Microwave-assisted extraction enhances wet-heating Maillard glycation of lupin protein with galacto-oligosaccharides: structural basis for improved interfacial functionality and gut fermentability

Abstract

Lupin protein isolate (LPI) is limited in food use by poor solubility and weak interfacial behavior. This study tested whether the extraction route conditions LPI for subsequent glycation. LPI was recovered by microwave-assisted extraction, applied only during extraction, and conjugated with galacto-oligosaccharides (GOS) by wet heating at 90 °C for 15 min. Spectroscopic, electrophoretic, thermal and microscopic analyses confirmed GOS attachment and partial unfolding. The conjugate from microwave-extracted LPI reached the highest apparent degree of grafting (22%), reduced particle size from 1775.1 to 89.7 nm and surface hydrophobicity from 443.3 to 62.5 arbitrary units, and lowered β-sheet content. Foaming (64%/36%), emulsifying activity (74.2 m²/g) and DPPH scavenging (79.6%) improved accordingly. In fecal fermentation it retained more acetate at 24 h (420.8 versus 212.1 μmol/L for the control) and gave the highest gas volume (13.9 mL) and microbial evenness. Extraction history therefore governs conjugate functionality.

Highlights

• •Microwave-assisted extraction enhanced Maillard conjugation of lupin protein.
• •Conjugation improved solubility, foaming, and emulsifying performance.
• •Glycation lowered β-sheet and raised β-turn and random-coil content.
• •α-Glucosidase inhibition rose from 17% to 60–66% after conjugation.
• •Conjugates retained more acetate and shifted microbiota toward beneficial taxa.

1. Introduction

Lupin protein isolate (LPI) has gained considerable attention as a promising plant-based protein source due to its high nutritional profile, well-balanced amino acid composition, favorable functional properties, and relatively low allergenic potential compared to traditional sources such as soy and pea proteins (Burgos-Díaz et al., 2019). Despite these advantages, the broader application of LPI in food formulations remains limited due to its poor solubility, reduced emulsification ability, and tendency to aggregate under processing conditions (Ma et al., 2024). To enhance the functional performance of LPI, glycation via the Maillard reaction has emerged as a viable non-enzymatic strategy. This reaction involves the covalent bonding of reducing sugars to nucleophilic amino groups on proteins, thereby improving solubility, antioxidant activity, and interfacial properties (Ma et al., 2025). Galacto-oligosaccharides (GOS), a class of prebiotic carbohydrates derived from lactose, are particularly suitable as glycation agents due to their ability to modify protein functionality while imparting prebiotic benefits. Through selective glycosylation, GOS can influence protein conformation, enhance emulsifying capacity, and promote favorable interactions with gut microbiota (Mao et al., 2024; Xu et al., 2024).

The Maillard reaction proceeds through well-defined chemical stages, beginning with the formation of Schiff bases between galacto-oligosaccharides (GOS) and free amino groups of proteins or peptides, followed by Amadori rearrangement products and, at advanced stages, the generation of melanoidins (de Oliveira et al., 2016; Ke & Li, 2023). When this reaction is carefully controlled, it can beneficially modify protein structure and functionality, as evidenced by enhanced emulsifying properties in casein–GOS conjugates (Li, Chen, et al., 2024). In contrast, excessive or poorly controlled heating promotes undesirable reactions, including excessive browning, protein crosslinking, and the formation of advanced glycation end-products (AGEs), which are linked to reduced nutritional and functional quality (Li, Jiang, et al., 2024). In this context, optimizing the protein extraction step prior to glycation is critical to ensure efficient conjugation and to preserve desirable functional attributes. Microwave-assisted extraction (MAE) has therefore gained attention as a promising alternative to conventional extraction methods, as it markedly reduces processing time while largely maintaining protein integrity, creating more favorable conditions for subsequent Maillard reactions. In combination with other emerging technologies such as ultrasound and pulsed electric fields, MAE has been shown to promote Maillard conjugate formation while limiting the generation of harmful by-products (Dumitrașcu et al., 2023; Ke & Li, 2023). Despite these advantages, the specific influence of MAE-induced structural modifications on Maillard reaction kinetics and the resulting functional properties of LPI–GOS conjugates remains insufficiently understood. Recent work on plant protein–polysaccharide systems has shown that the thermal and physical history of the protein before conjugation, rather than the heating step alone, governs the extent of Maillard coupling and the properties of the conjugate that forms (Khoshdouni Farahani et al., 2026).

Conjugation using wet-heating Maillard glycosylation is an established technique that enables controlled protein modification, reduces denaturation risks, and limits the formation of AGEs (Nomi et al., 2022). The inclusion of GOS in conjugation strategies not only contributes to functional improvements in protein properties but also exerts prebiotic effects by promoting the proliferation of beneficial gut bacteria such as Bifidobacteria and Lactobacilli. This activity increases the production of short-chain fatty acids (SCFAs), mainly acetate, propionate and butyrate, which are crucial for maintaining gut integrity and metabolic balance (Peled et al., 2024).

The composition and metabolic output of the gut microbiota respond to the chemical form in which a fermentable substrate reaches the colon. Chain length, glycosidic linkage, solubility and whether the sugar is free or bound to a protein carrier all influence which taxa can access it, how quickly it is fermented and the SCFA profile that results (Lin et al., 2025; Maqsood et al., 2025). Oligosaccharides recovered from pulse-processing streams have drawn attention on this basis, since their structure differs from that of commercial prebiotics and shifts the fermentation pattern accordingly (Yang et al., 2026). Covalent attachment of an oligosaccharide to a protein is therefore not a neutral modification from the microbial perspective: it alters both the accessibility of the sugar and the rate at which it becomes available during fermentation.

Recent advances have demonstrated that protein–galacto-oligosaccharide (GOS) conjugates can markedly improve protein solubility and bioavailability (Liu, Fan, et al., 2025), thereby enhancing their potential to act as effective prebiotic ingredients. Such conjugation strategies offer a promising approach for developing functional food components capable of modulating the gut microbiota more efficiently than free GOS alone. Nevertheless, despite the well-established prebiotic benefits of free GOS, the metabolic and microbiological consequences of GOS in conjugated forms remain insufficiently explored. In parallel, although glycation has been widely investigated in dairy, whey and soy protein systems, studies on lupin protein isolate (LPI) glycation with GOS under controlled Maillard conditions are still scarce (Liao et al., 2021; Zhong et al., 2021). The gap, however, is more specific than a shortage of reports. Microwave-assisted extraction is known to raise protein yield and purity (Işçimen & Hayta, 2022), but it remains unknown whether the conformational state in which the protein leaves the extraction step is what sets the rate and extent of the later glycation reaction, or whether the wet-heating step overrides that history entirely. Resolving this distinguishes extraction from conjugation as the controlling variable and determines where such a process would need to be optimized. The same question applies downstream. Hu et al. (2024) showed that the physiological response to GOS depends on the form in which it arrives in the colon, so protein-bound GOS may ferment differently from free GOS, yet the microbial consequences of the conjugated form remain largely untested.

This study addresses that gap by evaluating the structural, functional and gut microbial responses of LPI–GOS conjugates prepared by wet-heating Maillard glycosylation from either conventionally extracted or microwave-extracted LPI. The work first examined how microwave-assisted extraction of LPI influences the subsequent wet-heating glycation with GOS, covering the apparent degree of glycation, Maillard reaction progression, structural transformation and antioxidant capacity. It then characterized the physicochemical and functional consequences of conjugation. Finally, following the argument of Hu et al. (Hu et al., 2024) that the delivered form of GOS governs its physiological effect, it assessed how the conjugates modulate gut microbiota composition and SCFA profiles during in vitro fecal fermentation. The underlying hypothesis was that microwave-assisted extraction leaves LPI in a partially unfolded state with more accessible amino groups, so that an identical wet-heating treatment yields a more extensively glycated conjugate with better interfacial behavior and greater fermentability than one prepared from conventionally extracted LPI.

2. Materials and methods

2.1. Alkaline extraction methods for lupin protein isolation

The alkaline extraction method for lupin protein isolation was performed following the procedures of Bedin et al. (2020) with slight modifications. Briefly, lupin flour was mixed with water at a ratio of 1:10 (w/v), and the pH was adjusted to an alkaline range (8–12) with sodium hydroxide (NaOH) solution to facilitate protein solubilization. The mixture was stirred at room temperature for varying durations (30, 60, or 90 min) to enhance protein dissolution. The resulting solution was filtered and centrifuged at 3300×g to remove insoluble fibers and starch, with the supernatant containing solubilized proteins. Protein precipitation was induced by adjusting the pH to 4.5–4.7 using HCl, followed by incubation at 4 °C for 18 h to promote aggregation. The protein precipitates were collected and washed 2 times with Milli-Q water (1:10, w/v), each wash being followed by centrifugation at 3300×g and removal of the supernatant, in order to remove residual salts and non-protein solutes. The washed precipitate was then neutralized to pH 7.0 with 1 M NaOH and freeze-dried for further use.

2.2. Microwave-assisted extraction optimization

Microwave-assisted extraction of lupin protein isolates (LPI) was optimized using a screening design in which microwave power was set at three levels and irradiation time at three levels within each power setting, giving nine runs in total. The design contained no center or axial replicates and was therefore used to locate the most productive operating window rather than to fit a predictive response-surface model. Lupin flour was dispersed in distilled water at a ratio of 1:10 (w/v), and the pH was adjusted to 9. The slurry was then subjected to microwave heating in a Labotron 12T microwave system (SAIREM, Lyon, France) at nine combinations of microwave power (200, 400 and 800 W) and irradiation time (10–90 min, the levels applied at each power setting being those listed in Table S2). The three power levels were chosen to span the low, intermediate and high output range of the unit in approximately doubling increments; an intermediate 600 W setting was not included because it lies within the 400–800 W interval already covered and would not have extended the range examined. The condition selected for preparing the microwave-extracted LPI used in LPIGHM was 400 W for 20 min at pH 9. This condition gave the highest mean yield in the design (293.72 mg/g). The margin over 400 W for 10 min (270.54 mg/g) is modest, and the shorter run would be preferable on energy grounds; the higher-yielding condition was retained here so that the conjugation study proceeded from the most efficient extraction identified, but 400 W for 10 min represents a viable lower-energy alternative if the process were scaled. Microwave treatment was applied only during LPI extraction; the subsequent LPI-GOS conjugation step was performed by conventional water-bath wet heating at 90 °C for 15 min.

After microwave treatment, samples were immediately cooled to room temperature to prevent further thermal degradation. After the cooling step, all subsequent processing followed the conventional alkaline extraction method described in Section 2.1. In brief, the suspensions were filtered and centrifuged at 3300×g to remove insoluble material, and the supernatant containing solubilized proteins was collected. Proteins were precipitated by adjusting the pH to 4.5–4.7 with HCl and incubating at 4 °C for 18 h to promote aggregation. The resulting protein precipitates were washed with deionized or Milli-Q water (1:10, w/v), neutralized to pH 7.0, and finally freeze-dried for subsequent analyses.

2.3. Preparation of LPI-GOS mixtures and conjugates

Preparation of LPI-GOS mixtures and conjugates was conducted according to the methods in (Li, Xiong, et al., 2024), with slight modifications. GOS, a well-characterized oligosaccharide, was obtained from FrieslandCampina (Biotis™ GOS-P, Amersfoort, The Netherlands). Briefly, 2 g of lupin protein isolate (LPI) and 2 g of GOS were dissolved in 25 mL of 0.2 M phosphate buffer at pH 9.0 and stirred for 2 h at room temperature, with pH adjustments as necessary. The 1:1 protein:GOS mass ratio follows the equal-mass systems used for gelatin–xylooligosaccharide and casein–oligosaccharide wet-heating conjugates (Li, Chen, et al., 2024; Li, Xiong, et al., 2024), which give a workable degree of grafting without the excessive browning that accompanies higher sugar loads. The mixture was then transferred to a refrigerated cabinet and stirred gently overnight at 4 °C under continuous magnetic stirring, to ensure complete hydration of the biopolymers. For conjugate formation, the mixture was heated in a water bath at 90 °C for 15 min to facilitate protein-polysaccharide conjugation, while the non-heated mixture was left without this step. The heated samples were immediately cooled in an ice-water bath to stop the reaction. Subsequently, all samples were centrifuged at 10,000×g for 10 min, and the supernatants, containing dissolved biopolymers, were collected for further analysis or freeze-drying, whereas the pellets, containing insoluble or denatured material, were discarded. Finally, the supernatants were freeze-dried to obtain the final LPI-GOS conjugate or mixture powders. The four systems were: LPI, conventionally extracted lupin protein isolate; LPIG, non-heated mixture of conventionally extracted LPI and GOS; LPIGH, wet-heated LPI-GOS conjugate prepared from conventionally extracted LPI; and LPIGHM, wet-heated LPI-GOS conjugate prepared from microwave-extracted LPI. Microwave treatment was applied only during LPI extraction, whereas all Maillard conjugation was carried out by water-bath heating at 90 °C for 15 min.

2.4. Characterization

2.4.1. Structural characterization

SDS-PAGE was used to evaluate changes in apparent molecular-weight distribution of LPI and its conjugates following Maillard glycation Ma et al. (2022). FT-IR spectroscopy was conducted as described by Ayyash et al. (2020). Samples were placed on a diamond/ZnSe ATR crystal plate and spectra were recorded over 4000–400 cm⁻¹ at room temperature (23 °C), averaging 16 scans per sample at a resolution of 4 cm⁻¹ against a background collected under the same conditions. Surface morphology was examined by SEM (JEOL JSM-6010/LA, Tokyo, Japan) following Ayyash et al. (2020). Samples were sputter-coated with a thin gold layer and imaged at an accelerating voltage of 20 kV. Thermal characteristics were assessed by DSC (DSC 25, TA Instruments, USA), using 5.0 mg samples, according to Al-Hamayda et al. (2023). Samples were sealed in aluminum pans and heated from 50 to 250 °C at 10 °C/min under a constant nitrogen supply, with an empty sealed pan as reference. TGA was performed on a TGA-2 Star system (Mettler Toledo, Switzerland) to determine thermal degradation and weight loss, based on Tan et al. (2020). Samples (approximately 5 mg) were heated from 30 to 800 °C at 10 °C/min under a constant nitrogen supply. Zeta potential and particle size were measured using a NanoPlus Analyzer (Particulate Systems, GA, USA) following the method of Ayyash et al. (2020). RF-6000 spectrofluorophotometer (Shimadzu, Kyoto, Japan) was employed to measure the intrinsic fluorescence according to (Amiratashani et al., 2024), with excitation at 290 nm, emission recorded from 300 to 400 nm and an emission slit width of 5 nm, and to determine the surface hydrophobicity (H_O) using 8-anilinonaphthalene-1-sulfonic acid (ANS) according to Xu and Zhao (2019). The changes in the secondary structure were characterized using an FTIR spectrum according to Nadeeshani et al. (2026). The collected FTIR spectra were normalized, and the second-derivative spectrum was calculated, followed by curve fitting of the amide I region (1700–1600 cm⁻¹) using the Savitzky–Golay model with nine smoothing points in Origin 2021 software (OriginLab Corporation, MA, USA).

2.4.2. Functional characterization

The formation of Amadori and Melanoidin compounds in LPI and its conjugates were assessed by measuring UV absorbance at 304 nm and 420 nm, respectively, as markers of Maillard reaction progression, following Amiratashani et al. (2024). Browning degree was determined by Jiang et al. (2023) measuring absorbance at 294 nm and 420 nm. The free amino groups (FAG) and apparent degree of graft (DG) were determined using the o-phthaldialdehyde (OPA) method, following Zhao et al. (2021), with modifications. Native lupin protein isolate (LP-N) was used as the reference sample and was considered to contain 100% free amino groups. All samples were adjusted to the same protein concentration before analysis. DG and FAG were calculated as follows:

$$\mathrm{DG}\left(\%\right)=\left(\frac{{\mathrm{A}}_{\mathrm{r}}-{\mathrm{A}}_{\mathrm{s}}}{{\mathrm{A}}_{\mathrm{r}}}\right)\times 100$$ (1)

$$\mathrm{FAG}\left(\%\right)=\left(\frac{{\mathrm{A}}_{\mathrm{s}}}{{\mathrm{A}}_{\mathrm{r}}}\right)\times 100$$ (2)

where Ar is the absorbance of the reference sample, LP-N, and As is the absorbance of each treated sample. DG is reported as an apparent degree of graft based on the reduction in OPA-reactive free amino groups. Because the OPA assay measures the reduction in accessible free amino groups rather than the direct amount of covalently bound GOS, DG should be interpreted as an apparent degree of graft and not as direct grafting efficiency or direct GOS conversion.

Protein solubility at pH 7.0 was measured for LPI and its conjugates using the method of Lavaei et al. (2022), protein content in the supernatant was determined using the Biuret method. Solubility was calculated as:

$$\text{Solubility}\left(\%\right)=\left(\frac{{\mathrm{P}}_{\mathrm{s}}}{{\mathrm{P}}_{\mathrm{i}}}\right)\times 100$$ (3)

where Ps and Pi represent the protein concentrations in the supernatant and the initial solution, respectively.

Foaming properties were assessed by Amiratashani et al. (2024), Foaming capacity (FC) and foam stability (FS) were calculated as:

$$\mathit{FC}\left(\%\right)=\left(\frac{V_1-V_0}{V_0}\right)\times 100$$ (4)

$$\mathit{FS}\left(\%\right)=\left(\frac{V_2-V_0}{V_1-V_0}\right)\times 100$$ (5)

where V₁ is the total volume (mL) of solution and foam after whipping, and V₀ is the solution volume (mL) before whipping, where V₂ is the total volume of solution and foam after 30 min.

Water holding capacity (WHC) and oil holding capacity (OHC) were evaluated following Amiratashani et al. (2024), and were calculations as:

$$\mathit{WHC}\left(g/g\right)=\left(\frac{W_2-W_1-V_0}{W_0}\right)\times 100$$ (6)

$$\mathit{OHC}\left(g/g\right)=\left(\frac{O_2-O_1-O_0}{O_0}\right)\times 100$$ (7)

where W and O refer to water and oil sample measurements, respectively.

Emulsifying properties were measured as per Lavaei et al. (2022). Emulsifying activity index (EAI) and emulsion stability index (ESI) were calculated as:

$$\mathit{EAI}\left(m^{²}/g\right)=\left(\frac{2\times 2.30\times A0\times \mathit{DF}}{C\times L\times \left(1-\theta \right)\times 10000}\right)$$ (8)

$$\mathit{ESI}\left(\mathit{mins}\right)=\left(\frac{A_0\times 30}{A_0\times A_{30}}\right)$$ (9)

where A₀ and A₃₀ are the absorbance values of the diluted emulsion at 0 and 30 min, respectively; C is the initial protein concentration (mg/mL); DF is the dilution factor (50); θ is the oil volume fraction (0.25); and L is the optical path length (0.01 m).

2.5. Biological activities

The biological activities of LPI and its conjugates were comprehensively investigated to evaluate their functional and health-promoting potential. Their antidiabetic properties were determined through the inhibition of α-amylase and α-glucosidase enzymes following the protocol of Ayyash et al. (2018), with absorbance measurements taken at 540 nm and 400 nm, respectively. Assays were performed on freeze-dried LPI and conjugate powders redispersed over 12.5–100 mg/mL and applied without any prior enzymatic hydrolysis, so the values reported represent inhibition by the intact protein and conjugate rather than by released peptides. The percentage of inhibition was calculated using the formula:

$$\mathit{Inhibition}\left(\%\right)=\frac{\left(\mathit{Blank}\mathit{Abs}-\mathit{Sample}\mathit{Abs}\right)}{\mathit{Bl}\mathrm{a}\mathit{nk}\mathit{Abs}}\times 100$$ (10)

The angiotensin-converting enzyme (ACE) inhibitory activity was also assessed using the method of Ayyash et al. (2018), with absorbance measured at 228 nm and inhibition percentages calculated using the same formula.

$$\mathit{Inhibitio}\mathrm{n}\left(\%\right)=\frac{\left(\mathit{Blank}\mathit{Abs}-\mathit{Sample}\mathit{Abs}\right)}{\mathit{Blank}\mathit{Abs}}\times 100$$ (11)

The antiproliferative potential of LPI and its conjugates was tested on Caco-2 and MCF-7 cancer cell lines using the method described by Ayyash et al. (2018). The antiproliferative activity (%) was calculated as:

$$\mathit{Antiproliferative}\left(\%\right)=\left(\frac{R_{\mathit{sample}}-R_0}{R_{\mathit{control}}-R_0}\right)\times 100$$ (12)

where R_sample is the absorbance ratio at 570/605 nm in the presence of the test sample, R_control is the absorbance ratio at 570/605 nm in the absence of the test sample (vehicle (positive) control), and R_o represents the background (non-cell (negative) control) absorbance ratio at 570/605 nm.

Additionally, the antioxidant potential of LPI and its conjugates was thoroughly assessed using various antioxidant assays, including DPPH and ABTS radical scavenging activities, total antioxidant capacity (TAC), superoxide anion scavenging activity (SAS), and ferric reducing antioxidant power (FRAP), all conducted according to Subhash et al. (2024). Ferric reducing antioxidant power and total antioxidant capacity were quantified against an ascorbic acid standard curve, with absorbance read at 593 nm and 695 nm, respectively, and the results are expressed as μg ascorbic acid equivalents per mL.

2.6. Gut microbiota modulation by in vitro fecal fermentation

The gut microbiota modulation potential of LPI and its conjugates was assessed through in vitro fecal fermentation using fresh fecal samples from six healthy adult donors, three men and three women, aged 24–40 years, none of whom had used antibiotics in the six months preceding collection. Donors were not placed on a standardized diet before collection and dietary fiber intake was not controlled. Equal masses of fresh feces from each donor were combined into a single pooled slurry before inoculation, so that donor-to-donor variation was averaged rather than examined as an experimental factor. Each treatment was subsequently fermented in triplicate from that pooled inoculum, and the triplicates therefore represent analytical rather than inter-individual replication. Differences between individual microbiota are consequently not resolved by this design. The protocol for fecal sample collection was approved by the UAEU-Human Research Ethics Committee (UAEU-HREC) of United Arab Emirates University, UAE (No. ERH_2024_5441). Fecal slurry was prepared and homogenized in a basal fermentation medium following a modified protocol by Han et al. (2022). Throughout the fermentation process, changes in pH, total sugar, gas production, and reducing sugar content were monitored at intervals of 0, 6, 12, and 24 h following the method of Han et al. (2022). SCFAs production, including acetic, propionic, and butyric acids, was quantified during 24 h of fermentation using HPLC as described by Dobrowolska-Iwanek et al. (2020).

Microbial composition during fecal fermentation was analyzed after 24 h. Genomic DNA was extracted with the Tiangen Genomic DNA Kit, and the V3–V4 regions of the 16S rRNA gene were amplified and sequenced by BGI (Hong Kong). Library preparation and quality checks were conducted using Agencourt AMPure XP beads and an Agilent 2100 Bioanalyzer. Raw sequences were filtered with iTools Fqtools fqcheck (v0.25), and paired-end reads were merged using FLASH (v1.2.11). Operational taxonomic units (OTUs) were clustered at 97% similarity using UPARSE with chimera filtering by UCHIME (v4.2.40). OTU sequences were mapped using USEARCH (v7.0.1090) and taxonomically assigned with the RDP classifier (v2.2) at a 60% confidence threshold. Alpha and beta diversity indices were calculated using mothur (v1.31.2), QIIME (v1.80), and R (v3.1.1). LEfSe was used for differential abundance analysis, while microbial functional predictions were made using PICRUSt2 (v2.3.0-b). Correlation and predictive modeling were conducted in R (v3.4.1) and visualized with Cytoscape.

2.7. Statistical analyses

All experiments were performed in triplicate (n = 3) using purified LPI and its conjugates across a concentration range of 12.5–100 mg/mL. Antioxidant data were analyzed by two-way ANOVA with treatment (LPI, LPIG, LPIGH, LPIGHM) and concentration (12.5, 25, 50 and 100 mg/mL) as crossed fixed factors, including their interaction. Fermentation data were analyzed by two-way ANOVA with treatment (NC, LPIG, LPIGH, LPIGHM) and sampling time (0, 6, 12 and 24 h) as crossed fixed factors, again including the interaction term. Replicates were treated as independent experimental units and no random effects were fitted. Measurements involving a single factor, including particle size, zeta potential, surface hydrophobicity, degree of glycation, solubility and the interfacial properties, were analyzed by one-way ANOVA. In every case means were separated using Tukey's honestly significant difference test at P < 0.05. All analyses were carried out in RStudio version 2025.09.2 + 418, using the aov function of the base stats package to fit the models and TukeyHSD for the multiple comparisons. Normality and homogeneity of variance were assessed on the model residuals before parametric testing, and non-parametric tests were applied where those assumptions were not met. For the sequencing data, differences in alpha-diversity indices among treatments were evaluated by the Kruskal–Wallis test; beta-diversity was assessed on Bray–Curtis distances and group separation tested by ANOSIM; discriminative taxa were identified by LEfSe using thresholds of P < 0.05 and an absolute LDA score above 2.0; and associations among dominant taxa were evaluated by Spearman correlation. To examine the structure–function relationships directly, Pearson correlation coefficients were calculated between the structural parameters (particle size, zeta potential, surface hydrophobicity, β-sheet content and apparent degree of grafting) and the functional and antioxidant responses, using the four treatment means as input (Table S3). Because each coefficient is based on four points, these correlations were treated as exploratory trends rather than as independent evidence of causality.

3. Results and discussion

3.1. Optimization of conventional and microwave-assisted extraction of LPI

Lupin protein extraction was optimized under both conventional alkaline and microwave-assisted conditions using factorial screening designs rather than response-surface designs. Neither design included center or axial replicates, so the runs identify the most productive operating window but do not support a fitted predictive model; no response-surface polynomial was therefore derived, and the yield surfaces in Figs. S1 and S2 are interpolations of the measured points rather than model predictions. For the alkaline extraction, solvent pH, stirring time and temperature were varied across 12 runs (Table S1). The highest yield, 39.73 mg/g, was obtained at pH 9, 90 min and 20 °C, followed by pH 9, 60 min and 30 °C (37.74 mg/g) and pH 11, 60 min and 40 °C (37.59 mg/g). Yield was governed mainly by pH: the pH 9 runs outperformed pH 8, 10 and 11 at comparable stirring times, which is consistent with efficient solubilization of the major lupin globulins above their isoelectric range while avoiding the more aggressive alkaline conditions that begin to modify the protein at higher pH. Microwave-assisted extraction was screened across nine combinations of power and time (Table S2). Here power was the dominant variable. All 200 W runs remained close to 160 mg/g irrespective of time, indicating that the energy input was insufficient to disrupt the protein bodies, whereas 400 W for 20 min gave the highest value in the design (293.72 mg/g). Yields obtained under microwave heating were substantially higher than those from the alkaline runs, which reflects the rapid volumetric heating of the slurry and the associated release of protein from the cellular matrix rather than any change in the extraction chemistry itself.

3.2. Structural characterization

FTIR analysis showed that glycation between LPI and GOS caused notable structural changes (Fig. 1A). Increased O—H and N—H stretching (∼3300 cm⁻¹) indicated stronger hydrogen bonding and covalent interactions, while reduced Amide I (∼1650 cm⁻¹) and Amide II (∼1530 cm⁻¹) bands reflected a transition from ordered (α-helix, β-sheet) to disordered (random coil) structures. Enhanced C—O and C—N stretching (1200–1000 cm⁻¹) confirmed Maillard-type carbohydrate incorporation. Shifts in the fingerprint region (<1000 cm⁻¹) supported backbone rearrangements. Compared to LPIG and LPIGH treatments, LPIGHM induced the most pronounced modifications, suggesting that microwave-assisted extraction altered LPI conformation in a way that favored subsequent GOS glycation during wet heating. These results align with previous plant-protein glycation studies showing that reduced ordered structure and stronger O—H/N—H stretching accompany Maillard-type structural rearrangement (Namli et al., 2021). The chemical basis of these shifts is the condensation of a GOS carbonyl group with the ε-amino group of lysine and with the N-terminal α-amino group, which converts a charged primary amine into a secondary amine carrying a polyhydroxy chain. Removing those charged amines eliminates intermolecular salt bridges that help stabilize β-sheet packing, while the attached galactose units introduce a dense set of hydroxyls that hydrogen-bond preferentially to water rather than to neighboring peptide chains. The two effects act in the same direction: the amide I band weakens and broadens as ordered backbone hydrogen bonding is progressively replaced by hydration, and the 1200–1000 cm⁻¹ C—O envelope grows in proportion to the carbohydrate now covalently carried by the protein. The larger shifts in LPIGHM therefore indicate that more of these linkages were formed, not that a chemically different reaction took place.

Fig. 1.

FTIR spectra (A), DSC thermograms (B), TGA profiles (C), zeta potential (D), and secondary structure distribution (E) of lupin protein isolate (LPI) and LPI-GOS systems. LPI: lupin protein isolate obtained by conventional alkaline extraction; LPIG: non-heated mixture of conventionally extracted LPI and galacto-oligosaccharides (GOS); LPIGH: wet-heated LPI-GOS conjugate prepared from conventionally extracted LPI; LPIGHM: wet-heated LPI-GOS conjugate prepared from microwave-extracted LPI.

Differential scanning calorimetry revealed distinct thermal behaviors of LPI and its conjugates (Fig. 1B) between 50 and 250 °C. Native LPI showed minimal thermal transitions, indicating limited phase changes. LPIG exhibited a main endothermic peak at ∼115 °C and additional transitions between 140 and 160 °C. LPIGH displayed peaks at ∼110 °C and ∼140 °C, with further events at ∼180 °C and ∼200 °C. LPIGHM showed the most complex profile, with overlapping peaks from 100 to 150 °C and transitions at ∼160 °C, ∼180 °C, and ∼220 °C. These endothermic and exothermic events correspond to melting, dehydration, crystallization, and crosslinking. The increasing thermal complexity from LPIG to LPIGH to LPIGHM indicates successful functionalization of LPI, with each treatment inducing distinct changes in structural and molecular interactions. These results are consistent with findings on β-glucan–peanut/soy protein isolate conjugates applied in low-fat sausages (Zhang et al., 2022), and on soy protein–maltose glycates prepared in ionic liquids via the Maillard reaction (Xu & Zhao, 2019).

Thermogravimetric analysis revealed distinct degradation profiles for LPI and its conjugates (Fig. 1C) from 30 to 800 °C. Native LPI degraded in three stages: initial moisture loss (∼13%) below 200 °C, followed by major mass losses of 24% and 23% between 200 and 600 °C. LPIG showed lower thermal stability, with higher final mass loss (∼33%) and sharper DTG peaks around 200–300 °C. Heat-induced conjugation (LPIGH) improved resistance, reducing overall mass loss (∼57%) and broadening DTG peaks to higher temperatures, indicating stronger molecular interactions. The wet-heated conjugate prepared from microwave-extracted LPI (LPIGHM) exhibited similar behavior, with total mass loss (∼58%) and multiple broadened degradation steps, suggesting a more complex decomposition pathway. The trend from LPIG to LPIGHM demonstrates that wet-heating Maillard conjugation, especially when applied to microwave-extracted LPI, enhances LPI's structural integrity and thermal stability. These results are consistent with observations in soy protein–maltose glycates prepared in ionic liquids via the Maillard reaction (Xu & Zhao, 2019), and in soybean protein–chitosan oligosaccharide Maillard reaction products obtained through wet heating (Teng et al., 2022).

Wet-heating of conventionally extracted LPI induced the most pronounced charge alteration, whereas the conjugate prepared from microwave-extracted LPI showed a milder charge shift. As shown in Fig. 1D, zeta potential of LPI exhibited a low surface charge of −4.37 mV, while Maillard conjugation with GOS significantly enhanced the negative charge: LPIG showed the highest shift (−34.53 mV), followed by LPIGH at −29.52 mV and LPIGHM at −9.34 mV. This increase in charge may enhance the copolymer's ability to stabilize the interface (Caballero & Davidov-Pardo, 2021). The letters shown in Fig. 1D denote significance groupings among the four systems and are not ordered by magnitude; all four zeta potential values differed from one another (P < 0.05). The smaller negative charge of LPIGHM relative to LPIG and LPIGH is consistent with its higher extent of glycation rather than at odds with it. The covalently attached carbohydrate layer displaces the hydrodynamic shear plane outward and screens the underlying charged groups, so the measured zeta potential falls even though more GOS is bound. Colloidal stabilization in this system consequently shifts from a predominantly electrostatic to a steric and hydration basis, which is the interpretation developed further in Section 3.3.

Secondary-structure analysis indicated that native LPI adopted a predominantly β-sheet-rich conformation (Fig. 1E) (56.7%), followed by β-turns (19.1%) and α-helices (18.8%), with a small random coil component (5.4%). Following the addition of GOS, the overall β-sheet content was reduced in LPIG to 51.9%, accompanied by increased β-turn and random coil content. Wet-heating conjugation and prior microwave-assisted extraction resulted in additional, albeit small, conformational changes. The LPIGH and LPIGHM samples showed the lowest β-sheet content, at 50.3% and 50.0%, respectively, accompanied by increased β-turn content, at 21.7% and 22.4%, respectively, and a slightly increased random coil component, at 8.0% and 8.0%, respectively, whereas α-helix content remained almost constant at ∼19–20%. The β-sheet-dominant native conformation is supported by recent FTIR studies on white lupin protein isolates, whereby β-sheet was reported as the major component of the secondary structure, at 56–66%, and α-helix content was reported at 18–30%, respectively (Nadeeshani et al., 2026). Francisco et al. (2023) reported that, for lupin protein-grape seed extract microconjugates, heat treatment resulted in secondary structure modifications compared to their non-heated counterparts, thus suggesting that thermal processing plays a role in conformational rearrangement of lupin protein-based conjugates.

Particle size measurements showed that native LPI was strongly aggregated, with a mean hydrodynamic diameter of 1775.1 nm (Fig. 2A). Following conjugation treatments, particle size reduction was observed: LPIG decreased to 152.2 nm (11.7-fold reduction), LPIGH to 99.2 nm, and LPIGHM to 89.7 nm (19.8-fold reduction). This reduction is consistent with cold plasma-induced conjugates of pea protein and marine oligosaccharides, where particle size decreased from 3.80 μm to 1.31 μm due to structural modifications enhancing hydration and colloidal stability (Guiyun et al., 2025). The diminished particle size indicates enhanced hydration, electrostatic repulsion, and protein disaggregation, yielding smaller, more stable complexes with improved functionality. Notably, even the non-heated lupin protein isolate–galactooligosaccharide system (LPIG) exhibited a reduced particle size compared with native LPI, likely due to electrostatic interactions and steric stabilization provided by galactooligosaccharides, which can limit protein–protein aggregation prior to thermal conjugation, as similarly reported for protein–polysaccharide physical mixtures in previous studies (Kan et al., 2024). These fold changes should nonetheless be read with caution. Native LPI is highly polydisperse, and dynamic light scattering weights the intensity signal toward the largest particles present, so the 1775.1 nm value reflects a minority population of large aggregates rather than a typical particle. The apparent 19.8-fold reduction therefore quantifies the loss of that aggregate population rather than a uniform contraction of every particle in the dispersion. This also explains why the SEM images of the dried powders (Fig. 3) show larger apparent structures than the hydrodynamic diameters reported here, a point developed below.

Fig. 2.

Particle size (A), fluorescence spectra (B), SDS-PAGE profiles (C), and surface hydrophobicity (D) of LPI and LPI-GOS systems. LPI: conventionally extracted lupin protein isolate; LPIG: non-heated mixture of conventionally extracted LPI and GOS; LPIGH: wet-heated LPI-GOS conjugate prepared from conventionally extracted LPI; LPIGHM: wet-heated LPI-GOS conjugate prepared from microwave-extracted LPI.

Fig. 3.

SEM micrographs of LPI and LPI-GOS systems: LPI, conventionally extracted lupin protein isolate (A); LPIG, non-heated mixture of conventionally extracted LPI and GOS (B); LPIGH, wet-heated LPI-GOS conjugate prepared from conventionally extracted LPI (C); and LPIGHM, wet-heated LPI-GOS conjugate prepared from microwave-extracted LPI (D).

Fluorescence spectroscopy revealed pronounced tertiary structural modifications (Fig. 2B) in LPI upon glycosylation with GOS under different processing conditions. Upon excitation at 290 nm, native LPI showed maximum emission at 325 nm. Glycosylated LPI-GOS conjugates exhibited reduced fluorescence intensity and blue-shifted λ_max to 315 nm LPIG and 310 nm (LPIGH, LPIGHM), with LPIGHM displaying the most pronounced quenching. These spectral changes indicate increased environmental polarity around tryptophan residues, reduced solvent accessibility, and enhanced protein rigidity. The intensified quenching in LPIGHM reflects tertiary structural changes introduced during microwave-assisted extraction before wet-heating conjugation, whereas wet-heating of conventionally extracted LPI induced milder alterations, underscoring the role of extraction history and conjugation in dictating conformational outcomes. This correlation between glycosylation-induced tertiary structural changes and tryptophan fluorescence quenching aligns with findings in plant protein systems, where reduced aromatic residue exposure indicated increased molecular rigidity (Dong et al., 2024); additionally, extraction-induced structural modifications are consistent with improved glycation susceptibility and functionality reported for structurally modified protein–polysaccharide systems (Zhiyu et al., 2024).

SDS–PAGE analysis was conducted to examine the molecular weight distribution of the proteins and their conjugates, as shown in Fig. 2C. Lane 1 corresponds to the molecular weight marker (15–250 kDa), while lanes 2–5 and 6–9 represent the samples, with lanes 2 and 6 corresponding to LPI, lanes 3 and 7 to LPIG, lanes 4 and 8 to LPIGH, and lanes 5 and 9 to LPIGHM. The un-conjugated LPI samples (lanes 2 and 6) displayed several protein bands mainly distributed between approximately 15 and 40 kDa, based on comparison with the marker in lane 1. The LPIG samples (lanes 3 and 7) exhibited a similar electrophoretic pattern to LPI, with protein bands still predominantly observed within the same molecular weight region, indicating that the protein profile remained largely unchanged. In contrast, the conjugated samples LPIGH (lanes 4 and 8) and LPIGHM (lanes 5 and 9) showed the appearance of a prominent band located between the 50 and 75 kDa marker positions, corresponding to an approximate molecular weight of about 55–65 kDa. This band was not observed in the un-conjugated samples and appeared markedly more intense than the lower molecular weight bands present in LPI and LPIG. The emergence of this higher molecular weight band, together with the reduced intensity of bands below approximately 40 kDa, suggests the formation of protein conjugates or aggregates with increased molecular size compared with the original un-conjugated proteins. These observations indicate that the conjugation process promoted the association of smaller protein fractions into larger molecular structures, resulting in the formation of higher molecular weight complexes detectable in the SDS–PAGE profiles. The same pattern, a new band at roughly twice the mass of the parent subunits accompanied by weakening of the original bands, was reported for grass pea protein–xanthan gum and whey protein–carboxymethyl cellulose conjugates prepared by wet heating (Amiratashani et al., 2024; Jiang et al., 2023). In both cases the shift was attributed to covalent carbohydrate attachment rather than to non-covalent association, since the increase in apparent mass exceeded what protein–protein aggregation alone would produce.

The highest H₀ was obtained for native LPI (Fig. 2D), with an H₀ of 443.30 arbitrary units, expressed as the slope of relative fluorescence intensity against protein concentration. However, this was reduced (P < 0.05) in LPIG, with an H₀ of 347.96. A more dramatic reduction was noted for LPIGH and LPIGHM, with H₀ values of 71.09 and 62.49, respectively. The dramatic reduction in the surface hydrophobicity of LPI indicates that the conjugation process significantly altered the conformation of the protein. This could have resulted from the covalent attachment of the polysaccharide moiety, which introduces hydrophilic groups. In addition, the conjugation process could have resulted in the protein adopting a conformation that shields the hydrophobic groups from the surface of the protein. These hydrophobic groups could have been buried within the protein. Taken together, the fall in H₀ reflects a surface now dominated by hydroxyl-rich GOS chains rather than by exposed apolar side chains. The same change accounts for the higher solubility and water-holding capacity reported in Section 3.3, and for the shift in interfacial behavior from charge-driven to hydration-driven stabilization noted above.

SEM images recorded at 1000× magnification revealed distinct structural changes induced by Maillard conjugation (Fig. 3). Native LPI displayed a rough, irregular surface with large aggregates, whereas the conjugates showed smaller, more dispersed particles with reduced aggregation. Among them, LPIGHM exhibited the most distinct morphology, consistent with steric stabilization and controlled glycation. These changes support Maillard conjugation's role in improving plant protein functionality. A study by Huang et al. (2020) similarly reported structural changes in protein–polysaccharide composites, showing reduced aggregation and improved molecular dispersion in plant-based matrices via SEM analysis. It should be noted that the particle size values obtained from dynamic light scattering represent the hydrodynamic diameter of protein particles in aqueous dispersion, where hydration, electrostatic repulsion, and steric effects contribute to reduced apparent size. In contrast, SEM images reflect the morphology of dried samples under high vacuum, where dehydration and sample preparation can promote particle collapse and aggregation, resulting in larger apparent structures. Therefore, SEM was used to assess surface morphology and aggregation trends rather than absolute particle size. Importantly, both techniques consistently demonstrate reduced aggregation and improved dispersion of LPI after GOS conjugation, particularly in the wet-heated conjugate prepared from microwave-extracted LPI (LPIGHM).

3.3. Functional characterization

Browning was followed at 294 nm as an index of early Maillard intermediates and at 420 nm as an index of melanoidin formation (Fig. 4A). Absorbance values are reported to two decimal places, which corresponds to the working precision of the spectrophotometer. The non-heated mixture LPIG gave the lowest readings (2.07 at 294 nm and 1.31 at 420 nm), consistent with minimal reaction in the absence of heat. Wet heating raised both indices in LPIGH (2.51 and 1.70), and the conjugate prepared from microwave-extracted LPI gave the highest values (2.57 and 1.75). Amadori compound formation followed the same order (Fig. 4B): 2.50 and 0.79 for LPIG, 2.99 and 0.82 for LPIGH, and 3.08 and 0.84 for LPIGHM. The ordering follows directly from the reaction chemistry. Absorbance at 294 nm tracks the Amadori rearrangement product formed after the initial Schiff base, and the amount produced within a fixed heating period depends on how many amino groups are accessible to the GOS carbonyl when heating begins. Microwave-assisted extraction leaves the protein partially unfolded, so a larger fraction of lysine ε-amino groups is exposed at the start of the water-bath step and the reaction advances further within the same 15 min (Cipriano et al., 2018; Ke et al., 2023; Zhanmei et al., 2021). The parallel but smaller rise at 420 nm indicates that this additional reaction remained largely at the intermediate stage rather than progressing extensively to melanoidin formation, which is the intended outcome of a controlled wet-heating protocol.

Fig. 4.

Browning intensity (A), Amadori compound formation (B), and degree of glycation (DG) and free amino group content (FAG) (C) of LPI-GOS systems. LPIG: non-heated mixture of conventionally extracted LPI and GOS; LPIGH: wet-heated LPI-GOS conjugate prepared from conventionally extracted LPI; LPIGHM: wet-heated LPI-GOS conjugate prepared from microwave-extracted LPI.

The degree of glycation (DG) and the free amino group content (FAG) moved in opposite directions across the LPI-GOS systems (Fig. 4C). As set out in Section 2.4.2, DG here is an apparent value derived from the loss of OPA-reactive free amino groups; it reports how many amino groups have become unavailable to the reagent and is not a direct measurement of the mass of GOS covalently bound. The values below should be read on that basis. DG increased progressively from LPIG to LPIGH and LPIGHM, whereas FAG decreased in the same order, confirming the consumption of free amino groups during Maillard conjugation. LPIG showed the lowest DG and the highest FAG, reflecting limited reaction in the non-heated mixture. Wet heating increased DG and reduced FAG in LPIGH, while LPIGHM showed the highest DG and the lowest FAG, indicating that microwave-assisted extraction of LPI enhanced its susceptibility to subsequent wet-heating glycation with GOS. This behavior is consistent with previous reports showing that structural modification of proteins can expose reactive amino groups and facilitate Maillard-type protein-polysaccharide conjugation (Ertugrul et al., 2021; Namli et al., 2021). These results support the FTIR, fluorescence and SDS-PAGE observations, which collectively indicate greater protein-GOS interaction and formation of higher-molecular-weight conjugates in LPIGHM. Because the OPA method cannot distinguish an amino group consumed by GOS attachment from one rendered inaccessible by conformational change or by protein–protein crosslinking, the DG values are best treated as an upper bound on covalent grafting. The supporting evidence for genuine conjugation comes from the combination of measurements rather than from DG alone: the new 55–65 kDa band in SDS–PAGE, the growth of the C—O stretching envelope in FTIR, and the rise in browning and Amadori absorbance together indicate covalent attachment, since none of these would follow from conformational change by itself.

Protein solubility at pH 7.0 rose modestly across the series (Fig. 5A). Unmodified LPI was the least soluble at 31.46%, followed by LPIG at 31.86%, with LPIGH and LPIGHM the most soluble at 32.96% each; the last two did not differ from one another. The increase is small because solubility was measured at pH 7.0, well away from the isoelectric region where glycation makes its largest difference to plant proteins. This aligns with findings that SPI-GOS conjugates under novel processing (e.g., homogenization) enhance solubility through polar group exposure, demonstrating the role of glycation in mitigating plant protein hydrophobicity (Xie et al., 2022). Further, microwave glycation of soy protein similarly promotes structural modifications that favor solubility (Namli et al., 2021).

Fig. 5.

Solubility (A), foaming properties (B), water- and oil-holding capacities (C), and emulsifying properties (D) of LPI and LPI-GOS systems. LPI: conventionally extracted lupin protein isolate; LPIG: non-heated mixture of conventionally extracted LPI and GOS; LPIGH: wet-heated LPI-GOS conjugate prepared from conventionally extracted LPI; LPIGHM: wet-heated LPI-GOS conjugate prepared from microwave-extracted LPI.

Conjugation improved foaming far more than it improved solubility (Fig. 5B). Native LPI reached only 10% foaming capacity and 4% foam stability. LPIG reached 34% capacity and 11% stability; LPIGH improved to 46% and 18%; LPIGHM had the highest at 64% capacity and 36% stability, corresponding to 6.4- and 9-fold increases, respectively. Microwave-assisted extraction enhances protein–polysaccharide interactions and structural flexibility, boosting foam formation and stability, a mechanism corroborated by studies demonstrating microwave-induced structural modifications in plant proteins that optimize foam stabilization (Sayeh et al., 2023). Comparable effects of microwave heating on the expansion and stability of foamed food systems have been reported for foamed mango pulp (Duangkhamchan et al., 2025).

Water- and oil-holding capacities increased in the same direction (Fig. 5C): native LPI exhibited values of 1.2 g/g WHC and 12.4 g/g OHC, while LPIGHM reached 2.9 g/g WHC and 14.6 g/g OHC, corresponding to 2.4-fold and 1.2-fold improvements, respectively. Microwave-assisted extraction enhanced hydrophilic and lipophilic binding sites through optimized protein–polysaccharide interactions and conformational changes, a mechanistic alignment observed in ultrasound-microwave enzymatic treatments of plant protein–carbohydrate complexes where structural modifications significantly improved emulsion functionality (Sun et al., 2022).

Emulsifying activity index (EAI) and emulsion stability index (ESI) improved in the same order (Fig. 5D), starting from native LPI (7.2 m²/g and 40.2 min). LPIG had 31.4 m²/g EAI and 63.9 min ESI; LPIGH reached 67.2 m²/g and 67.8 min; LPIGHM showed the highest at 74.2 m²/g and 71.6 min, corresponding to 10.3- and 1.8-fold increases, respectively, and demonstrating the superior emulsifying performance of the wet-heated conjugate prepared from microwave-extracted LPI, aligning with microwave-treated soy protein-pectin conjugates exhibiting significantly improved emulsion stabilization capacity (Xiuxiu et al., 2023).

The improved foaming and emulsifying properties of LPI-GOS conjugates can be linked to the combined effects of wet-heating Maillard glycation and LPI extraction history. Reduced particle size, lower surface hydrophobicity and partial unfolding likely improved molecular flexibility and dispersion, facilitating adsorption at air-water and oil-water interfaces. Although LPIGHM showed a less negative zeta potential than LPIG and LPIGH, its smaller particle size, higher glycation level, reduced hydrophobicity and GOS-derived steric hydration may have contributed more strongly to interfacial stabilization than electrostatic repulsion alone. Thus, the functional performance of LPIGHM should be interpreted as the result of combined steric, hydration and conformational effects rather than zeta potential alone. The magnitudes are consistent with this reading. Surface hydrophobicity fell by a factor of about seven between LPI and LPIGHM (443.3 to 62.5 arbitrary units) and the hydrodynamic diameter by a factor of about twenty (1775.1 to 89.7 nm), while emulsifying activity rose about tenfold and foaming capacity about sixfold. A smaller, more hydrated and more flexible particle diffuses to a fresh interface faster and unfolds there more readily, so the rate-limiting step in film formation is relaxed; the attached GOS chains then extend into the aqueous phase and provide the steric and hydration barrier that keeps bubbles and droplets apart once the film has formed. The two contributions are separable in principle but not in these data, because particle size and surface hydrophobicity fell together across the series.

Pearson correlation analysis across the four treatment means was used to test these relationships directly (Table S3). With four points per coefficient the analysis is exploratory, and significance at P < 0.05 requires an absolute coefficient above 0.95, so the pattern across the matrix is more informative than any single value. The apparent degree of grafting was the single strongest predictor of functional performance, correlating positively with foaming capacity (r = 0.99), emulsifying activity index (r = 0.98), foam stability (r = 0.96) and ABTS scavenging (r = 1.00), and falling just short of significance for solubility (r = 0.95). Surface hydrophobicity ran a close second and in the opposite direction: it correlated negatively with solubility (r = −1.00), emulsifying activity index (r = −0.99) and ABTS scavenging (r = −0.97), and negatively though not significantly with foaming capacity (r = −0.92). Particle size correlated negatively with emulsion stability index (r = −0.98) and DPPH scavenging (r = −0.95), and β-sheet content correlated negatively with the same two responses (r = −0.99 and −1.00). The direction of every one of these coefficients is the one the mechanism predicts: smaller, less hydrophobic and less ordered particles perform better at the interface and scavenge radicals more effectively. Zeta potential behaved differently. It showed no strong association with any interfacial property, the largest coefficient being with total antioxidant capacity (r = −0.91), which supports the argument made above that stabilization in this system is steric and hydration-driven rather than electrostatic. Superoxide anion scavenging and FRAP correlated with none of the structural variables, including the degree of grafting, consistent with the assay-dependent antioxidant pattern described in Section 3.4. These correlations do not establish causality, and with four treatments they cannot; they indicate that the structural and functional changes move together in the way the proposed mechanism requires.

3.4. Biological activities

Enzyme inhibition responded to conjugation differently depending on the enzyme. α-Amylase inhibition was highest for native LPI at every concentration tested, reaching 73.6% at 100 mg/mL, and was not improved by conjugation: LPIG, LPIGH and LPIGHM gave 67.3%, 68.2% and 70.9% respectively at that concentration, where the differences among treatments were not significant, and the gap widened as concentration fell, with LPIGHM dropping to 10.9% at 12.5 mg/mL against 40.9% for native LPI (Fig. 6A). A steric explanation fits this pattern. α-Amylase is an endo-acting enzyme with an extended substrate-binding cleft, and inhibition by a protein depends on that protein occupying or obstructing the cleft; the hydrated GOS chains carried on the conjugate surface hinder rather than assist such an interaction, and the effect becomes more evident as the amount of available material falls.

Fig. 6.

Inhibition of α-amylase (A), α-glucosidase (B), and angiotensin-converting enzyme (ACE) (C), and antiproliferative activity against MCF-7 and Caco-2 cells (D) of LPI and LPI-GOS systems. LPI: conventionally extracted lupin protein isolate; LPIG: non-heated mixture of conventionally extracted LPI and GOS; LPIGH: wet-heated LPI-GOS conjugate prepared from conventionally extracted LPI; LPIGHM: wet-heated LPI-GOS conjugate prepared from microwave-extracted LPI.

α-Glucosidase behaved in the opposite way. Native LPI was the weakest inhibitor at every concentration, giving 17.0% at 100 mg/mL, whereas all three GOS-containing systems reached 60.0–66.1% at the same concentration, a roughly fourfold increase, and the difference between native LPI and the conjugates was significant (Fig. 6B). Conjugation therefore improved α-glucosidase inhibition substantially. The improvement was already present in the non-heated mixture LPIG (66.1%), which indicates that the GOS component contributes to it directly rather than the effect depending on covalent attachment alone. This enzyme-specific divergence is consistent with the architecture of the two active sites: α-glucosidase has a small pocket-shaped site that binds carbohydrate-like structures, so an oligosaccharide-decorated protein is a more plausible competitive inhibitor for it than for α-amylase. Native LPI gave a higher value at 12.5 mg/mL (44.3%) than at 100 mg/mL, a non-monotonic response that we attribute to aggregation of the unmodified isolate at the higher concentrations, which reduces the surface available for interaction with the enzyme; no comparable reversal occurred in any of the conjugated systems, all of which are far less aggregated (Section 3.2).

ACE inhibition was modest throughout and varied little with either treatment or dose, ranging from 7.0% to 14.9% across all systems and concentrations (Fig. 6C). LPIG was marginally the most active at 13.7–14.9% and native LPI the least active, but the differences are small and no consistent dose response was evident, so Maillard conjugation had at most a slight positive effect on ACE inhibition in this system. A comparable but larger enhancement has been reported for soy protein hydrolysates conjugated with mannose or allulose (Anuduang et al., 2024). In that work the starting material was already hydrolysed and the inhibitory peptide sequences were free in solution, whereas the intact conjugates assayed here would require gastrointestinal digestion before such sequences became available, which the present design did not include.

Taken together, the enzyme results show that the biofunctional consequences of LPI–GOS conjugation are enzyme-specific rather than uniformly beneficial: α-glucosidase inhibition improved markedly, ACE inhibition slightly, and α-amylase inhibition not at all.

This study reveals differential antiproliferative effects of LPI-GOS conjugates on MCF-7 breast and Caco-2 colorectal cancer cells under various processing conditions. Caco-2 cells were markedly more sensitive, with inhibition rates from 38.3% to 86.2%, versus minimal effects in MCF-7 cells (2.9–5.0%). LPI alone showed the highest inhibition in Caco-2 cells (86.2%), while LPIGHM, the wet-heated conjugate prepared from microwave-extracted LPI, showed higher activity (61.1%) than LPIG (38.3%) and LPIGH (44.9%) conjugates (Fig. 6D). Trend analysis suggests that using microwave-extracted LPI partially preserved bioactivity after wet-heating conjugation. The limited MCF-7 response indicates cell-specific mechanisms, with Caco-2 cells more susceptible to lupin conjugate antiproliferative effects.

The LPI extract exhibited dose-dependent antimicrobial activity, with Staphylococcus aureus demonstrating the highest susceptibility (90.3% to 58.2% inhibition), followed by Listeria monocytogenes (70.1–58.2%), Escherichia coli O157:H7 (75.2–35.7%), and Salmonella Typhimurium (79.1–27.5%). The combination of LPI with LPIG showed moderate, strain-specific effects, being most effective against S. aureus (86.8–41.5%) and L. monocytogenes (86.4–22.6%), while efficacy against E. coli and S. typhimurium declined (from 86.8% and 60.9% to 29.0% and 22.6%, respectively). LPIGH displayed enhanced activity at higher concentrations, particularly against L. monocytogenes (96.2–43.2%) and E. coli (90.0–21.2%), with strong responses from S. aureus (81.5–43.2%) and S. typhimurium (93.9–28.0%). LPIGHM provided the most balanced and consistent inhibition across all strains (72.5–83.1% to 19.7–41.1%), indicating broader-spectrum and stable antimicrobial activity (Fig. S3). All antimicrobial values in this section and in Fig. S3 are expressed as percentage inhibition of growth relative to an untreated control, not as inhibition-zone diameters. These results align with studies showing that microwave processing enhances the functional properties and the bioactivity of polysaccharide complexes, improving their interaction with bacterial pathogens (Mohammadi et al., 2023; Subhash et al., 2024).

Antioxidant activities differed among treatments and increased with concentration across all assays (Table 1). For DPPH scavenging, all samples showed dose-dependent activity, with conjugated samples performing slightly better than native LPI. At 100 mg/mL, LPIGHM showed the highest activity (79.6%), followed by LPIGH (79.5%) and LPIG (79.0%), while LPI had the lowest value (77.9%). These modest but consistent improvements after conjugation agree with previous reports on protein–polysaccharide Maillard conjugates (Kan et al., 2024; Liu, Hou, et al., 2025). ABTS•⁺ scavenging showed clearer treatment effects, particularly at lower concentrations. At 100 mg/mL, LPIGHM exhibited the highest activity (68.4%), followed by LPIGH (67.5%), LPIG (65.9%), and LPI (64.9%). At 12.5 mg/mL, LPIGHM showed a markedly higher activity (43.9%) compared with LPI (12.8%), indicating enhanced antioxidant efficiency in the wet-heated conjugate prepared from microwave-extracted LPI. This stronger ABTS response compared with DPPH is consistent with earlier studies on protein–polysaccharide conjugates (Kan et al., 2024; Liu, Hou, et al., 2025). Superoxide anion scavenging activity varied substantially among treatments. LPIG showed the highest activity across all concentrations, reaching 99.7% at 100 mg/mL, while LPI exhibited the lowest values. LPIGH and LPIGHM showed intermediate activity, suggesting that the non-heated LPI-GOS mixture was more effective for superoxide scavenging. In contrast, reducing power and total antioxidant capacity were mainly enhanced by heat-assisted conjugation. LPIGH showed the highest FRAP (100.0 μg/mL) and TAC (741.3 μg/mL) values at 100 mg/mL, while LPIG and LPIGHM showed smaller or moderate improvements. Overall, the results indicate assay-dependent antioxidant enhancement, with LPIGHM favoring ABTS radical scavenging and LPIGH showing the strongest reducing power and total antioxidant capacity. Divergence of this kind between assays is expected rather than anomalous, since DPPH, ABTS, superoxide scavenging, FRAP and TAC each probe a different reaction (hydrogen atom transfer, single electron transfer or metal reduction) and respond differently to the solvent polarity and pH of the assay medium (Alias & Shafie, 2025). The pattern reported here therefore reflects the chemistry of the individual assays as much as a single underlying antioxidant capacity, and no one assay should be taken as representative on its own.

Table 1.

In vitro antioxidant activities of LPI, LPIG, and conjugates at different concentrations (12.5–100 mg/mL).

Antioxidant activities2	Concentration (mg/mL)	LPI1	LPIG	LPIGH	LPIGHM
DPPH (%)	100	77.9 ± 0.24dA,3	79.0 ± 0.24dB	79.5 ± 0.26dB	79.6 ± 0.26dB
	50	74.9 ± 0.31cB	73.3 ± 0.26cA	73.3 ± 0.26cA	77.3 ± 0.26cC
	25	69.2 ± 0.24bB	71.1 ± 0.25bC	67.9 ± 0.26bA	73.8 ± 0.26bD
	12.5	66.8 ± 0.24aC	59.5 ± 0.26aA	64.3 ± 0.26aB	69.0 ± 0.26aD
ABTS (%)	100	64.9 ± 0.23dA	65.9 ± 0.23dB	67.5 ± 0.26dC	68.4 ± 0.26dD
	50	47.1 ± 0.23cA	48.0 ± 0.26cB	55.5 ± 0.26cC	62.4 ± 0.26cD
	25	43.1 ± 0.26bC	35.4 ± 0.26bB	21.2 ± 0.26bA	58.4 ± 0.26bD
	12.5	12.8 ± 0.26aB	11.1 ± 0.26aA	11.3 ± 0.26aA	43.9 ± 0.26aC
SAS (%)	100	32.7 ± 0.21dA	99.7 ± 0.25dD	53.4 ± 0.14dC	47.4 ± 0.14dB
	50	31.2 ± 0.26cA	67.6 ± 0.13cD	48.2 ± 0.28cC	35.1 ± 0.24cB
	25	28.7 ± 0.42bA	49.8 ± 0.59bD	36.8 ± 0.29bC	30.9 ± 0.42bB
	12.5	24.7 ± 0.26aB	42.1 ± 0.21aD	34.1 ± 0.24aC	17.6 ± 0.13aA
FRAP (μg/mL)	100	72.6 ± 0.78dB	57.8 ± 0.83dA	100.0 ± 0.76dC	58.3 ± 0.82dA
	50	63.4 ± 0.75cC	54.8 ± 0.78cB	67.5 ± 0.78cD	51.5 ± 0.76cA
	25	53.4 ± 0.77bB	51.6 ± 0.80bB	61.7 ± 0.76bC	47.3 ± 0.83bA
	12.5	47.4 ± 0.72aB	44.7 ± 0.79aA	46.5 ± 0.84aAB	45.2 ± 0.78aA
TAC (μg/mL)	100	535.1 ± 0.37cA	681.0 ± 0.73dC	741.3 ± 0.37dD	587.1 ± 0.93cB
	50	534.7 ± 0.73cA	539.3 ± 0.77cB	590.0 ± 0.64cD	586.2 ± 0.73cC
	25	521.0 ± 0.64bA	528.3 ± 0.37bB	565.8 ± 0.64bC	565.2 ± 0.73bC
	12.5	515.0 ± 0.43aB	501.1 ± 0.80aA	552.1 ± 0.37aD	540.8 ± 0.64aC

^(a-d/A-D) Different lowercase letters within the same column indicate significant differences among concentrations for the same treatment, while different uppercase letters within the same row indicate significant differences among treatments at the same concentration (p < 0.05).

¹LPI, lupin protein isolate obtained by conventional alkaline extraction; LPIG, non-heated mixture of conventionally extracted LPI and galacto-oligosaccharides (GOS); LPIGH, wet-heated LPI-GOS conjugate prepared from conventionally extracted LPI; LPIGHM, wet-heated LPI-GOS conjugate prepared from microwave-extracted LPI.

²DPPH: 2,2-diphenyl-1-picrylhydrazyl radical scavenging activity; ABTS: 2,2′-azino-bis(3-ethylbenzothiazoline-6-sulfonic acid) radical cation scavenging activity; SAS: superoxide anion scavenging activity; FRAP: ferric reducing antioxidant power, measured at 593 nm; TAC: total antioxidant capacity, measured at 695 nm. Both were quantified against an ascorbic acid standard curve and are expressed as μg ascorbic acid equivalents per mL of assayed sample solution. Values are reported on a per-volume rather than a per-gram basis because all treatments were assayed at matched concentrations over 12.5–100 mg/mL.

³Values are expressed as mean ± standard deviation (n = 3).

3.5. In vitro fecal fermentation

3.5.1. SCFAs and fermentation parameters

Three limitations should be kept in view when reading this section. First, the fermentation was carried out on the prepared LPI-GOS systems without prior simulated gastrointestinal digestion, so the substrate presented to the microbiota is not the one that would reach the colon in vivo. Second, the system is closed, with no absorption of acids across an epithelium and no host contribution, so acid concentrations reflect production and consumption within the vessel alone. Third, the functional pathways reported in Section 3.5.2 are predicted from 16S rRNA data rather than measured. With those constraints stated, SCFA concentrations showed treatment- and time-dependent changes rather than continuous accumulation. The relatively high initial SCFA values suggest that part of the measured acids may have originated from the basal medium, inoculum or sample matrix at the start of fermentation. During incubation, SCFA levels generally declined, which may reflect microbial cross-feeding, utilization of acids as secondary substrates, or matrix-dependent buffering and metabolism in the closed in vitro system. Therefore, the results are best interpreted as treatment-dependent SCFA retention and stabilization rather than direct evidence of net SCFA production (Kan et al., 2024; Liu, Hou, et al., 2025).

The design allows a partial separation of the contributions of GOS and of the modified protein. LPIG contains the same protein and the same mass of GOS as LPIGH and LPIGHM but in unbound form, so any difference between LPIG and the two conjugates is attributable to covalent attachment rather than to the presence of GOS as such. On that comparison the conjugates retained more acetate at 24 h than LPIG (476.2 and 420.8 versus 413.3 μmol/L) and produced substantially more gas (11.4 and 13.9 versus 8.8 mL), indicating that binding the oligosaccharide to the protein changed the fermentation rather than simply delivering the same substrate. The likely reason is that covalently bound GOS is released gradually as the protein backbone is degraded, whereas free GOS is fermented rapidly and early. What the design cannot do is separate the contribution of the protein moiety itself, since no protein-only fermentation control was included; that comparison would require a separate experiment.

Acetic acid showed strong treatment and time dependency. At 0 h, LPIGH and LPIGHM exhibited the highest acetic acid concentrations (934.4 and 904.1 μmol/L, respectively), while NC and LPIG showed lower levels (636.6 and 593.7 μmol/L). During fermentation, acetic acid generally decreased in NC, reaching 212.1 μmol/L at 24 h. In contrast, LPIG, LPIGH, and LPIGHM maintained relatively higher concentrations at 24 h (413.3, 476.2, and 420.8 μmol/L, respectively), indicating a more stable acetic acid production in treated samples.

A similar dominance and stabilization of acetate in conjugated systems was reported in in vitro fermentation of whey protein–gum Arabic Maillard conjugates and chickpea protein–citrus pectin conjugate emulsions, where conjugation enhanced sustained saccharolytic fermentation (Kan et al., 2024; Liu, Hou, et al., 2025). In contrast, in vivo models showed net acetate accumulation rather than decline, which reflects continuous microbial activity and host absorption processes absent in in vitro systems (Kan et al., 2022).

Propionic acid also varied among treatments. At the start of fermentation, LPIGH and LPIGHM showed the highest propionic acid levels (165.5 and 164.2 μmol/L), followed by NC (123.4 μmol/L) and LPIG (85.1 μmol/L). Over time, propionic acid decreased in all treatments. By 24 h, LPIGH still retained the highest level (46.4 μmol/L), while LPIG and NC showed moderate values (34.4 and 33.5 μmol/L), and LPIGHM recorded the lowest concentration (29.7 μmol/L) (Table 2).

Table 2.

SCFA profile, pH, and gas generation during the in vitro fecal fermentation.

Attribute	Time (h)	NC1	LPIG	LPIGH	LPIGHM
pH	0	6.9 ± 0.02aB2	5.8 ± 0.07aA	5.9 ± 0.08aA	6.0 ± 0.03aA
	6	6.8 ± 0.10aB	6.0 ± 0.39aA	6.1 ± 0.39aA	6.1 ± 0.23aA
	12	6.8 ± 0.16aB	6.0 ± 0.37aA	6.2 ± 0.29aA	6.1 ± 0.19aA
	24	6.8 ± 0.12aB	6.0 ± 0.40aA	6.2 ± 0.36aA	6.2 ± 0.21aA
Gas (mL)	0	3.9 ± 0.58aA	7.1 ± 3.24aAB	9.3 ± 3.80aAB	11.3 ± 4.48aB
	6	4.0 ± 0.69aA	7.2 ± 3.34aAB	9.4 ± 3.89aAB	11.5 ± 4.65aB
	12	4.3 ± 0.63aA	6.6 ± 2.70aAB	8.9 ± 3.35aAB	11.0 ± 4.14aB
	24	4.4 ± 0.54aA	8.8 ± 0.93aAB	11.4 ± 0.78aB	13.9 ± 0.75aB
Acetic acid (μmol/L)	0	636.6 ± 45.34bA	593.7 ± 44.60bA	934.4 ± 18.40cB	904.1 ± 36.41cB
	6	642.8 ± 72.06bBC	683.1 ± 45.04bC	417.2 ± 54.04aA	553.1 ± 36.03bB
	12	272.5 ± 9.31aA	389.5 ± 27.37aB	547.4 ± 18.54bC	453.0 ± 45.38aBC
	24	212.1 ± 21.67aA	413.3 ± 40.02aB	476.2 ± 67.68abB	420.8 ± 47.89aB
Propionic acid (μmol/L)	0	123.4 ± 2.10dB	85.1 ± 5.05cA	165.5 ± 5.09dC	164.2 ± 3.07dC
	6	96.3 ± 5.02cB	105.2 ± 5.02dB	66.5 ± 6.02bA	106.3 ± 6.02cB
	12	79.1 ± 3.18bB	67.0 ± 6.97bA	107.4 ± 7.15cC	88.4 ± 8.09bB
	24	33.5 ± 6.29aA	34.4 ± 3.64aAB	46.4 ± 5.09aB	29.7 ± 5.72aA
Butyric acid (μmol/L)	0	20.6 ± 1.88cB	13.7 ± 0.60cA	22.2 ± 2.77bB	12.7 ± 0.31cA
	6	7.7 ± 0.56bA	8.4 ± 0.47bA	7.7 ± 0.56aA	9.5 ± 0.56bA
	12	1.7 ± 0.19aA	5.5 ± 0.19aBC	6.7 ± 0.47aC	4.3 ± 0.66aB
	24	3.0 ± 0.55aA	4.2 ± 0.77aAB	6.0 ± 0.65aB	5.7 ± 0.29aB

^(a-d/A-D) Different lowercase letters within the same column indicate significant differences among fermentation times for the same treatment, while different uppercase letters within the same row indicate significant differences among treatments at the same time point (p < 0.05).

¹NC, negative control; LPIG, non-heated mixture of conventionally extracted lupin protein isolate (LPI) and galacto-oligosaccharides (GOS); LPIGH, wet-heated LPI-GOS conjugate prepared from conventionally extracted LPI; LPIGHM, wet-heated LPI-GOS conjugate prepared from microwave-extracted LPI.

²Values are expressed as mean ± standard deviation (n = 3).

This trend agrees well with in vitro reports showing that propionate production is highly responsive to protein–polysaccharide conjugation but may decline at later fermentation stages due to secondary microbial utilization or cross-feeding (Kan et al., 2024; Liu, Hou, et al., 2025). In vivo studies reported propionate enrichment under conjugate treatment, but without the same temporal decline, again reflecting differences between in vitro and in vivo fermentation kinetics (Kan et al., 2022).

Butyric acid concentrations were generally lower compared to other acids and showed limited accumulation during fermentation. At 0 h, LPIGH and NC had slightly higher butyric acid levels (22.2 and 20.6 μmol/L), while LPIG and LPIGHM showed lower values. As fermentation progressed, butyric acid decreased in all treatments. At 24 h, LPIGH and LPIGHM maintained relatively higher levels (6.0 and 5.7 μmol/L), compared to LPIG (4.2 μmol/L) and NC (3.0 μmol/L) (Table 2).

This modest butyrate response is in agreement with in vitro fermentation studies, where acetate and propionate dominate early fermentation, while butyrate production remains limited due to the slower response of butyrate-producing bacteria (Kan et al., 2024; Liu, Hou, et al., 2025). In contrast, stronger butyrate recovery has been reported in in vivo models following conjugate supplementation, likely due to enrichment of butyrate-producing taxa under physiological conditions (Kan et al., 2022).

Cumulative gas production increased gradually over time for all treatments. NC consistently showed the lowest gas production throughout fermentation, reaching 4.4 mL at 24 h. In contrast, LPIGHM produced the highest gas volume at all time points, reaching 13.9 mL at 24 h. LPIGH also showed high gas production (11.4 mL at 24 h), while LPIG exhibited intermediate values (8.8 mL at 24 h) (Table 2). These results indicate that the treated samples enhanced microbial fermentation activity compared to the control. Higher gas production in conjugated systems has been consistently reported in in vitro fermentation studies and is generally associated with increased carbohydrate fermentability and microbial metabolic activity (Kan et al., 2024; Liu, Hou, et al., 2025). Gas production was not assessed in vivo, therefore direct comparison with animal models is not applicable (Kan et al., 2022).

Changes in pH during fermentation were relatively small but treatment-dependent. At 0 h, NC had the highest pH value (6.9), while LPIG, LPIGH, and LPIGHM ranged between 5.8 and 6.0. During fermentation, pH values remained fairly stable for all treatments. NC maintained a higher pH around 6.8 across all time points, whereas LPIG, LPIGH, and LPIGHM stabilized around 6.0–6.2 by 24 h. This suggests that the treatments slightly moderated acidification during fermentation. Comparable pH stabilization under conjugate treatment has been reported in in vitro studies using protein–polysaccharide conjugates, where buffering effects and gradual SCFA release limited sharp pH drops (Kan et al., 2024; Liu, Hou, et al., 2025). Greater pH fluctuations observed in in vivo studies are likely influenced by host absorption and intestinal buffering mechanisms (Kan et al., 2022) (Table 2).

3.5.2. Gut microbiota modulation

The effects of LPI–GOS conjugates on gut microbiota were examined using an in vitro fecal fermentation model. Analysis of the Venn diagram (Fig. 7A) showed that 292 OTUs were shared among all groups (NC, LPIG, LPIGH, and LPIGHM), suggesting the presence of a stable core microbiota that was not eliminated by treatment. At the same time, each group contained a number of unique OTUs, with NC showing 20, LPIG 7, LPIGH 21, and LPIGHM 2 unique features. Partial overlaps between groups ranged from 10 to 33 OTUs, indicating that although the core community was conserved, conjugation and LPI extraction method induced treatment-specific microbial shifts. This pattern agrees with previous reports showing that protein–polysaccharide conjugates modulate microbial composition without disrupting the conserved core microbiota during in vitro fermentation, as observed for whey protein–gum Arabic Maillard conjugates and conjugate-stabilized emulsions (Kan et al., 2022; Kan et al., 2024). The very low number of unique OTUs in LPIGHM, two against 20 for the control and 21 for LPIGH, deserves a cautious reading. It indicates that this treatment established almost no distinct taxa of its own, and that its effect on the community operated by redistributing the abundance of organisms already present rather than by recruiting new ones. Whether that is desirable depends on the objective. It is favorable for a food ingredient, in that the core community is left intact, but it also means the shifts reported below are shifts in proportion rather than evidence that LPIGHM supports a distinct microbial population.

Fig. 7.

Venn diagram of shared and unique OTUs (A), alpha-diversity indices with Kruskal–Wallis p-values (B), NMDS ordination with ANOSIM statistics and stress value (C), principal coordinate analysis with group ellipses (D), UPGMA clustering with genus-level relative abundance (E), and genus-level phylogenetic tree (F), following in vitro fecal fermentation. NC: negative control; LPIG: non-heated mixture of conventionally extracted LPI and GOS; LPIGH: wet-heated LPI-GOS conjugate prepared from conventionally extracted LPI; LPIGHM: wet-heated LPI-GOS conjugate prepared from microwave-extracted LPI. Panel A is discussed in the first paragraph of Section 3.5.2, panel B in the third, panel F in the fourth, and panels C, D and E in the fifth; higher-resolution versions of the composition data are provided in Fig. S5A and S5B.

Alpha diversity rarefaction curves (Fig. S4) based on Ace, Chao, and observed species indices increased gradually with sequencing depth and then reached a plateau, confirming that sequencing coverage was sufficient to capture most microbial taxa. Shannon and Simpson indices indicated higher diversity and improved evenness in conjugate-treated samples, especially LPIGHM. Coverage values exceeded 99% for all treatments, supporting the reliability of the dataset. These findings are consistent with earlier studies on protein–polysaccharide Maillard conjugates, where conjugation enhanced microbial diversity and community evenness by providing sustained fermentable substrates (Kan et al., 2024). A similar stabilization of diversity indices under conjugate treatment was also reported for protein–polysaccharide emulsion systems delivering bioactives (Liu, Hou, et al., 2025).

Further alpha- and beta-diversity analysis (Fig. 7B) demonstrated clear treatment-dependent effects on microbial structure. Species richness was slightly higher in the control group than in LPIG (329 against 304 observed species, P = 0.044), whereas community evenness was clearly improved in LPIGHM, as reflected by a higher Shannon index and a lower Simpson index (Shannon 3.23 for LPIGHM against 1.85 for the control, P = 0.016; Simpson 0.11 against 0.40, P = 0.019; Kruskal–Wallis). This observation aligns with studies showing that conjugation, rather than the protein or polysaccharide alone, favors a more balanced microbial community by slowing digestion and extending substrate availability in the colonic phase (Kan et al., 2024; Liu, Hou, et al., 2025). In contrast, some in vivo models reported a simultaneous increase in richness and evenness following conjugate supplementation (Kan et al., 2022). This difference is likely due to the simplified in vitro system used here, which lacks host–microbe interactions and typically limits overall richness expansion.

Microbial composition analysis (Fig. S5) revealed clear shifts at both phylum and genus levels following conjugate treatment (Fig. S5A,S5B). Conjugate-supplemented groups showed increased abundance of beneficial genera such as Bifidobacterium and Lactobacillus, along with reduced levels of opportunistic taxa including Escherichia and Klebsiella. LEfSe analysis identified discriminative taxa associated with each treatment (Fig. S5E), and the genus-level phylogenetic relationships among the taxa detected are shown in Fig. 7F. These trends strongly agree with findings from whey protein–gum Arabic Maillard conjugates and conjugate-stabilized emulsions, where enrichment of Bifidobacterium and suppression of Enterobacteriaceae were consistently observed (Kan et al., 2022; Kan et al., 2024). Comparable shifts toward beneficial taxa were also reported for chickpea protein isolate–citrus pectin conjugate emulsions, supporting the role of conjugation in shaping gut microbial composition (Liu, Hou, et al., 2025).

Beta-diversity analyses further confirmed distinct clustering among treatments, by NMDS (Fig. 7C), principal coordinate analysis (Fig. 7D) and UPGMA clustering (Fig. 7E). These group differences were supported by ANOSIM (R = 0.91, P = 0.001; Fig. S5D), and the NMDS ordination in Fig. 7C returned comparable statistics (R = 0.91, P = 2 × 10⁻⁴) with a stress value of 0. A stress of zero indicates that the two-dimensional ordination reproduces the rank order of the distance matrix exactly, which is expected with only twelve samples rather than a sign of an unusually good fit; the ordination is therefore presented as a visual summary of the group separation, with the statistical support provided by ANOSIM. NC samples showed greater dispersion, suggesting higher inter-sample variability, while LPIG formed a more compact cluster. LPIGH and LPIGHM showed intermediate but clearly separated clusters, with enrichment of genera such as Parabacteroides, Ruminococcus, Succinivibrio, and Catenibacterium. This agrees with previous reports indicating that Maillard conjugation alters substrate accessibility and selectively promotes saccharolytic and SCFA-producing bacteria (Kan et al., 2024). The stronger shift observed in LPIGHM is consistent with findings showing that structural modification of conjugates enhances microbial selectivity and metabolic specialization (Liu, Hou, et al., 2025).

Functional prediction using PICRUSt2 revealed that metabolic pathways dominated across all treatments (Fig. 8A–8C). The KEGG pathway heatmap is provided in Fig. S5C. Enrichment of carbohydrate, amino acid, lipid, and energy metabolism pathways was particularly pronounced in LPIGHM. These functional enhancements are in agreement with whey protein–gum Arabic conjugate studies, where conjugation enriched carbohydrate and amino acid metabolism pathways and reduced inflammatory signaling (Kan et al., 2024). Similar metabolic pathway enrichment related to energy and amino acid metabolism was reported for protein–polysaccharide conjugate emulsions delivering curcumin, indicating functional adaptation of gut microbiota to structurally modified substrates (Liu, Hou, et al., 2025). It should be emphasized that PICRUSt2 infers gene content from 16S rRNA marker genes against reference genomes and does not measure gene abundance, transcription or metabolite production. The pathway differences described here are therefore predictions of metabolic potential, and the enrichment reported for LPIGHM indicates that the taxa favored by this treatment carry the relevant genes, not that those pathways were more active during fermentation.

Fig. 8.

Predicted microbial functionality of LPI-GOS systems after in vitro fecal fermentation based on PICRUSt2 analysis: KEGG pathways (A-C), COG functional categories (D), MetaCyc pathways (E), and microbial co-occurrence network correlations (F). NC: negative control; LPIG: non-heated mixture of conventionally extracted LPI and GOS; LPIGH: wet-heated LPI-GOS conjugate prepared from conventionally extracted LPI; LPIGHM: wet-heated LPI-GOS conjugate prepared from microwave-extracted LPI. Panels A–C are discussed in the sixth paragraph of Section 3.5.2 and panels D–F in the seventh; the corresponding KEGG heatmap is provided in Fig. S5C.

COG and MetaCyc analyses further supported enhanced biosynthetic and metabolic capacity in LPIGHM, including pathways linked to protein synthesis, nucleotide metabolism, and vitamin biosynthesis (Fig. 8D and E). These findings agree with studies demonstrating that conjugate-based systems enhance microbial functional diversity and metabolic potential through delayed digestion and sustained fermentation (Kan et al., 2024; Liu, Hou, et al., 2025). Correlation and co-occurrence network analysis revealed stronger positive interactions among beneficial taxa, particularly Bifidobacterium longum and Blautia species, whereas Escherichia showed negative associations (Fig. 8F). Enhanced microbial connectivity under conjugate treatment has also been reported for protein–polysaccharide conjugate systems, supporting improved network stability and cooperative microbial behavior (Kan et al., 2022).

Overall, our results demonstrate that LPI-GOS conjugation, particularly when prepared from microwave-extracted LPI, effectively reshaped gut microbial communities during in vitro fecal fermentation. While a conserved core microbiota was maintained across all treatments, conjugate supplementation promoted clear shifts in community structure, diversity, and functional potential toward a more balanced and metabolically active profile. The LPIGHM group consistently showed improved microbial evenness, stronger clustering patterns, enrichment of beneficial taxa, and enhanced functional pathways related to carbohydrate and energy metabolism, indicating improved microbial adaptation to the conjugated substrate. These observations are in line with previous studies on protein–polysaccharide conjugates and conjugate-stabilized emulsions, which reported that structural modification of proteins through glycation improves microbial accessibility and fermentation efficiency. Minor differences in richness trends compared with some in vivo studies likely reflect the limitations of the in vitro model, where host interactions are absent. Overall, the findings suggest that wet-heating Maillard glycation of microwave-extracted LPI conjugates represents an effective strategy to modulate gut microbiota composition and functionality without disrupting microbial core stability.

Microwave-assisted extraction was used in this study as a protein pre-conditioning strategy to improve subsequent wet-heating Maillard conjugation of LPI with GOS, rather than as an independent treatment to evaluate the direct effects of microwave processing on foaming, emulsification, antioxidant activity or gut microbial responses. Therefore, the improved responses observed for LPIGHM should be interpreted mainly in relation to enhanced glycation efficiency, structural rearrangement and conjugate formation. In addition, fecal fermentation was conducted using the prepared LPI-GOS systems without prior simulated gastrointestinal digestion to maintain consistent substrate concentrations across treatments at the start of fermentation and to allow direct comparison of their fermentation behavior. Two consequences follow. First, because microwave extraction and wet heating were applied in sequence and not in a factorial arrangement, the design does not isolate the independent contribution of each; a microwave-extracted LPI carried through without conjugation would be required for that, and its absence means the effects reported for LPIGHM are effects of the combined process. Second, omitting the digestion step keeps the substrate concentration equal across treatments and makes the fermentation comparison clean, but it also means the material entering the vessel is more intact than what would reach the colon after gastric and small-intestinal transit. Digestion would be expected to hydrolyze part of the protein backbone and to release a portion of the bound GOS earlier, so the fermentation profiles reported here should be read as a comparison between the systems as prepared rather than as a prediction of colonic behavior.

4. Conclusion

Wet-heating Maillard glycation of microwave-extracted LPI produced a conjugate that differed from its conventionally extracted counterpart in a consistent way, and the first two aims of the study are supported. Microwave-assisted extraction left the protein in a state that reacted further with GOS under identical wet-heating conditions, giving the highest apparent degree of glycation, the largest fall in β-sheet content and the greatest loss of surface hydrophobicity; these structural changes translated into the strongest foaming, emulsifying, water- and oil-holding and ABTS-scavenging performance of the four systems. The third aim is only partly met. LPIGHM gave the greatest gas volume and the most even microbial community and retained more acetate at 24 h than the control, but SCFA concentrations declined over the incubation rather than accumulating, and the PICRUSt2 outputs are predictions rather than measurements, so the metabolic claims remain provisional. Confirming them requires evidence that the present design cannot supply: mapping of the glycation sites by LC-MS/MS or MALDI-TOF mass spectrometry, quantification of advanced glycation end-products and of available lysine to establish the nutritional cost of the reaction, cell-based antioxidant assays alongside the chemical ones, simulated gastrointestinal digestion before fermentation, and direct measurement of microbial genes or metabolites in place of predicted pathways. Performance in a real food matrix, and the energy and cost balance of combining microwave extraction with wet heating at production scale, would also need to be established before the approach could be recommended for manufacturing.

Abbreviations

LPI	lupin protein isolate obtained by conventional alkaline extraction
LPIG	non-heated mixture of conventionally extracted LPI and galacto-oligosaccharides (GOS)
LPIGH	wet-heated LPI-GOS conjugate prepared from conventionally extracted LPI
LPIGHM	wet-heated LPI-GOS conjugate prepared from microwave-extracted LPI
MAE	microwave-assisted extraction
OPA	o-phthalaldehyde
DPPH	2,2-diphenyl-1-picrylhydrazyl
ABTS	2,2′-azino-bis(3-ethylbenzothiazoline-6-sulfonic acid)
FRAP	ferric reducing antioxidant power
TAC	total antioxidant capacity
SAS	superoxide anion scavenging
SCFA	short-chain fatty acids
ACE	angiotensin-converting enzyme
NC	negative control
PICRUSt2	Phylogenetic Investigation of Communities by Reconstruction of Unobserved States 2
KEGG	Kyoto Encyclopedia of Genes and Genomes
COG	Clusters of Orthologous Groups
MetaCyc	Metabolic Pathway Database
OTU	operational taxonomic unit
NMDS	non-metric multidimensional scaling
UPGMA	unweighted pair group method with arithmetic mean
PCoA	principal coordinate analysis.

Availability of data and materials

All data supporting the findings of this study are available within the paper and its supplementary materials. The raw gut microbiota data are available from the corresponding author upon reasonable request.

CRediT authorship contribution statement

Naeem Ullah: Writing – original draft, Visualization, Validation, Methodology, Investigation, Formal analysis, Data curation. Gafar Bamigbade: Writing – review & editing, Validation, Methodology. Maduni Jayahansi Paththuwe Arachchi: Writing – review & editing, Validation, Methodology. Abdelmoneim Ali: Writing – review & editing, Supervision, Resources. Afaf Kamal-Eldin: Writing – review & editing, Supervision, Resources. Fuzhen Zhou: Writing – review & editing, Validation, Funding acquisition. Song Miao: Writing – review & editing, Validation, Funding acquisition. Mutamed Ayyash: Writing – review & editing, Validation, Supervision, Resources, Project administration, Methodology, Funding acquisition, Formal analysis, Conceptualization.

Informed consent statement

Informed consent was obtained from human subjects involved in the study. The protocol for fecal sample collection was approved by the UAEU-Human Research Ethics Committee (UAEU-HREC) of United Arab Emirates University, UAE (No. ERH_2024_5441).

Funding

This study was funded by the United Arab Emirates University via the AUA grant scheme (grant number #12F106) and the National Natural Science Foundation of China (Grant No. 32502408).

Declaration of competing interest

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

Appendix A. Supplementary data

Supplementary material.

Data availability

Data will be made available on request.