Impact of Co-Fermentation with Bifidobacterium animalis subsp. lactis IU100 and Type III Resistant Starch on the Aroma Profile of Fermented Milk

TL;DR

Adding B. lactis IU100 and resistant starch to fermented milk improves flavor, texture, and fermentation time by enhancing key aroma compounds and metabolic pathways.

Contribution

The study introduces a combined strategy using B. lactis IU100 and RS3 to optimize flavor and texture in fermented milk through metabolic reprogramming.

Findings

Co-supplementation reduced fermentation time by 1 hour and improved texture properties like hardness and water holding capacity.

115 key aromatic compounds were enriched, with ethyl caprylate and ethyl n-butyrate contributing fruity and creamy notes.

KEGG analysis showed 24 differential metabolites linked to purine and amino acid metabolism pathways.

Abstract

The addition of Bifidobacterium animalis subsp. lactis and prebiotics to fermented milk can enhance its flavor and sensory properties; however, research on the effects of their combined supplementation on flavor profiles remains limited. This study investigated the impact of simultaneously adding B. lactis IU100 and resistant starch type III (RS3) to fermented milk on flavor and texture. The results showed that co-supplementation shortened the fermentation time by 1 h. It also increased hardness by 28.8%, springiness by 1.14 mm, and water holding capacity by 12.45%, accompanied by the formation of a more continuous and dense gel network. Headspace solid-phase microextraction coupled with gas chromatography–mass spectrometry (HS-SPME-GC-MS) combined with odor activity value analysis indicated the enrichment of 115 key aromatic compounds, among which ethyl caprylate, ethyl n-butyrate, 1-octanol, and 2,3-heptanedione were identified as representative flavor compounds associated with fruity and creamy notes. KEGG pathway analysis revealed that 24 differential metabolites were predominantly enriched in purine metabolism and amino acid-related pathways. Within these pathways, coordinated enzymatic reactions convert α-keto acids and fatty acid metabolites into key flavor esters and catalyze the formation of volatile alcohols from amino acids and aromatic fatty acid precursors. Overall, this combined strategy effectively optimized fermentation efficiency, texture, and flavor through the targeted reprogramming of microbial metabolic flux.

Figures

Full text

1. Introduction

Fermented milk is widely consumed worldwide as a nutrient-dense food, and it also serves as an effective delivery system for viable beneficial microorganisms [1,2]. This intrinsic functional property has promoted the transformation of traditional fermented milk into enhanced functional variants. In recent decades, increasing consumer health awareness has substantially driven demand for functional dairy products, particularly yogurt formulated with probiotics and prebiotics to enhance health benefits [3]. The adequate consumption of probiotics and prebiotics is linked to a range of health benefits, such as regulating gut microbiota and improving abnormal glucose and lipid metabolism [4,5,6]. In addition to their health-promoting effects, the incorporation of these ingredients contributes to the formation of key flavor compounds and improves the textural properties of fermented milk.

The incorporation of probiotics or prebiotics can significantly enrich the profile of flavor compounds and improve the textural properties of fermented milk. For example, Lacticaseibacillus paracasei is an important adjunct culture in fermented dairy products, as its abundant carbohydrate-active enzymes efficiently degrade a wide range of carbon sources, including lactose, glycoproteins, and glycolipids within dairy matrices. Such enzyme-mediated action boosts the biogenic synthesis of short-length fatty acids, encompassing acids like butyrate and hexanoate, which function as key aroma-active metabolites and enhance the flavor complexity of the final products [7,8]. Similarly, Limosilactobacillus fermentum can produce characteristic flavor compounds, including dibenzoyl-L-tartaric acid anhydride—responsible for a distinctive wine-like aroma—through strain-specific glycoside hydrolase-mediated carbohydrate metabolism and amino acid transformation [9]. Moreover, the co-cultivation of Bifidobacterium with conventional starter cultures enhances proteolytic activity, promoting the release of peptides and free amino acids [10]. These compounds serve not only as essential nitrogen sources for microbial growth but also as important flavor precursors that are further metabolized into volatile aldehydes, ketones, and other aroma-active compounds, thereby enriching the overall sensory quality of yogurt [10].

Prebiotics can significantly increase hardness, improve cohesiveness and uniformity, and reduce centrifugal losses in fermented dairy products. Through their strong hygroscopicity and water binding capacity, prebiotics enhance protein–water interactions, thereby promoting the formation of a more viscous and elastic gel network. Such effects have been observed with prebiotics including inulin, polydextrose, xylooligosaccharides, and resistant starch [11,12]. Among these, resistant starch exhibits superior performance. Owing to its high water binding capacity and ability to form a supportive structural matrix, resistant starch markedly increases product viscosity and hardness. In addition, it enhances foam stability and fat destabilization while reducing the meltdown rate, collectively contributing to improved texture and overall stability [13]. Although the individual effects of Bifidobacterium and resistant starch have been well documented, studies examining their combined application in yogurt systems remain limited, particularly for Bifidobacterium animalis subsp. lactis (B. lactis) and resistant starch type III (RS3). Moreover, their potential interactions in modulating the textural and flavor attributes of fermented milk are not yet fully understood. Elucidating these interactions is both important and challenging for the development of fermented milk products that are optimized for both functionality and sensory quality.

Bifidobacterium animalis subsp. lactis IU100 (hereinafter referred to as IU100) is a novel probiotic strain isolated from the feces of healthy Chinese adults. More importantly, recent research has demonstrated that [14] IU100 shows significant immunomodulatory functions in a cyclophosphamide (CTX)-induced immunosuppressed mouse model. This study was designed to systematically investigate the synergistic effects of co-culturing B. lactis IU100 and RS3 with conventional yogurt starter cultures on fermentation performance, microstructure development, volatile compound profiles, and associated metabolic pathways in fermented milk. The fermentation kinetics and texture attributes were first evaluated, and microstructural alterations were visualized using scanning electron microscopy. HS-SPME-GC-MS was then applied to characterize the volatile flavor compounds and delineate the specific contribution of the probiotic–prebiotic combination to the overall flavor profile. To elucidate the underlying biochemical mechanisms, an untargeted metabolomics approach employing UHPLC–Q Exactive HF-X MS was performed, followed by a Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis of differentially abundant metabolites. This integrated analytical strategy successfully identified central metabolic pathways implicated in flavor biosynthesis. Together, these results clarify the metabolic networks regulating flavor formation and provide a robust scientific framework for the rational design of high-quality, functionally enhanced fermented dairy products with optimized sensory and potential health attributes, thereby contributing to advancements in dairy biotechnology and targeted synbiotic applications.

2. Materials and Methods

2.1. Preparation of the Milk Matrix

Milk (containing 3.2% (w/w) protein and 4.0% (w/w) fat) served as the experimental substrate. Granulated sugar was added at 5.0% (w/w), and the mixture was agitated using an electronic stirrer (FLUKO Technology, Shanghai, China) at 550 rpm until a homogenous blend was attained, followed by hydration for 30 min. RS3 was then incorporated into the milk base at 1.5% (w/w) and mixed using the same stirrer at 2000 rpm until a homogenous mixture was obtained. Afterward, the mixture was warmed up to 60 °C and underwent a two-step homogenization procedure using a high-pressure homogenizer (APV-2000, SPX Flow Technology Co., Ltd., Shanghai, China) at 60 °C under pressures of 5 MPa and 15 MPa. The resulting preparation served as the fermentation milk matrix.

2.2. Preparation of Set-Type Yogurt

Four types of set-type fermented milk samples were prepared:

• (1).Fermented milk with only the 4.0 starter culture;
• (2).Fermented milk supplemented with the 4.0 starter culture and B. lactis IU100;
• (3).Fermented milk containing the 4.0 starter culture and 1.5% RS3;
• (4).Fermented milk with the 4.0 starter culture, B. lactis IU100, and 1.5% RS3.

For each type, three replicate samples were prepared in parallel.

The dairy substrate underwent initial thermal processing following the procedures outlined in Section 2.1 (post-homogenization). A direct vat set starter culture (4.0), containing Streptococcus salivarius subsp. thermophilus (S. thermophilus) and Lactobacillus delbrueckii subsp. bulgaricus (L. bulgaricus) (provided by the Ministry of Education-Beijing Municipal Key Laboratory of Functional Dairy Products, China Agricultural University (Beijing, China)), was then added to the pretreated milk at a dosage of 100 g/t, and B. lactis IU100 was incorporated at a concentration of 5 × 10^6^ CFU/mL. This blend was incubated for fermentation at 42 °C until its titratable acidity attained 70 °T, so as to obtain set-style yogurt. Right after the fermentation process, the resulting yogurt was kept at 4 °C.

2.3. Measurement of pH, Titratable Acidity, and Viable Cell Counts

The pH of the set-type fermented milk samples was measured directly using a calibrated pH meter (Model S210, Mettler Toledo, Zurich, Switzerland). Titratable acidity was measured in accordance with the national standard of China, namely GB 5009.239–2016 [15]. Briefly, around 10 g of the prepared sample was precisely weighed out, mixed with an ethanol solution containing phenolphthalein as an indicator, and the sample solution was titrated using a 0.1 mol/L standard NaOH solution until the endpoint was reached. All measurement results were presented in the unit of degrees Thorner (°T) [16].

Lactic acid bacteria enumeration was carried out based on a method reported in prior work [17] with slight modifications made to the protocol. Specifically, an exact 1 g portion of every yogurt sample was weighed out and underwent serial dilution in 0.85% sterile physiological saline. Aliquots of 1 mL taken from each dilution gradient were spread evenly over MRS agar plates. To selectively enumerate B. animalis, Lin-Mupirocin was added to the medium at 5 mg/100 mL. Plates containing Lin-Mupirocin were incubated at 37 °C, whereas standard MRS agar plates (without Lin-Mupirocin) were incubated at 42 °C, both for 36–48 h. Colony count results were presented as log_10_ colony-forming units per gram (log_10_ CFU/g), and each dilution gradient was inoculated onto plates in three replicate trials.

2.4. Texture Analysis

Texture profile attributes of set-style yogurt, encompassing hardness and springiness, were assessed based on a revised approach reported by Moghiseh et al. [18], with the help of a Brookfield CTX texture analyzer (Ametek Brookfield, Middleborough, MA, USA). A TA10 probe was employed. Cylindrical yogurt samples (height: 30 mm; diameter: 40 mm) were carefully prepared and equilibrated to 4 °C prior to testing. The probe was aligned to contact the sample surface before initiating the test. The target strain was set at 35% of the original sample height. The probe’s descent, test, and return speeds were set at 1, 3, and 3 mm/s, respectively. A cylindrical probe (40 mm diameter) was used to penetrate the sample to a depth of 30 mm for each measurement.

2.5. Measurement of Water Holding Capacity

Water holding capacity (WHC) was assessed with a revised procedure derived from the method reported by Delikanli et al. [19]. Briefly, samples were accurately weighed to record their initial mass and then transferred into centrifuge tubes. The sample tubes were subjected to centrifugation at 4000× g for 15 min, following which the resulting supernatant was collected and weighed to determine its mass.

2.6. Microstructure Analysis

Scanning electron microscopy (SEM) was employed to characterize the microstructure of set-type yogurt gel networks, following the methodology described by Pang et al. [20]. Specifically, the gel samples were held at 10 °C for a 48-h period prior to sample preparation. The specimens were immobilized in glutaraldehyde solution at ambient temperature. Subsequently, fixed samples were dehydrated through a graded ethanol series (50%, 70%, 90%, and 100% v/v, each step for 15–20 min) to gradually replace water. To preserve the delicate gel structure and prevent collapse, they were dehydrated through a graded ethanol series and dried with a CO_2_ critical point drying apparatus (Tousimis Automatic, Rockville, MD, USA). Post-drying, the samples were sputter-coated with a thin platinum layer and viewed under a scanning electron microscope (JEOL 6610, Tokyo, Japan) with an accelerating voltage set at 10 kV.

2.7. Non-Targeted Metabolomics Analysis of Volatile Metabolites by HS-SPME-GC/MS

Volatile compounds underwent detection via headspace solid-phase microextraction coupled with gas chromatography–mass spectrometry (HS-SPME-GC/MS) [21,22]. Volatile metabolomics analysis was performed on each of the three biological replicates per treatment group. Specifically, 1 g of the sample was placed in a 20 mL headspace vial, and 10 μL of d_13_-hexanol solution (10 mg/L) was added thereto. The vial was then incubated at a constant temperature of 60 °C for 10 min with continuous stirring. Prior to extraction, the SPME fiber was pre-conditioned at 270 °C in the gas chromatography (GC) inlet for 10 min. Subsequently, the pre-treated extraction fiber was placed in the sample headspace at 60 °C for 15 min to promote the adsorption of volatile substances. The analysis was performed using an Agilent 8890A gas chromatograph coupled with a LECO Pegasus^®^ 4D mass spectrometer. Separation was achieved on a DB-WAX capillary column (30 m × 0.25 mm × 0.25 μm; Agilent Technologies, Santa Clara, CA, USA). The carrier gas was helium at a constant flow rate of 1.1 mL/min. The injector temperature was set at 250 °C in splitless mode. The oven temperature program was as follows: initial temperature 40 °C held for 2 min, then ramped at 8 °C/min to 120 °C, followed by a second ramp at 5 °C/min to 230 °C and held for 10 min. The mass spectrometer operated in electron ionization (EI) mode at 70 eV, with the ion source temperature set at 230 °C and the interface temperature at 200 °C. The mass scan range was 30–400 m/z.

Peaks were extracted and deconvolved using MassHunter Quantitative Analysis (Agilent Technologies) and matched against the NIST17 library with a similarity threshold > 75%. Results are expressed as μg equivalents of d_13_-hexan-1-ol per 10 mg of sample.

2.8. Analysis of Non-Volatile Components Using UHPLC-Q Exactive HF-X

Non-volatile metabolite isolation and identification were carried out using the method of Zhao et al. [23]. Non-volatile metabolomics analysis was conducted on each of the three independent biological replicates. Specifically, 50 mg of freeze-dried yogurt samples were accurately weighed, and their bioactive components were extracted with 400 μL of 80% (v/v) methanol solution containing 0.02 mg/mL L-2-chlorophenylalanine as the internal standard. The mixture was homogenized at a frequency of 50 Hz for 6 min, followed by ultrasonication in an ice-water bath for 30 min. After the above treatments, the samples were incubated at −20 °C for 30 min and then centrifuged at 12,000 rpm for 15 min at 4 °C. The resulting supernatant was gently transferred into sample vials, and metabolite profiling was conducted using an UHPLC-Q Exactive HF-X system (Thermo Fisher Scientific, Waltham, MA, USA). Throughout the testing process, all specimens were stored under 4 °C conditions.

2.9. Data Processing and Statistical Analyses

HS-SPME-GC/MS and LC-MS data were processed as described previously [23]. Statistical analyses were based on three independent biological replicates (n = 3). For volatile compounds, peaks were extracted and deconvolved using MassHunter Quantitative Analysis (Agilent Technologies) and matched against the NIST17 library with a similarity threshold > 75%. Results are expressed as µg equivalents of d_13_-hexan-1-ol per 10 mg of the sample. Raw LC-MS datasets were pre-processed using Progenesis QI software (version 3.0, Waters Corporation, Milford, MA, USA). A three-dimensional data matrix comprising sample information, peak retention time, and m/z values was generated. This matrix was subsequently imported into the Majorbio Cloud platform (https://cloud.majorbio.com (accessed on 1 January 2026)) for comprehensive analysis. The pre-processing pipeline included the following: (1) Feature filtering: Only metabolic features detected in at least 80% of samples within any experimental group were retained. Missing values were imputed with the minimum observed value for the respective feature. (2) Normalization: Data were normalized by the total sum of all features in each sample to correct for overall concentration differences. (3) Quality Control (QC): To ensure data reliability, metabolic features with a relative standard deviation (RSD) > 30% in the QC samples were removed, indicating poor instrumental reproducibility. (4) Transformation: The filtered and normalized data matrix was subjected to a base-10 logarithmic transformation to approximate a normal distribution and reduce heteroscedasticity for subsequent statistical analysis. Raw LC-MS datasets underwent pre-processing via Progenesis QI software, with further analyses conducted on the Majorbio Cloud platform. Metabolic signatures detected in no less than 80% of specimens within any group were retained, and absent values were filled using the lowest observed concentration. Datasets were normalized to their total sum, while signatures with a relative standard deviation (RSD) over 30% in quality control (QC) specimens were removed before log10-transformation. Statistical analysis was performed at two levels. First, multivariate statistical analysis was conducted using MetaboAnalyst 5.0. Principal Component Analysis (PCA) was initially applied to assess overall clustering and detect outliers. Subsequently, supervised Partial Least Squares Discriminant Analysis (PLS-DA) was employed to maximize the separation between pre-defined sample groups and to identify metabolite signatures contributing most to the differentiation. The robustness of the PLS-DA model was evaluated by the parameter R^2^Y (goodness of fit), and a permutation test (n = 200) was performed to rigorously validate the model against overfitting. Metabolites with a Variable Importance in Projection (VIP) score > 1.0 from the PLS-DA model were considered significant contributors to group separation. Second, univariate statistical analysis was carried out. The normalized data were imported into GraphPad Prism 9. After confirming the data distribution, appropriate parametric (e.g., Student’s t-test) or non-parametric tests were applied for inter-group comparisons. Metabolites were considered statistically significant only if they concurrently satisfied VIP > 1.0 and a p-value < 0.05 from the univariate test. And they were annotated using the KEGG database. Non-volatile metabolites were annotated through Progenesis QI, relying on exact mass (mass error ≤ 10 ppm) and MS/MS fragmentation patterns, by searching the HMDB, ChemSpider, and METLIN databases. Pathway assessments were performed via MetaboAnalyst, with a significance cutoff set at p < 0.05. Statistical tests and graph generation were carried out in GraphPad Prism 9. Final data are shown as mean ± SD, and variations were deemed statistically meaningful when p < 0.05.

3. Results

3.1. Changes in the Fermentation Characteristics of Fermented Milk

Titratable acidity and pH are key indicators of the fermentation progress in fermented milk. Supplementation with B. lactis IU100 and/or 1.5% RS3 shortened the time required to reach the fermentation endpoint (Figure 1A). Specifically, the control group (4.0) reached the endpoint (69.65 °T, pH 4.47) at 7 h, whereas the 4.0 + IU100, 4.0 + 1.5%RS3, and 4.0 + IU100 + 1.5%RS3 groups achieved the endpoint (70.83 °T, 71.72 °T, and 70.80 °T, respectively) at 6 h. During the early stages of microbial growth, no significant differences (p > 0.05) were observed in the viable counts of starter cultures among the groups (6.74–6.94 log_10_ CFU/mL) (Figure 1B,C). After 5 h, the viable counts in the 4.0 + IU100, 4.0 + 1.5%RS3, and 4.0 + IU100 + 1.5%RS3 groups were significantly higher than that in the control group (4.0), increasing from 8.79 log_10_ CFU/mL to 9.06, 8.99, and 8.96 log_10_ CFU/mL, respectively (p < 0.05) (Figure 1D). At the fermentation endpoint, no significant differences were observed in viable starter counts among the four groups (9.03–9.14 log_10_ CFU/mL). For Bifidobacterium species, viable counts did not differ significantly between the 4.0 + IU100 and 4.0 + IU100 + 1.5%RS3 groups at either the initial stage or after 5 h of fermentation (5.51–5.56 log_10_ CFU/mL and 6.90–6.98 log_10_ CFU/mL, respectively) (Figure 1E). Lactose, a key determinant of acid production rate in fermented milk, can be efficiently hydrolyzed by enzymatic systems of the starter cultures, such as β-galactosidase, thereby promoting rapid bacterial growth and shortening the fermentation time [24]. Bifidobacterium secretes extracellular amylases that partially degrade the granular structure of RS3, producing various saccharides such as oligosaccharides, maltose, and glucose [25]. These sugars are transported into the cell via ABC transporters and metabolized through the unique bifid shunt pathway. Catalyzed mainly by fructose-6-phosphate phosphoketolase, this pathway generates lactate and acetate as the main metabolites, provides energy for bacterial growth, and thereby accelerates fermentation [26].

3.2. Textural Properties and Water Holding Capacity of Fermented Milks

The hardness and springiness of the four fermented milk samples were evaluated at the fermentation endpoint. The control group (4.0) had a hardness of 6.90 g, while the 4.0 + IU100 and the 4.0 + 1.5% RS3 groups had hardness values of 7.08 g and 6.41 g, respectively (Figure 2A). Notably, the 4.0 + IU100 + 1.5% RS3 group exhibited a significantly higher hardness of 8.88 g (p < 0.05). Springiness values for the 4.0, 4.0 + IU100, and 4.0 + 1.5%RS3 groups were 0.85, 0.97, and 1.14 mm, respectively, with no significant differences observed (p > 0.05) (Figure 2B). In contrast, the 4.0 + IU100 + 1.5%RS3 group showed a significantly higher springiness of 2.33 mm (p < 0.05). The water holding capacity (WHC) of the control group (4.0) was 59.08% (Figure 2C). The 4.0 + IU100 group had a WHC of 68.84%, which did not exhibit obvious discrepancy compared with the control’s reference value (p > 0.05). In comparison, the 4.0 + 1.5%RS3 and 4.0 + IU100 + 1.5%RS3 groups exhibited WHC values of 71.26% and 71.53%, respectively, both significantly higher than that of the control (p < 0.05). Compared with the control group (4.0), the addition of B. lactis IU100 alone (4.0 + IU100) resulted in a slight increase in hardness (from 6.90 g to 7.08 g) and springiness (from 0.85 mm to 0.97 mm), although these changes were not statistically significant. In contrast, the addition of RS3 alone (4.0 + 1.5% RS3) significantly improved the water holding capacity (WHC) from 59.08% to 71.26%, owing to its strong water-binding and gel-filling properties. However, hardness decreased slightly to 6.41 g, which may be attributed to the disruption of protein network continuity by RS3 in the absence of sufficient microbial interaction. Most notably, the co-supplementation group (4.0 + IU100 + 1.5% RS3) exhibited a pronounced synergistic effect: hardness increased significantly to 8.88 g, springiness markedly improved to 2.33 mm, and WHC reached 71.53%.

The synergistic interaction between probiotics and prebiotics can markedly enhance the textural properties of fermented milk. In general, the texture of fermented milk is governed by the physical interactions among casein micelles. A significant increase in hardness is attributed to the strengthened binding capacity of casein macromolecules and the improved gelation ability of the system [27]. For example, Costa et al. [28] reported that adding 5% (w/v) inulin to goat milk yogurt containing Lactobacillus acidophilus LA-5 significantly increased hardness compared to a prebiotic-free control (p < 0.05). This effect is likely due to inulin acting as a soluble fiber and water-structuring agent, forming complexes with casein aggregates and integrating into the three-dimensional gel network developed during fermentation. Furthermore, the molecular properties of prebiotics enhance interactions among casein molecules, minimize structural loosening caused by whey separation during storage, and promote the formation of a more stable gel network with increased hardness. The enhancement in springiness may be attributed to the ability of prebiotics to optimize gel network integrity by improving water distribution and fat dispersion and the ability of filling gel pores to strengthen network toughness. Such a composite network can better resist external stress and recover from deformation, resulting in improved springiness [29]. Wang et al. [30] reported that supplementing fermented milk containing Bifidobacterium Bb-12 with 1.0% potato resistant starch significantly increased the WHC (p < 0.05). Potato resistant starch can efficiently bind water through its hydrophilic groups, reducing the amount of free water without competing with proteins for hydration or disrupting the gel matrix.

3.3. Microstructural Analysis of Fermented Milk by Scanning Electron Microscopy

The microstructure of the control group (4.0) exhibited a loose, granular protein network composed of irregularly shaped and unevenly dispersed aggregates (Figure 2D). The voids between aggregates were large, heterogeneous in size, and unevenly distributed, with some regions forming cavity-like pores. This discontinuous structure contributed to poor compactness and reduced overall stability. In the 4.0 + IU100 group, the protein network exhibited enhanced cohesiveness, with localized clusters and improved interparticle connectivity (Figure 2E). The 4.0 + 1.5% RS3 group exhibited a denser, more agglomerated protein structure (Figure 2F). The interparticle voids were largely filled, resulting in a substantial reduction in pore number and size. The protein matrix appeared more homogeneous, forming a relatively continuous and uniform gel network. In the 4.0 + IU100 + 1.5% RS3 group, a dense, continuous composite gel network incorporating resistant starch was observed (Figure 2G). The matrix surface was smooth and uniform, with no visible loose or particulate regions. Overall, this group exhibited the highest structural homogeneity and compactness among all four samples.

The pronounced microstructural improvement observed with the addition of RS3 highlights its key role in regulating the formation of the milk protein gel network. This research outcome corresponds with earlier documented studies regarding the regulatory effects of prebiotics on acid-induced milk gels [31]. In yogurt without RS3, the protein matrix typically exhibits a loose, discontinuous structure with irregularly dispersed protein particles, large and unevenly distributed pores, and poor network continuity—a characteristic of unstable acid gels lacking effective stabilizers, resulting from insufficient casein–casein interactions. In contrast, RS3 incorporation markedly modifies the microstructure: protein particles form dense aggregates, pore number and size are substantially reduced, and a more continuous and uniform gel network is established. This improvement can be attributed to two main mechanisms: (1) RS3 acts as a prebiotic that interacts with milk proteins, for example, through hydrogen bonding between RS chains and casein molecules, thereby strengthening cross-links between protein particles and reducing loose interstitial spaces [32,33], and (2) RS3 serves as a physical filler, occupying voids between protein aggregates, which densifies the gel matrix and enhances structural homogeneity. These microstructural modifications underlie the observed improvements in textural properties, such as firmness, and functional performance, including WHC, as a continuous and dense gel network more effectively traps water and resists external deformation [33,34]. This stabilizing effect of RS3 as a filler and structure promoter aligns with findings in acid-induced casein gel models, where modified starches like acetylated distarch phosphate were shown to integrate into the protein network, reduce porosity, and enhance gel homogeneity and strength at optimal concentrations [35].

3.4. GC-MS Analysis of Volatile Metabolites

Volatile metabolites in the four fermented milk samples at the fermentation endpoint were analyzed using HS-SPME-GC/MS. A total of 115 volatile metabolites were identified and classified into eight groups: 30 ketones, 29 acids, 20 alcohols, 14 aldehydes, 6 esters, 7 heterocyclic compounds, 3 furans, and 3 aromatic hydrocarbons. Compared with the control group (4.0), the relative content of acids in the 4.0 + IU100 group increased significantly from 5409.85 μg/L to 5894.50 μg/L (p < 0.05) (Figure 3A). In contrast, the relative contents of furans and aromatic hydrocarbons decreased significantly, from 357.67 μg/L to 161.19 μg/L and from 319.89 μg/L to 56.80 μg/L, respectively (p < 0.05) (Figure 3A). In the 4.0 + 1.5% RS3 group, ester content increased by 27.82%, and furan content increased significantly by 28.46% (p < 0.05) (Figure 3A). In the 4.0 + IU100 + 1.5% RS3 group, acid content increased significantly from 5409.85 μg/L to 6819.59 μg/L, while ester content also increased by 25.08% (p < 0.05) (Figure 3A). In contrast, the relative content of aromatic hydrocarbons decreased significantly in this group (p < 0.05).

Principal component analysis (PCA) was performed for the relative contents of the 115 annotated volatile compounds to explore metabolic differences among the four sample groups (Figure 3B). At the fermentation endpoint, the initial two principal components, namely PC1 and PC2, accounted for 24.2% and 13.6% of the overall variance, correspondingly. Along PC1, the control group (4.0) was clearly separated from the 4.0 + 1.5% RS3 and 4.0 + IU100 + 1.5% RS3 groups, thereby indicating distinct volatile profiles. In contrast, the control (4.0) and 4.0 + IU100 groups exhibited partial overlap in the PCA score plot, suggesting a degree of similarity in their volatile metabolite composition.

3.5. Odor Activity Value (OAV) Analysis of Key Volatile Compounds

The overall aroma of fermented milk is determined by both the relative abundances and odor thresholds of volatile organic compounds (VOCs). Odor activity value (OAV) analysis was performed for key VOCs detected in the four fermented milk groups. Differential metabolites with an OAV > 1 between groups were identified and visualized in volcano plots using the criteria of variable importance in projection (VIP ≥ 1.0 and p < 0.05). In the 4.0 + IU100 group, three VOCs were significantly upregulated: ethyl caprylate, ethyl n-butyrate, and 1-octanol. In contrast, three compounds were downregulated: 2,3-dihydrofuran; 2,7-nonadien-5-one, 4,6-dimethyl-; and lauryl aldehyde (Figure 4A). In the 4.0 + 1.5% RS3 group, six compounds were upregulated, ethyl n-butyrate, ethyl caprylate, 1-octanol, 1-nonanol, 2,3-heptanedione, and 2-xylene, while four were downregulated, γ-dodecalactone, lauryl aldehyde, 4-ethylbenzaldehyde, and styrene (Figure 4B). In the 4.0 + IU100 + 1.5% RS3 group, the four upregulated compounds were ethyl caprylate, 1-octanol, ethyl n-butyrate, and 2,3-heptanedione, whereas the four downregulated compounds were 2,2,4-trimethyl-1,3-pentanediol diisobutyrate, 4-ethylbenzaldehyde, 2,7-nonadien-5-one, 4,6-dimethyl-, and lauryl aldehyde (Figure 4C).

The addition of probiotics and prebiotics to fermented milk enhances its flavor profile, primarily by facilitating the generation of critical volatile substances including esters, alcohols, and ketones. Esters, produced through the esterification of free fatty acids and alcohols, are important aroma-active components in foods. They typically impart floral–fruity aromatic traits to fresh milk-derived products, thereby helping to mask the pungent odors of free fatty acids. For example, Dan et al. [36] inoculated 17 strains of Lactobacillus bulgaricus into a fermented milk system and identified eight characteristic esters, including formic acid ethenyl ester, through volatile compound analysis. Notably, the OAV of formic acid ethenyl ester reached 12.42, imparting a distinct fruity note and enhancing the overall flavor of the product. Walsh et al. [37] demonstrated that incorporating Lactobacillus kefiranofaciens into kefir boosted the synthesis of ester compounds derived from lipids, including ethyl hexanoate and ethyl octanoate, that impart prominent fruity and creamy aromatic notes. Alcohols, such as 1-hexanol, impart herbal, grassy, and subtle fruity notes to fermented milk [38]. This study confirms that partially substituting camel milk with chickpea milk as a prebiotic antioxidant significantly enhances the fermented milk’s antioxidant activity and improves its sensory profile, primarily attributed to its promotion of probiotic growth and increased phenolic and flavonoid compounds [39]. López-Salas et al. [40] observed that the concentration of 1-hexanol increased from 2.10 μg·mL^−1^ to 6.01 μg·mL^−1^ during the fermentation of Habanero pepper puree inoculated with Lactiplantibacillus plantarum, introducing a distinct herbal aroma. Similarly, Yang et al. [41] reported that alcohols (333.09 ± 19.27 μg/L) were the predominant volatile compounds in apple juice fermented with Limosilactobacillus fermentum CICC 21828 and supplemented with galactooligosaccharides, contributing fruity and floral aroma notes. Ketones such as 2,3-heptanedione, 2,3-pentanedione, and 2,3-butanedione are known to impart buttery and fresh dairy aromas [42]. Papaioannou et al. [43] reported that goat milk yogurt supplemented with Lactobacillus acidophilus LA-5 contained higher levels of 2,3-pentanedione (5.98 ± 0.58 mg/kg) compared to the control (4.30 ± 0.10 mg/kg), resulting in a more pronounced creamy and buttery flavor. In addition, the supplementation of probiotics and prebiotics can help mitigate undesirable flavors. Fischer et al. [44] demonstrated that adding L. acidophilus and 1.5% inulin to yogurt significantly reduced hexanal content from 1.33 mg/kg to 0.42 mg/kg, effectively decreasing the “beany” off flavor associated with excessive hexanal [45]. Similarly, carob powder as a prebiotic has been shown to enhance probiotic viability and may consequently support flavor metabolite production in food matrices [46]. Carob powder as a prebiotic has been shown to enhance probiotic viability and may consequently support flavor metabolite production in food matrices [47].

3.6. Untargeted Metabolomics Profiling

Untargeted metabolomics analysis using LC-MS/MS was employed to profile the non-volatile metabolites in fermented milk. At the fermentation endpoint, a total of 1707 metabolites were detected, with the most abundant classes being lipids and lipid-like molecules (399, 23.46%), organic acids and derivatives (385, 22.63%), organic oxygen compounds (237, 13.93%), organoheterocyclic compounds (228, 13.40%), and benzenoids (139, 8.17%) (Figure 5A). Partial least squares discriminant analysis (PLS-DA) was implemented to clarify variances in non-volatile flavor compound composition among the treatment groups (4.0, 4.0 + IU100, 4.0 + 1.5% RS3, and 4.0 + IU100 + 1.5% RS3) at the initial fermentation stage. The first two components explained 30.1% (Component 1) and 17.7% (Component 2) of the variance, with a cumulative variance of 47.8%, indicating that the model effectively captured inter-group differences (Figure 5B). The model exhibited an R^2^ value of 0.9446 in the permutation test, confirming statistically significant discrimination among groups and supporting the reproducibility and reliability of the data (Figure 5C). In the PLS-DA score plot, the four groups exhibited distinct clustering patterns: the 4.0 and 4.0 + IU100 groups clustered in the right quadrants, the 4.0 + 1.5% RS3 group was distributed in the upper left quadrant, and the 4.0 + IU100 + 1.5% RS3 group clustered in the lower left quadrant. These results visually indicate that significant differences in non-volatile metabolite profiles were already present at the early stage of fermentation. Such differences likely reflect the influence of added functional components (e.g., probiotics, resistant starch) and/or the specific microbial combinations in each treatment, providing a clear foundation for further investigation into the regulatory mechanisms underlying flavor formation in fermented milk.

To investigate the specific effects of Bifidobacterium and RS3 on the metabolome of fermented milk, metabolites were filtered using the criteria of VIP ≥ 1.0 and p < 0.05. In the 4.0 + IU100 group, 135 differential metabolites were identified, including 44 upregulated compounds (e.g., L-asparagine, 4-pyridoxolactone, L-acetylcarnitine) and 91 downregulated compounds (Figure 6A,B). In the 4.0 + 1.5% RS3 group, 281 differential metabolites were detected, of which 80 were significantly increased (e.g., NAD, cyclic AMP, phloretic acid) and 201 were decreased (Figure 6C,D). The 4.0 + IU100 + 1.5% RS3 group exhibited the most pronounced changes, with 373 differential metabolites identified, including 98 upregulated and 275 downregulated compounds (Figure 6E,F). Representative upregulated metabolites included NAD, cyclic AMP, and 4-pyridoxolactone. These findings indicate that the combined supplementation of B. animalis IU100 and RS3 exerts a significant influence on the non-volatile metabolome of fermented milk.

3.7. KEGG Pathway Enrichment Analysis

To investigate the metabolic regulatory mechanisms underlying these changes, an enrichment analysis of KEGG pathways was carried out. In the 4.0 + IU100 group, 24 differential metabolites were mapped to 12 metabolic pathways, including alanine, aspartate, glutamate metabolism, nucleotide metabolism, purine metabolism, and vitamin B6 metabolism (Figure 7A). In the 4.0 + 1.5% RS3 group, 26 distinct metabolites showed enrichment in 12 metabolic pathways, including purine metabolism, nucleotide metabolism, and aldosterone synthesis and secretion (Figure 7B). For the 4.0 + IU100 + 1.5% RS3 group, 24 metabolites were enriched across 12 pathways, including purine metabolism, alcoholism, and aldosterone synthesis and secretion (Figure 7C). The biosynthesis of ethyl caprylate and ethyl n-butyrate is closely associated with adenosine diphosphate ribose from purine metabolism, NAD^+^ metabolism within the aldosterone synthesis and secretion pathway, and L-asparagine from alanine, aspartate, and glutamate metabolism (Figure 7D). These esters are generated from α-keto acids and fatty acid metabolites as key precursors under the coordinated regulation of these metabolic pathways. Specifically, α-keto acids are decarboxylated by decarboxylases to form acetaldehyde, which is then reduced to ethanol by NADH-dependent alcohol dehydrogenases. Finally, ethanol and acyl-CoA undergo esterification catalyzed by ester synthases, resulting in the formation of ethyl caprylate and ethyl n-butyrate (Figure 7D). Purine metabolism and aldosterone synthesis and secretion serve as key pathways supporting ester biosynthesis. NAD^+^ produced via purine metabolism enhances NADPH-dependent reductive reactions, ensuring an adequate supply of precursors for the formation of ethyl caprylate and ethyl n-butyrate [48]. Within the aldosterone synthesis and secretion pathway, NAD^+^ functions as an essential critical coenzyme that drives the production of the precursor octanoate and modulates alcohol dehydrogenase activity [49,50]. Acetyl-CoA acts as a central metabolic node for alcohol precursor formation, with fatty acids derived from acetyl-CoA—supplied through L-asparagine metabolism—being subsequently reduced to their corresponding alcohols [51].

The volatile alcohols 1-octanol and 1-nonanol are closely associated with L-asparagine and NAD^+^ within the alanine, aspartate, and glutamate metabolism and aldosterone synthesis and secretion pathways (Figure 7E). The formation of the volatile aldehyde 4-ethylbenzaldehyde is related to glutamic acid within the biosynthesis of cofactors pathway. 4-Ethylbenzaldehyde is produced from the aromatic fatty acid p-ethylphenylacetic acid through decarboxylation and oxidation reactions catalyzed by the aromatic compound metabolic enzyme system of lactic acid bacteria (Figure 7E). Additionally, the synthesis of 2,3-dihydrofuran relies on linoleic acid as the core precursor and is regulated in part by the vitamin B6 metabolism pathway. Linoleic acid serves as the direct substrate for the non-enzymatic oxidative formation of 2,3-dihydrofuran, whereas p-ethylphenylacetic acid is converted into 4-ethylbenzaldehyde through microbial enzymatic decarboxylation and oxidation [51]. Notably, key lipid-oxidizing bacteria, such as Lactococcus and Pediococcus, exhibit a strong positive correlation with glutamate, the predominant free amino acid contributing to flavor. This relationship highlights the functional interplay of these microorganisms, which simultaneously drive lipid oxidation (producing 4-ethylbenzaldehyde) and proteolysis (generating glutamate), thereby linking lipid- and protein-derived flavor formation pathways [52]. It is worth noting that the integrated metabolomics and microbial metabolic pathway analysis approach adopted in this study is not only applicable for elucidating flavor formation mechanisms in fermented milk systems but also provides an analytical framework for investigating microbe–diet component interactions in other complex fermented food systems, such as cheese, soybean products, and fermented cereals. This methodology contributes to systematically revealing the regulatory patterns of metabolic networks in different matrices, offering methodological support for the flavor design and functional optimization of diversified fermented foods.

4. Conclusions

The combined supplementation of B. lactis IU100 and 1.5% RS3 synergistically enhanced the overall quality of fermented milk. This combination promoted starter culture growth, accelerated acidification, and shortened fermentation time. It also facilitated the formation of a denser, more homogeneous, and stable protein gel network, resulting in improved textural properties. Moreover, the supplementation markedly influenced both volatile and non-volatile metabolite profiles. Volatile analysis revealed increased levels of acids and esters, and OAV analysis confirmed the enrichment of key aroma-active compounds—such as ethyl caprylate, ethyl n-butyrate, 1-octanol, and 2,3-heptanedione—contributing fruity and creamy notes to the final product. Untargeted metabolomics combined with KEGG pathway analysis revealed that the observed flavor enhancements were driven by alterations in core metabolic pathways, including purine metabolism, alanine/aspartate/glutamate metabolism, and aldosterone synthesis and secretion. Changes in key metabolites such as L-asparagine and NAD^+^—which act as precursors or cofactors—facilitate coordinated enzymatic reactions that convert α-keto acids and fatty acid derivatives into flavor esters (e.g., ethyl caprylate and ethyl n-butyrate) and volatile alcohols (e.g., 1-octanol), providing a mechanistic basis for the improved aroma profile.