. The reported cumulative values can therefore be interpreted as near-exhaustive rather than as the sum of three arbitrary fractions.

When compared to other species, the cumulative TPC obtained here (≈23 mg GAE/g DW) falls within the mid-upper range reported for aqueous or ethanolic leaf extracts from other species under conventional extraction (typically 10–20 mg GAE/g DW) [58,98,99], but remains slightly lower than the values achieved with more intensive methods such as deep eutectic solvents or ultrasound assistance (40–80 mg GAE/g DW) [100,101], yet without requiring specialised solvents or equipment. These results demonstrate that a compact multistage scheme recovers most of the antioxidant phenolic pool within short contact times, which is relevant for informing preliminary process design considerations. However, further work is needed at pilot scale to assess solvent recovery, energy demand, and operational feasibility before drawing conclusions on industrial scale-up.

Studies focusing on the life-cycle assessment of phenolic recovery from woody and agro-industrial residues indicate the material and energy needs for such processes. In an assessment of polyphenol extraction from spruce bark, electricity used for heating dominated most impact categories, followed by the production and use of ethanol as solvent [102]. A comparison of extraction techniques applied to pine needles reached a related conclusion; simple maceration carrying the lowest impact despite its long duration and microwave-assisted extraction the highest, since electricity consumption rather than contact time governed the result [103]. Analyses of agro-industrial residues have similarly identified solvent production and electrical energy as the principal contributors, with solvent recovery as the main means of reducing them [104,105]. The process described here operates at or below 60 °C for contact times of ten minutes or less and requires no specialised equipment, and in its aqueous configuration uses no organic solvent at all. The aqueous route avoids the solvent production requirements and confines energy demand to moderate heating, whereas the hydroethanolic route incurs environmental impacts associated with solvent use and production that would have to be mitigated through recovery, which was not attempted here. To compare the present process to the studies cited, a life-cycle inventory would be required, which was not collected in this work.

3.5. Chemical Profiles

UPLC-ESI-MS/MS identified nineteen phenolic and related constituents across the optimised extracts (Table 5; full qualifier ion transitions in Table S6), spanning phenolic acids, flavonoids, and other secondary metabolites. The comparisons that follow are semi-quantitative: peak areas were not converted to concentrations, and because ionisation efficiency differs between compounds, they indicate the relative distribution of a given compound between the two extracts but do not support comparison of abundance between different compounds.

Table 5. UPLC-ESI-MS/MS identification of phenolic and related constituents in optimised aqueous and hydroethanolic M. pomifera leaf extracts. ND: not detected.

Compound	ESI	Precursor (m/z)	Quantifier (m/z)	Aq. RT (min)	Hy. RT (min)	Aq. Area	Hy. Area
Phenolic acids
Ferulic acid	+	195.00	38.65	2.949	2.953	795,532	1,332,750
Sinapic acid	+	225.00	207.15	3.853	4.613	3,938,154	2,385,705
2-Methoxybenzoic acid	+	153.05	135.00	4.365	4.364	646,889	265,713
Salicylic acid	−	137.20	92.95	4.674	4.679	188,177	83,365
p-Coumaric acid	−	163.05	118.95	4.226	4.278	15,011	12,550
Caffeic acid	−	179.15	135.00	8.507	8.511	18,942	33,659
Flavonoids
Naringenin	+	273.05	153.00	5.174	4.221	23,168	152,519
Quercetin	+	303.05	153.10	4.206	4.464	2843	56,594
Rutin	+	611.00	303.10	4.215	4.237	90,964	429,730
Luteolin	+	286.75	153.00	5.104	4.805	10,839	2983
Chrysin	+	254.85	152.80	4.861	ND	13,706	ND
Epicatechin	+	290.80	139.15	ND	3.464	ND	761,150
Other secondary metabolites
Vanillin	+	153.05	64.75	4.144	4.143	1,132,341	300,270
Thymol	+	151.15	108.90	4.843	4.717	103,284	77,796
Kojic acid	+	142.75	69.05	5.637	ND	53,276	ND
Resveratrol	+	229.05	107.10	4.984	4.989	103,442	151,022
Curcumin	+	368.90	177.05	5.080	5.086	23,758	16,840
Oleuropein	+	540.50	523.40	5.590	5.592	99,292	80,775
Oleanolic acid	+	457.20	411.05	6.488	3.902	1,411,612	1,193,094
β-Carotene	+	537.10	280.95	7.910	7.923	1,335,943	381,112

A comparison of relative peak areas revealed distinct solvent selectivity, governed by the dielectric properties of the extraction media and by the physicochemical features of the individual analytes. Among phenolic acids and simple aromatics, the aqueous extract showed higher relative abundances of vanillin, sinapic acid, 2-methoxybenzoic acid and salicylic acid. These are low-molecular-weight compounds in which polar functional groups, principally carboxyl and aldehyde moieties, account for a large proportion of the molecular surface, and the high dielectric constant and hydrogen-bond-donor capacity of water favour the solvation of such compact, strongly polar structures [21,44]. Caffeic and ferulic acids were the two members of this group recovered preferentially in the hydroethanolic system. Within the hydroxycinnamic acids the distribution tracked the availability of free hydroxyl groups ortho to the ring hydroxyl: caffeic acid, bearing an unsubstituted catechol, and ferulic acid, in which one position is methylated, favoured the hydroethanolic extract, whereas p-coumaric acid, which lacks ortho substitution, was distributed similarly between the two systems, and sinapic acid, in which both positions are methylated, favoured the aqueous extract. Free ortho-hydroxyl groups form strong hydrogen bonds with cell-wall polysaccharides, and their retention appears to shift the balance towards the ethanol-containing medium. Chrysin and kojic acid were detected only in the aqueous extract.

The hydroethanolic extract was consistently enriched in flavonoids, including naringenin, quercetin and rutin, with epicatechin detected exclusively in this system. The C6–C3–C6 skeleton confers an amphiphilic character, combining several hydrogen-bonding hydroxyl groups with a comparatively apolar ring framework, so that neither pure water nor pure ethanol provides an optimal solvation environment. The addition of ethanol lowers the bulk dielectric constant of the medium to an intermediate value better matched to this dual character, and simultaneously weakens the hydrogen bonding that retains these compounds within the matrix, enhancing their recovery [21,83]. The behaviour observed here is not ordered by lipophilicity: rutin, the most polar flavonoid identified, was recovered preferentially in the hydroethanolic system, indicating that the relevant factor is the accommodation of hydroxyl functionality rather than hydrophobic surface area alone. A comparable distinction is apparent in the flavonoid skeleton itself. Quercetin and luteolin share the same hydroxylation pattern and catechol B-ring and differ only by the hydroxyl at C-3, yet were recovered preferentially in opposite systems; rutin, epicatechin and naringenin, each carrying a C-3 substituent or a saturated C2–C3 bond, followed quercetin, while chrysin, the remaining unsubstituted flavone, followed luteolin. These findings agree with reports that ethanol-containing mixtures favour the extraction of flavonoid aglycones and glycosides in Moraceae and related taxa [106,107,108].

Luteolin therefore diverged from the broader class trend, showing higher relative abundance in the aqueous extract, which indicates that solvent selectivity is not reducible to class membership. Two explanations are consistent with the present data. Its 3′,4′,5,7-tetrahydroxylated structure may support an extensive hydrogen-bonded hydration shell that stabilises the molecule in aqueous medium. Alternatively, the result may be analytical in origin: more polar luteolin glycosides, preferentially recovered in water, could undergo in-source fragmentation and contribute to the monitored transition at m/z 286.75 → 153.00. Distinguishing between these possibilities would require chromatographic resolution of the intact glycosides. The predominance of most flavonoids in the hydroethanolic extract nevertheless aligns with the higher total flavonoid content and stronger SNPAC response recorded for this system (Section 3.2, Section 3.3 and Section 3.4), both determined independently of the chromatographic analysis.

Among other secondary metabolites, resveratrol gave a moderately higher relative peak area in the hydroethanolic extract, while curcumin and oleuropein were similar in both systems, and oleanolic acid was recovered preferentially in the aqueous extract. Glycosylation was not determining, since rutin favoured the hydroethanolic extract while oleuropein, also a glycoside, favoured the aqueous system. Given the number of compounds identified within each class, the structural relationships described above are presented as observations consistent with the present data rather than as established structure–solvent relationships.

Overall, despite comparable antioxidant activities between the two solvent systems (Section 3.1, Section 3.2, Section 3.3 and Section 3.4), the chemical profiles differed markedly: aqueous extraction favoured polar phenolic acids and simple aromatics, while hydroethanolic extraction favoured most flavonoids, supporting the compositional fractionation described in Section 3.4.

3.6. Cytotoxicity of the Optimised Extracts in Mammalian Cell Lines

The pooled optimised aqueous and hydroethanolic extracts from all three stages were evaluated in BHK21, IBRS2, MDBK, and Vero cells using a resazurin assay after 24 h of exposure to concentrations ranging from 0.2 to 6.5 mg/mL. Dose–response curves were fitted using a four-parameter logistic model, and the corresponding IC₅₀ values are reported in Table 6 and illustrated in Figure 7.

Table 6. IC₅₀ (mg/mL) of the pooled aqueous and hydroethanolic extracts, and of ascorbic acid, in BHK21, IBRS2, MDBK and Vero cell lines after 24 h of exposure.

Cell Line	Aqueous Extract (mg/mL)	Hydroethanolic Extract (mg/mL)	Ascorbic Acid (mg/mL)
BHK21	6.42	3.10	0.57
IBRS2	5.47	3.02	0.39
MDBK	5.40	0.70	<0.20
Vero	5.19	1.60	0.36

IC₅₀ values were calculated using four-parameter logistic dose–response models. <0.20 indicates an IC₅₀ below the lowest concentration tested.

Figure 7. Dose–response curves for the optimised aqueous and hydroethanolic M. pomifera leaf extracts and ascorbic acid in BHK21, IBRS2, MDBK, and Vero cells after 24 h of exposure, fitted using a four-parameter logistic model.

The hydroethanolic extract was more cytotoxic than the aqueous extract in all four cell lines. Its IC₅₀ values were 3.10 mg/mL in BHK21, 3.02 mg/mL in IBRS2, 0.70 mg/mL in MDBK, and 1.60 mg/mL in Vero. The aqueous extract was less cytotoxic, with IC₅₀ values of 6.42, 5.47, 5.40, and 5.19 mg/mL in BHK21, IBRS2, MDBK, and Vero, respectively. MDBK cells were the most sensitive to the hydroethanolic extract, while BHK21 cells were the least sensitive among the hydroethanolic-treated lines.

The residual–ethanol control did not differ from the PBS control, indicating that the stronger response observed for the hydroethanolic extract was not attributable to residual solvent under the conditions tested. Ascorbic acid produced the lowest IC₅₀ values in all cell lines, with values of 0.57 mg/mL in BHK21, 0.39 mg/mL in IBRS2, <0.20 mg/mL in MDBK, and 0.36 mg/mL in Vero.

These results indicate that both plant extracts were less cytotoxic than ascorbic acid over the tested concentration range, although the hydroethanolic extract showed a clearly stronger effect than the aqueous extract. Ascorbic acid served as a reference for non-specific cytotoxicity rather than as a safety threshold. The greater response of the hydroethanolic extract may reflect its higher abundance of flavonoids and less polar constituents, but the present data do not identify the compounds responsible for the effect. These findings are consistent with other reports on M. pomifera, in which fruit extracts showed selective antiproliferative or pro-apoptotic effects in malignant cell lines and lower cytotoxicity on normal cell lines [16,109,110,111,112]. Data for the leaves remain scarce. Aqueous leaf extracts have been reported to cause no significant loss of viability or change in morphology in HEp-2 cells at concentrations up to 25 mg/mL over 24 h, while retaining antibacterial and antiviral activity [113]. These concentrations are not directly comparable with those reported here, since the cell line, the viability endpoint and the extract preparation all differ, but the direction of that result is consistent with the comparatively high IC₅₀ values obtained for the aqueous extract in the present work.

The present cytotoxicity results should be interpreted as a preliminary in vitro screening rather than a full safety assessment. The resazurin assay primarily reflects cellular metabolic activity and does not reveal the mechanism of the response. In addition, the assay was limited to 24 h exposure and four cell lines. All four lines are of renal origin, and the panel does not include models of the epithelia that would be exposed first in the food, cosmetic and materials uses envisaged for these extracts. The results accordingly a support comparison between the two extracts and the reference antioxidant under common conditions, and do not constitute a tissue-specific or application-specific safety assessment.

Further work should therefore include additional non-tumour and primary cell models, longer or repeated exposures, and complementary assays addressing membrane integrity, apoptosis, and oxidative stress before any application-specific safety conclusions are drawn.

4. Conclusions

This study demonstrated the feasibility of a short, laboratory-scale crosscurrent extraction process for the valorisation of Maclura pomifera pruning leaves as a source of antioxidant-rich extracts. Water and aqueous ethanol produced comparable cumulative total phenolic contents, ABTS capacities, and PFRAP responses, whereas the hydroethanolic process gave higher total flavonoid content and SNPAC activity. These differences indicate that the choice of solvent and the sequence of extraction stages can be used not only to increase recovery but also to influence the composition and in vitro antioxidant characteristics of the resulting fractions.

The first stage accounted for approximately 65–77% of cumulative recovery for most measured responses, and the second stage provided a further 16–28%. In the aqueous process, these results indicate that two stages were sufficient to recover more than 90% of the cumulative TPC and ABTS response within approximately 15 min of effective contact time. The third stage therefore offered limited additional bulk phenolic recovery in water. In the hydroethanolic process, however, the ethanol-rich third stage contributed 39.2% of the cumulative flavonoid response and 28.8% of SNPAC, supporting its use as a targeted step for recovering a flavonoid-enriched fraction rather than simply as an additional exhaustion stage.

UPLC–ESI–MS/MS confirmed that the two solvent systems generated chemically different extracts. The aqueous extract generally contained higher relative levels of selected phenolic acids and simple aromatics, whereas the hydroethanolic extract was richer in several flavonoids, including naringenin, quercetin, rutin, and epicatechin. Luteolin did not follow this general trend and was more abundant in the aqueous extract, illustrating that solvent selectivity cannot be predicted solely from broad compound classes. The cytotoxicity results are preliminary. Both extracts gave higher IC₅₀ values than ascorbic acid in the four cell lines after 24 h of exposure. This comparison was obtained under a single set of experimental conditions and does not establish the definitive safety of either extract.

Overall, the results support crosscurrent aqueous extraction as a simple route for recovering most of the polar antioxidant fraction from M. pomifera pruning leaves, with an optional ethanol-rich stage when flavonoid-oriented fractionation is desired. The process remains at laboratory scale, and the present work does not constitute an assessment of its environmental performance. Further studies should now address process scale-up to the pilot scale, solvent recycling to evaluate and enhance process sustainability, full gravimetric tracking of the solid phase to close a complete mass balance, and the valorisation of the residual solids, alongside comprehensive environmental impact assessments, specifically quantifying mass and energy balances, solvent recycling efficiencies, and life-cycle impacts. Additionally, broader toxicological and application-oriented studies are needed to define safe and effective use levels of the different fractions in specific food, cosmetic or green materials formulations.