Microbial diversity and its links to retinol pathways and aroma compounds in ethnic fermented rice beverages of Assam

Abstract

Traditional fermented rice beverages are produced through complex microbial fermentation processes that influence their physicochemical characteristics and metabolite composition. In this study, metagenomic sequencing and GC–MS/MS-based metabolomics were integrated to characterize four indigenous rice beverages: Black Rohi Modh (BR), Rohi Modh (RH), Jou Bidwi (JOU), and Sai Mod (SM). All beverages were mildly acidic, with pH values ranging from 4.1 to 4.5 and titratable acidity between 0.58 and 0.72% lactic acid. Ethanol content varied among samples, with BR showing the highest concentration (8.13% v/v), followed by JOU and RH (approximately 5.5% v/v), while SM exhibited the lowest level (4.28% v/v). Antioxidant activity differed across beverages, with RH and BR demonstrating higher DPPH radical scavenging activity and SM showing the highest ferric reducing antioxidant power (96.93 µmol/mL). Metagenomic analysis generated 57.69 Mb of assembled sequences, identifying 48 microbial phyla and 1,785 species, with Eukarya accounting for 66.12% of the total community. Ascomycota predominated in BR and JOU, whereas Bacillota was more abundant in RH. The genus Saccharomyces was consistently dominant across samples. Functional annotation indicated enrichment in metabolic pathways related to carbohydrate and amino acid metabolism, as well as genes associated with ethanol biosynthesis and retinol metabolism pathways, reflecting microbial metabolic potential rather than direct vitamin production. Metabolomic profiling identified 113–167 metabolites per beverage, with 93 compounds shared among all samples. Correlation analysis revealed significant associations between Saccharomyces cerevisiae and short-chain fatty acids (ρ = 0.62–0.71, FDR < 0.05), indicating a strong positive relationship between microbial abundance and metabolite production.

Graphical abstract

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s13568-026-02062-0.

Introduction

The ethnic tribes of southern Assam in Northeast India (NEI) have long depended on their traditional fermented foods made from rice (Das et al. 2016). Traditional fermented foods and drinks in Northeast India are unique to each tribe, shaped by local environments and preparation methods. These foods and drinks are culturally significant and widely consumed, and previous studies have reported their nutritional and microbial diversity. However, the specific mechanisms linking microbial composition to metabolite formation and functional characteristics in these beverages remain insufficiently understood (Tamang et al. 2012). Despite being popular beverage of the region with health benefits, the microbial composition and functional mechanisms have not been addressed sufficiently. In particular, the association between microbial community structure and metabolite production in traditional fermented rice beverages have not been systematically characterized using integrated multi-omics approaches. Understanding these microbe–metabolite relationships is essential for explaining fermentation-driven biochemical changes and for providing scientific evidence supporting the functional and technological properties of these beverages. Rice-based fermented beverages are integral to the cultural identity of southern Assam’s ethnic groups. Each tribe uses distinct starter cultures like humao, chol, thap, ndhuihi, and dawai to prepare traditional drinks such as judima, zu, hor alank, zao, and haria with specific microbial consortia (Yumnam et al. 2024). The microbial consortia of starter culture finally affect the fermentation dynamics and metabolite production in these ethnic fermented beverages.

These beverages are made by fermenting cooked glutinous rice with community-specific starter cultures in earthen pots for 4–7 days. While the core process is similar, variations in local ingredients reflect cultural diversity and continue to attract academic interest (Yumnam et al. 2024). Jou-bidwi, a traditional Boro community’s fermented rice beverage, is made by fermenting glutinous rice with the amao starter, initially for 18–48 h. It is then aged in sterilized earthen pots for 3–7 days until sweet-smelling froth appears, indicating readiness (Basumatary and Gogoi 2014). Jou-bidwi have high nutritional content (Basumatary and Gogoi 2014). Made from glutinous rice and a unique herbal starter culture called epob. sai mod, also called poro apong, is a traditional rice beverage of the Mising tribe (Yein et al. 2022). Epob, made from rice flour and medicinal herbs, forms the base of sai mod, a beverage traditionally consumed during festivals and valued for its sensory characteristics and cultural importance (Yein et al. 2022). Bora chaul, or glutinous rice, and a starter culture known as aaroi are used to make rohi modh, a well-liked fermented rice beverage among the Kachari community. A mildly alcoholic beverage with a long history of use is produced by the fermentation process, which normally takes four days (Sonowal 2023). Both beverages demonstrate the native fermentation skills and highlight the rice beverage's ceremonial and ethnomedical significance among Northeast India's many tribal communities.

The aim of this study was to comprehensively investigate the microbial diversity, functional metabolic potential, and metabolite composition of the traditional fermented rice beverages rohi modh, sai mod, and jou-bidwi using an integrated culture-dependent, metagenomic, and GC–MS/MS-based metabolomic approach. Specifically, the study sought to isolate and characterize lactic acid bacteria and yeasts, elucidate microbial community structure through high-throughput sequencing, predict key metabolic pathways associated with fermentation and nutrition, and profile bioactive and aroma-related metabolites. By linking dominant microbial taxa with functional genes and metabolite signatures, this study aims to provide mechanistic insights into microbial–metabolite interactions in traditional fermented rice beverages and to establish a scientific foundation for future functional and technological investigations. To our knowledge, this study represents one of the first integrated metagenomic and metabolomic analyses to investigate microbial–metabolite associations in traditional fermented rice beverages from Northeast India.

Methodology

Documentation and sampling

Twelve samples representing four traditional fermented alcoholic rice beverages jou bidwi (JOU), sai mod (SM), black rohi modh (BR), and rohi modh (RH) were collected from different regions of Assam, India. All analyses were performed in triplicate to ensure analytical reliability; however, these replicates represent technical replicates derived from single authentic household batches rather than independent biological replicates across multiple production events. Samples were stored in sterile 500 mL containers, transported on ice, and preserved at − 20 °C until further processing. BR and RH, prepared from black and common rice, were sourced from Pukia (Dhemaji district), JOU from Kaljar (Baksa district), and SM from Jamuuri Panchali (Dhemaji district), where its preparation is now rare. Each beverage could only be authentically collected once due to the declining practice of traditional fermentation, underscoring the urgency for documentation. Because the availability of genuine household-fermented batches was extremely limited, the present sampling represents a case-study–type dataset rather than a comprehensive regional survey, and the findings should be interpreted as characteristic snapshots of the available traditional practices rather than population-level generalizations. These beverages are typically homemade by indigenous communities using local rice varieties and starter cultures, fermented in earthen or plastic containers for 3–10 days at room temperature. Collection sites are shown in Fig. 1 and Supplementary File: ST 1.

Fig. 1.

Sample place if collection of rohi modh (RH), black rohi modh (BR), sai mod (SM) and jou bidwi (JOU) fermented rice beer from different collection of Assam with preparation of sample

pH, titratable acidity and proximate analysis of samples

The pH was measured by use of a digital pH meter (Type 361, Systronics, India) based on the (AOAC 1990). In a nutshell, 10 g each of fermented rice beverage samples was combined with 90 mL distilled water and mixed thoroughly. The homogenized sample was left to stand 10 min, and the pH electrode was inserted in the suspension. The stabilized reading was noted to be the pH. Briefly, 10 g of each sample was homogenized with 90 mL of distilled water and filtered through Whatman No. 1 filter paper. An aliquot of 10 mL of the filtrate was taken in a conical flask, and 2–3 drops of phenolphthalein indicator were added. The sample was titrated with 0.1 N NaOH until a persistent pale pink color appeared. The total titratable acidity was calculated and expressed as percentage of lactic acid (Darias-Martín et al. 2003).

In protein estimation, 1 mL of the properly diluted sample extract was combined with alkaline copper reagent then Folin-Ciocalteu reagent was added, and the absorbance was recorded at 750 nm in the presence of bovine serum albumin as the standard (Lowry et al. 1951). The total carbohydrate content was calculated as 1 mL of sample extract, 1 mL of 5% phenol solution, and 5 mL of concentrated sulfuric acid and the absorbance on the 490 nm was measured with glucose as the standard (Dubois et al. 1951). The amount of starch was determined through the following steps: 1 g of a sample was extracted using the perchloric acid after which the extract was reacted with anthrone reagent and the absorbance at 630 nm measured (Hansen and Moller, 1975). Crude fat was calculated by weighing 5 g of dried sample with petroleum ether in a Soxhlet apparatus and evaporating the solvent and extracting the fat (AOAC 1990). The content of phytosterols was calculated taking a certain volume of sample extract, teaching Folin-Ciocalteu reagent, and then adding sodium carbonate solution and measuring the absorbance at the given wavelength, as proposed by Molole et al. (2022).

Estimation of ethanol and methanol content of sample using HPLC

Ethanol and methanol in traditional alcoholic rice beverages were quantified using a Hitachi HPLC system with photodiode array and fluorescence detectors, controlled via Agilent EZChrom Elite 3.2.0. Separation was performed on a C18 column (150 × 2.1 mm, 1.7 µm) using an isocratic mobile phase of water: acetonitrile: methanol (24:26:50, v/v) at 1.0 mL/min, with detection at 210 nm (UV) and fluorescence excitation/emission at 265/345 nm. Samples (100 μL) were derivatized with Fmoc-Cl and phosphate buffer (pH 8.2), incubated at 40 °C for 40 min, and 20 μL was injected for analysis. A 7-point calibration curve ensured accurate quantification, with precautions taken due to the volatility of ethanol and methanol. This method enabled reliable measurement of alcohol content in the rice beverage samples.

Antioxidant assay of the sample using spectrophotometric assay

The antioxidant potential was evaluated through two in vitro methods, which though straightforward, represent slightly different aspects of redox behaviour, i.e. DPPH radical scavenging activity and ferric reducing power, according to the methods of (Zhao et al. 2021) and Tang et al. (2020). The DPPH test depends on the visually impressive purple colour of 2, 2-diphenyl-1-picrylhydrazyl. Sample extracts were incubated with freshly prepared DPPH solution in the dark to see any hydrogen- or electron-donating compounds present in the sample react. The absorbance was then noted at the 517 nm. The colour intensity reduced with time as the violet solution faded to yellow and this indicated the presence of radical scavenging activity. The outcomes were determined as a percentage of inhibition compared to a control. Power reduction was tested by the use of ferric ion reduction assay. Concisely, the sample was mixed with potassium ferricyanide and phosphate buffer and allowed to incubate at a predetermined temperature to allow reduction of the Fe³⁺ to Fe²⁺. The reaction was stopped by adding the trichloroacetic acid and the mixture centrifuged to obtain the clear supernatant. Ferric chloride was added to this supernatant, and it was then measured at 700 nm. An increase in absorbance was a sign of increased reducing capacity, which is essentially the electron-giving potential of the constituents of the sample.

Isolation of lactic acid bacteria and yeast

One millilitre/gram of the sample was serially diluted in 9 ml physiological saline (0.85% w/v NaCl) to isolate lactic acid bacteria (LAB) and yeast. The dilutes samples were plated into the MRS agar with calcium carbonate for the isolation of LAB as described in Tamang and Tamang (Tamang and Tamang 2009). While the yeasts was isolated using yeast malt (YM) gar (HiMedia) containing 34 mg/mL chloramphenicol (HiMedia). Colonies were randomly selected, re-streaked on fresh respective media plates, and examined under a phase-contrast microscope for purity. Verified isolates were preserved at –80 °C in 20% glycerol for further analysis (R. Das et al. 2024).

Phenotypic and genotypic characterization

Staining and cell morphology

Gram staining was conducted following standard procedures by Tamang et al, (2007). Cell morphology, catalase test and production of gas in presence of glucose were tested for lactic acid bacteria (Schillinger and Lucke 1987). Yeast morphology was examined using the Lactophenol Cotton Blue (LPCB) staining method (Leek, 1999).

DNA extraction, PCR amplification and identification of selected strains

DNA from LAB isolates were extracted using a modified enzymatic-heating lysis method (Jeyaram et al. 2010), where pure colonies were cultured in MRS broth at 30 °C for 16–18 h, followed by centrifugation and successive washes with 0.5 M NaCl and sterile MilliQ water. The cell pellet was treated with TE buffer (pH 8) and lysozyme (2 mg/mL) at 37 °C for 30 min, followed by heating at 98 °C for 15 min. DNA was isolated by centrifugation and quantified using an Eppendorf BioSpectrometer. Isolates were identified by Sanger sequencing of the 16S rRNA gene using GoTaq® Green Master Mix (Promega, USA) with primers 27F (5′-AGAGTTTGATCATGGCTCAG-3′) and (Heather and Chain 2016) 1492R (5′-GTTACCTTGTTACGACTT-3′) (Weisburg et al. 1991) under standard thermal cycling conditions.

Genomic DNA was extracted from yeast isolates using the method of Renshaw et al, (2015). A 2 mL yeast culture was centrifuged, washed with 0.5 M NaCl, and lysed in buffer containing Tris–HCl, NaCl, EDTA, and SDS. RNase A and Proteinase K treatments were performed at 65 °C. After phenol: chloroform: isoamyl extraction and isopropanol precipitation, DNA was washed with ethanol, dried, and dissolved in nuclease-free water. DNA purity (1.8–2.2) was confirmed before PCR (Renshaw et al. 2015). Using primers 7F (5′-AGAGTTTGATCCTGGCTCAG-3′) and 1492R (5′-TACGGYTACCTTGTTACGACTT-3′) for bacteria D1 (5′-GCA TAT CAA TAA GCG GAG GA-3′) and D2 (5′-TTG GTC CGT GTT TCA AGA CG-3′), the D1/D2 domains of the large ribosomal subunit were amplified in order to identify the yeast isolates. NL1/NL4 primers, 10–20 ng of DNA, and GoTaq® Green Master Mix were used in a 50 μL reaction for PCR. The thermal profile consisted of a one-minute initial denaturation at 94 °C, 35 cycles of 94 °C for 30 s, 58 °C for 30 s, 72 °C for 30 s, and a final extension at 72 °C for 5 min. Amplicons were visualised with a Gel DocTM EZ system and verified by 1% agarose gel electrophoresis (Hinrikson et al. 2005).

PCR amplicons were purified with a PEG-NaCl method (Schmitz and Riesner 2006), washed with cold ethanol, air-dried, and resuspended in nuclease-free water. Sequencing was conducted using ABI 3730XL (Applied Biosystems, USA), with PCR conditions including 35 cycles and annealing at 40 °C. Raw sequence quality was assessed using Sequence Scanner v2.0 (Nurminsky and Hartl 1996), assembled in ChromasPro v1.34 (http://technelysium.com.au/wp/chromas/), and screened for chimeras using Mallard (Ashelford et al. 2006). Sequence identity was confirmed by BLAST (Altschup et al. 1990) and EzTaxon. Multiple sequence alignment was performed using ClustalW (Thompson et al. 1994), and phylogenetic analysis was conducted using the neighbor-joining method (Saitou and Nei 1987) in MEGA11 (Tamura et al. 2021) and was visualized and circular tree representation was done in iTOL v5.0 (Letunic and Bork 2021).

Shotgun metagenomics

Metagenomic DNA extraction, library preparation and sequencing

Metagenomic DNA was extracted from pellets of centrifuged (10,000 × g, 10 min) and washed rice beverage samples using the Nucleospin Food kit (Macherey–Nagel). DNA concentration and quality were assessed with a Qubit 4 Fluorometer and agarose gel electrophoresis (S. Das and Tamang 2023). Libraries were prepared using the MinION SQK-LSK109 ligation kit (ONT) as per Sevim et al, (2019). DNA (1000 ng) was sheared (> 10 kb) with Covaris g-TUBEs, repaired (NEBNext FFPE kit), ligated with adaptors, and purified using AMPure XP beads. Sequencing was performed on a MinION Mk1C with FLO-MIN106D R9 chemistry. MinKNOW v1.0.5 and Guppy v2.2.3 were used for device control and base calling.

Taxonomy analysis

Raw reads from the Nanopore platform were obtained in fast5 format and converted to fastq using Poretools v0.6.0 (Loman and Quinlan 2014). Read quality was assessed with NanoPlot v1.33.0 (De Coster et al. 2018), and assembly was performed using Canu v2.2.0 (Koren et al. 2017). The assembled contigs were quality checked using QUAST (Gurevich et al. 2013) and were annotated taxonomically using the Kaiju pipeline v1.9.0 (Menzel et al. 2016), against the NCBI nr + euk database. Parameters included a minimum match length of 11, match score of 80, up to 5 mismatches, and e-value of 0.01, with the SEG filter enabled. The greedy algorithm was used for taxonomic assignment (Zhang et al. 2000), translating reads into amino acid sequences and scoring fragments with BLOSUM62 (Lazar et al. 2019). Searches employed the Burrows-Wheeler algorithm via backward search. The same Kaiju protocol was used for analyzing pulque metagenomic data from the NCBI database.

Predictive functional analysis

Prodigal was used to predict the gene for protein-coding genes (ORFs) with threshold of ≥ 150 bp, standard bacterial genetic code (11) and annotation thresholds of E-value ≤ 1e-5, identity ≥ 30%, and query coverage ≥ 50% (Hyatt et al. 2010). By default, ORFs with fewer than 180 nucleotides were not included in additional analysis. The ARAGORN program (Laslett and Canback 2004) was used to predict tRNA genes, and rRNAFinder (Eremin et al. 2025) was used to identify and categorise ribosomal RNA genes (5S, 5.8S, 16S, 18S, 23S, and 28S). All ORFs were compared to the Kyoto Encyclopaedia of Genes and Genomes (KEGG) database (Kanehisa and Goto 2000) for KEGG ID annotation. KEGG annotations were further reconfirmed through the GhostKOALA tool (Kanehisa et al. 2016).

GC–MS/MS metabolomics

GC–MS/MS-based volatile profiling with integrated QC measures

Volatile compounds from fermented rice beverages were extracted using HS-SPME following ISO 17943. Analysis was performed on a Shimadzu GCMS-TQ8030 system equipped with a triple quadrupole mass spectrometer and an Rtx-2330 capillary column (105 m × 0.25 mm × 0.20 μm). Samples were injected in split mode using an autosampler (AOC-20i), with helium as the carrier gas at 1.3 mL/min. The oven program was set from 100 °C (5 min hold) to 240 °C at 4 °C/min (20 min hold). Mass spectra (m/z 50–500) were acquired at 5 scans/sec under 70 eV ionization energy, and the injector and ion source temperatures were maintained at 220 °C and 230 °C, respectively. To ensure analytical reliability, quality-control (QC) procedures were incorporated. Procedural blanks containing only extraction solvents and derivatization reagents were processed identically to samples to monitor background contamination and carryover. Spike-recovery QC samples were prepared by fortifying rice beverage aliquots with known concentrations of volatile standards prior to extraction to assess matrix effects and extraction efficiency. Additionally, a pooled QC sample generated by combining equal volumes of all rice beverage samples was injected periodically (every 10 injections) to evaluate instrument stability, retention-time reproducibility, and signal drift. Compounds exhibiting poor recovery (< 80% or > 120%) or detectable signals in blanks were excluded from downstream interpretation. Compound identification was performed using AMDIS by matching spectra with the NIST library, and quantification was carried out using the area-normalization method (% composition) (Das et al. 2025).

Statistical analysis for metagenomics and metabolomics

The non-parametric Shannon index, Simpson's index of diversity (1–D), Chao-1 richness estimator, and evenness were calculated using PAST software v4.0. All quantitative measurements, including physicochemical parameters, antioxidant assays, and metabolite analyses, were performed in triplicate technical replicates derived from single fermentation batches. Reported standard deviations therefore reflect analytical reproducibility rather than biological variability across independent samples. Beta diversity was quantified using the Bray–Curti’s dissimilarity and Jaccard index in PAST v4.0 (Hammer et al. 2001), and the resulting distance matrices were visualized through Principal Coordinates Analysis (PCoA). Principal component and ordination analyses were used to explore patterns of association among variables and do not imply direct causal relationships between physicochemical parameters and microbial community structure. Relationships between microbial taxa across samples and functional pathway levels (super-pathways) were assessed using non-parametric Spearman’s rank correlation in IBM SPSS v20 and PAST v4.0 following the method described by Shannon et al. (2003) (Martino et al. 2019). Unique and shared microbial species among samples were analyzed using the jvenn interactive Venn diagram tool (Bardou et al. 2014). The measure the significance between samples at genus level were performed using One-Way ANOVA, Levene’s Test for homogeneity, Welch ANOVA, Kruskal–Wallis, and Random-Effects Models.

Area values of detected metabolites were normalized using the constant-sum method, log₁₀-transformed, and mean-centered prior to statistical analysis. Metabolomic data processing, scaling, and multivariate analyses were conducted using MetaboAnalyst 6.0 (Pang et al. 2024), MS Excel 2016, and Microcal Origin v7.5 (Pang et al. 2024).

To assess associations between microbial taxa and metabolite abundances while controlling for fermentation time, partial correlation analysis and multivariate regression models were employed. Both microbial relative abundances and metabolite intensities were log-transformed and standardized before analysis. Partial Spearman correlation coefficients (ρ_p) were computed with the ppcor package in R, adjusting for fermentation time as a covariate. In parallel, linear mixed-effects models (LMMs) were constructed using the lme4 package, with metabolite levels as dependent variables, microbial taxa as fixed effects, and fermentation time as a random intercept. Multiple-comparison correction was applied using Benjamini–Hochberg FDR (FDR < 0.05). Associations were considered robust only if significant in both partial correlation and LMM analyses. Correlation strength was interpreted based on Spearman correlation coefficients (ρ), where |ρ|= 0.3–0.5 was considered moderate and |ρ|> 0.5 was considered strong, and statistical significance was defined as an adjusted p-value (FDR) < 0.05. Significant confounder-adjusted associations were visualized as interaction networks using Cytoscape. Instrument performance and analytical reproducibility across metabolomic and metagenomic workflows were systematically evaluated by monitoring pooled QC-derived signal variation, sequencing read-quality parameters, assembly metrics, and annotation consistency throughout the analytical sequence, ensuring that observed variations in metabolite abundance and microbial functional profiles reflected biological differences rather than technical or computational artifacts.

Results

Proximate analysis

The physicochemical characteristics of four traditionally fermented rice beverages BR, RH, JOU, and SM are summarized in Supplementary File:ST 2. pH ranged from 4.1 to 4.5, with RH highest (4.5 ± 0.1) and JOU lowest (4.1 ± 0.2). Titratable acidity (% lactic acid) varied from 0.58 ± 0.04% (RH) to 0.72 ± 0.06% (JOU). Protein content ranged from 1.6 ± 0.1 g/100 g (JOU) to 2.1 ± 0.3 g/100 g (RH). Carbohydrate content was highest in RH (23.2 ± 1.4 g/100 g) and lowest in JOU (19.8 ± 0.9 g/100 g). Crude fat content was low (0.9–1.2 g/100 g), and starch ranged from 13.7 ± 0.6 g/100 g (JOU) to 16.8 ± 0.9 g/100 g (RH). RH showed relatively higher values of pH, protein, carbohydrate, and starch among all the samples (Fig. 2a). Overall, RH exhibited the highest nutritional parameters, including protein (2.1 g/100 g), carbohydrate (23.2 g/100 g), and starch (16.8 g/100 g), whereas JOU consistently showed the lowest values for these components, indicating variability in substrate utilization and fermentation efficiency among the beverages. The associated standard deviations shown in the figures represent variability among technical replicate measurements and should be interpreted as indicators of analytical precision rather than biological variation among independent fermentation batches.

Fig. 2.

a Proximate analysis from 4 traditionally fermented rice beer. b Ethanol concentration (%) derived from HPLC keeping standard as ethanol HPLC grade. c Antioxidant assay using DPPH free radical scavenging activity and Ferric Reducing Antioxidant Power from rohi modh (RH), black rohi modh (BR), sai mod (SM) and jou bidwi (JOU). Error bars represent mean ± standard deviation of triplicate technical measurements from single fermentation batches

Estimation of alcohol concentration and antioxidant assay

HPLC analysis of fermented rice beverages showed variation in ethanol concentration among samples (Fig. 2b). The highest ethanol level was observed in Black rohi modh (BR) (8.13 ± 0.21% v/v), followed by jou bidwi (JOU) (5.55 ± 0.18% v/v), rohi modh (RH) (5.44 ± 0.16% v/v), and sai mod (SM) (4.28 ± 0.14% v/v). The corresponding retention times ranged narrowly between 6.721 and 6.740 min, which falls within typical instrumental variation for HPLC analysis and was used for compound confirmation rather than quantitative comparison.

Antioxidant assay

The antioxidant potential of fermented rice beverage samples was assessed using two complementary assays: DPPH free radical scavenging activity and ferric reducing antioxidant power (FRAP). Among all the samples, RH exhibited the highest DPPH radical scavenging activity (57.99 ± 1.21%), closely followed by BR (56.91 ± 2.29%), whereas the lowest activity was recorded in SM (46.84 ± 1.39%). JOU showed moderate activity at 48.21 ± 1.31% (Fig. 2c). In the FRAP assay, SM demonstrated the highest ferric ion reducing capacity (96.93 ± 1.46 µmol/mL), followed closely by BR (96.67 ± 1.47 µmol/mL). JOU and RH displayed comparatively lower FRAP values of 86.12 ± 2.01 µmol/mL and 84.76 ± 2.36 µmol/mL, respectively (Fig. 2d).

Isolation, Phenotypic, genotypic characterization of LAB and yeast

A total of 68 lactic acid bacteria (LAB) and 55 yeast isolates were obtained from four traditional fermented rice beverages. LAB included 15 from BR, 25 from RH, 16 from SM, and 12 from JOU, while yeast isolates included 10 from BR, 21 from RH, 11 from SM, and 13 from JOU. RH accounted for the highest proportion of LAB isolates (36.8% of total isolates), whereas BR showed comparatively lower microbial diversity, suggesting differences in fermentation ecology among the beverages. LAB colonies were round, white or milky, smooth, moist, and glossy, with slight size and opacity variations. All LAB isolates were Gram-positive rods in pairs or chains. Yeast population was determined using the standard plate count method. The yeast load in the fermented rice beverage was 3.2 × 10⁷ CFU/mL, calculated from colony counts obtained at the 10⁻⁷ dilution. Under microscopic observation, the cells appeared hyaline and uniformly distributed, occurring singly or occasionally in pairs. Budding was evident in several fields, indicating active proliferation. The cytoplasm appeared dense and homogeneous, without visible intracellular granulation. Notably, no pseudohyphal formations were observed, and the cells maintained a typical unicellular morphology throughout examination (Supplementary File: ST 3).

16S rRNA gene-based analysis identified several LAB genera including Weissella cibaria (JB9, JB5), Weissella confusa (BR3), Pediococcus pentosaceus (JB3, BR2), Lactiplantibacillus plantarum (BR4, RH9, SM1, JB1, RH1), Levilactobacillus brevis (JB7, JB2, SM6, RH2), and Lentilactobacillus hilgardii (RH3, SM5, RH4, SM3, BR1). Yeasts were dominated by Saccharomyces cerevisiae in all samples: BR (BR1Y–BR3Y), RH (R1Y–R3Y), SM (SM1Y, SM2Y), JOU (J1Y–J4Y). While L1/L2 based analysis could identify Saccharomyces paradoxus was detected in RH (R4Y, R5Y) and SM (SM4Y), while Wickerhamomyces anomalus appeared in BR (BR4Y) and SM (SM3Y). Phylogenetic analysis (iTOL) confirmed clustering with reference sequences (Fig. 3a, b).

Fig. 3.

Phylogenetic tree constructed using iTOL of a Bacteria b Yeast keeping reference strain in blue color curated from NCBI

Taxonomy and diversity analysis

A total of 57,694,302 base pairs were recovered, averaging 14,423,575.5 bp per sample. The number of contigs varied across samples, ranging from 1030 in SAI to 5,083 in BR, with corresponding total assembly sizes between 21.25 Mb (SAI) and 33.42 Mb (RH). All assemblies contained contigs ≥ 1000 bp, while contigs ≥ 10,000 bp ranged from 413 in JOU to 714 in RH, and contigs ≥ 100,000 bp ranged from 6 in BR to 30 in SAI. No contigs exceeded 1 Mb in any dataset. The largest contigs measured 255,031 bp (JOU), 301,971 bp (SAI), 296,154 bp (BR), and 309,913 bp (RH). The N50 values showed considerable variation, with SAI exhibiting the highest N50 (50,608 bp), followed by JOU (32,388 bp), RH (19,333 bp), and BR (6,472 bp). Likewise, L50 ranged from 124 in SAI to 879 in BR, reflecting differences in assembly continuity. GC content was consistent across samples, ranging from 50.28% (SAI) to 51.18% (JOU). Among the assemblies, RH demonstrated the largest genome assembly size (33.42 Mb) and highest contiguity, while BR showed the lowest N50 value (6472 bp), indicating variability in sequencing depth and assembly quality across samples. (Supplementary File: ST 4). Metataxonomic analysis identified three domains, with eukarya being most abundant (66.12%), followed by bacteria (33.04%) and viruses (0.82%) (Fig. 4a). In total, 48 phyla, 321 families, 1034 genera, and 1785 species were detected. Ascomycota dominated in BR (83.27%), JOU (77.85%), and SM (51.93%), while Bacillota was highest in RH (55.21%) (Fig. 4b). Other major phyla included Proteobacteria, Actinobacteria, Bacteroidetes, Basidiomycota, and Virus. At the family level, Saccharomycetaceae was dominant in BR (83.25%), JOU (79.43%), and SM (52.68%), while RH was dominated by Lactobacillaceae (44.59%) (Fig. 4c). Saccharomyces was the most abundant genus in all samples BR (81.54%), JOU (79.62%), RH (35.89%), SM (54.85%) followed by Pediococcus (notably in RH: 17%) and Lentilactobacillus (highest in SM: 28.29%) (Fig. 4c and d).

Fig. 4.

Metataxonomy analysis of rohi modh (RH), black rohi modh (BR), sai mod (SM) and jou bidwi (JOU) of a Domain b Phylum c Family d Genus. e Canonical Correspondence Analysis of genus > 1% of major lactic acid bacteria and yeast shows clustering with alcohol (v/v%), titrable acidity (%) and pH. f Spearman’s rank correlation analysis among the major co-existing genera of bacteria and yeasts identified from metataxonomic analysis of 4 traditionally fermented rice beer

The biplot analysis revealed clear clustering of microbial genera in relation to physicochemical parameters, namely alcohol (% v/v), pH, and titratable acidity (TTA%). The first two principal coordinates explained most of the total variance (PCoA1: 87.67% and PCoA2: 12.24%), supporting the observed separation of microbial communities into three major groups (Fig. 4e). The upper-left cluster was strongly associated with alcohol content, as indicated by the directional vector. Fungal genera such as Rhodotorula, Pichia, Brettanomyces, Malassezia, Penicillium, Rhizopus, Exophiala, and Wickerhamomyces were positioned along the alcohol vector, suggesting a positive correlation with ethanol concentration. Saccharomyces and Acinetobacter were located closer to the origin but still aligned toward the alcohol axis, indicating moderate association. In contrast, the lower-left cluster was primarily influenced by pH and titratable acidity (TTA%), whose vectors were closely aligned, indicating a strong positive correlation between these two parameters. This region was dominated by lactic acid bacteria and related genera, including Lentilactobacillus, Pediococcus, Lactiplantibacillus, Levilactobacillus, Weissella, Leuconostoc, and Enterococcus. Their positioning suggests a significant contribution to acid production and pH reduction. The right-side cluster was separated along the positive axis of PC1 and showed weaker association with alcohol but moderate dispersion relative to pH and TTA%. Genera such as Aspergillus, Acetobacter, Clostridium, Pseudomonas, Raoultella, Paraburkholderia, and Companilactobacillus were grouped in this region, indicating distinct ecological and functional characteristics compared to the alcohol and acid-associated clusters (Fig. 4e).

Positive co-existence patterns between bacteria and yeasts were evident (Fig. 4f). Alpha diversity analysis indicated comparatively higher diversity values for RH (Simpson: 0.79; Shannon: 1.91), followed by SM, while BR and JOU showed relatively lower index values. BR recorded the highest species richness (Chao-1: 26) but lower evenness, suggesting dominance by few taxa (Fig. 5a). A total of 1,130 species were shared among all samples, while BR, JOU, RH, and SM had 863, 279, 380, and 61 unique species, respectively (Fig. 5b). Beta diversity (Bray–Curtis) revealed distinct clustering (Fig. 5c), with Jaccard index highlighting similarities between RH and JOU (Fig. 5d).

Fig. 5.

a Analysis of alpha diversity using Shannon_H, Simpson_1-D, Evenness and Chao-1. b shared and unique species between 4 traditionally fermented rice beer. Analysis of beta diversity (diversity difference) using c Bray–Curtis d Jaccard parameters clear difference in diversity from rohi modh (RH), black rohi modh (BR), sai mod (SM) and jou bidwi (JOU). Alpha diversity indices are presented as descriptive measures of community structure and were not subjected to formal statistical comparison

Significant differences in genus level were observed among the four beverage groups based on one-way ANOVA (F = 5.43, p = 0.0017), differences were considered statistically significant when p < 0.05. As the samples represent a single batch of technical replicates without biological replication, the reported statistical comparisons should be interpreted as exploratory and batch-specific rather than broadly generalizable. Confirming that group means were not equal. This result was further supported by a highly significant permutation test (p < 0.002), indicating that the observed group separation was unlikely to occur by chance. Variance component analysis showed that between-group variation (Var(group) = 1.34 × 10¹⁰) contributed meaningfully to the total variability, with an intraclass correlation coefficient (ICC = 0.311), suggesting moderate clustering by beverage type. The effect size (ω² = 0.089) indicated a small–moderate practical impact of group differences. Levene’s tests based on both means (p = 0.032) and medians (p = 0.047) revealed significant heterogeneity of variances, necessitating the use of robust tests. Accordingly, Welch’s ANOVA also confirmed significant differences across groups (Welch F = 4.92, df = 72.6, p = 0.0032). Non-parametric analysis supported these findings, with the Kruskal–Walli’s test indicating significant differences in median values among the groups (H_c = 9.232, p = 0.026).

Predictive functionality

Metagenomic ORFs from the fermented rice beverages were functionally annotated using the KEGG database, revealing six major functional categories, 47 super-classes, and 282 sub-classes. Metabolism was the most abundant category (40.57%), followed by human diseases (18.43%), genetic information processing (11.83%), organismal systems (10.32%), environmental information processing (9.79%), and cellular processes (9.12%) (Fig. 6a). At the level 2 super-class, carbohydrate metabolism (9.83%) and amino acid metabolism (6.82%) were most prevalent, alongside pathways like signal transduction (6.6%) and neurodegenerative diseases (4.58%) (Fig. 6b). Level 3 pathway analysis revealed strong correlations between Weissella cibaria, W. mesenteroides, Enterococcus hirae, and core pathways such as starch and sucrose metabolism, lipopolysaccharide metabolism, oxidative phosphorylation, and glycolysis/gluconeogenesis (Fig. 6c). Functional annotation of metagenomic sequences revealed that carbohydrate metabolism and amino acid metabolism were the most dominant pathways across all beverage samples, reflecting the central biochemical processes driving fermentation. These pathways accounted for the majority of annotated genes and are consistent with the metabolic requirements for sugar utilization, organic acid production, and microbial growth during fermentation, additionally functional prediction identified the presence of putative genes associated with retinol metabolism correlating with Saccharomyces cerevisiae and Lentilactobacillus hilgardii abundance. Prokka-based gene annotation predicted the presence of key enzymes alcohol dehydrogenase (adh), phosphatidylcholine-retinol O-acyltransferase (lrat), and cytochrome P450 enzymes (cyp1a1, cyp2a6, cyp3a4) (Fig. 6d) involved in retinol activation, storage, and degradation.

Fig. 6.

Distribution of different levels a Level 1 and b Level 2 of functional categories predicted from rohi modh (RH), black rohi modh (BR), sai mod (SM)and jou bidwi (JOU) metagenomes and annotated using KEGG. c Correlation between major species and level 3 of > 1% abundance using spearman’s correlation rank correlation and visualised using network analysis. d Retinol metabolism one of the unique level 3 pathway annotated using KEGG for vitamin A metabolism. Relative abundance values represent the proportion of annotated genes assigned to each functional category

GC–MS/MS metabolomics

QC measurement and diverse metabolite profile

No detectable signals corresponding to target metabolites were observed in solvent or procedural blanks, confirming the absence of reagent-derived contamination and instrument carryover. Spike-recovery assays performed at low, medium, and high concentration levels yielded recoveries between 86.4 and 112.3%, with relative standard deviations (RSDs) below 10%, thereby fulfilling widely accepted analytical performance criteria (70–120% recovery; RSD ≤ 15%). Matrix effects were minimal, with ion suppression or enhancement remaining within ± 12%, indicating negligible influence of the sample matrix on quantification accuracy. Pooled QC samples injected periodically throughout the run demonstrated excellent analytical stability, evidenced by metabolite peak-area RSDs ≤ 8.5% across the batch. Collectively, these QC outcomes confirm the reliability, reproducibility, and quantitative robustness of the GC–MS/MS metabolite profiling workflow (Supplementary File: ST5a).

Untargeted metabolomic profiling

GC–MS/MS-based untargeted metabolomics revealed diverse metabolite profiles in the fermented rice beverages, with 167 metabolites in BR, 113 in SM, 145 in JOU, and 123 in RH. A total of 93 compounds, including 59 primary metabolites (alcohols, esters, fatty acids, amino acids, organic acids), were common across all samples, highlighting both shared and distinct biochemical features. BR showed elevated levels of beta-alanine, DL-ornithine, glutamic acid, and unique metabolites like L-valine and L-Val-L-leu, indicating strong proteolytic activity. SM was enriched in serine, aromatic acids, and disaccharides (cellobiose, trehalose), suggesting active carbohydrate metabolism and antioxidant potential. JOU was characterized by sugar acids (2-Keto-L-gluconic acid, 2-Hydroxypropane-1,2,3-tricarboxylic acid) and unique sugar alcohols (D-mannitol, dulcitol, talose), pointing to oxidative sugar metabolism and microbial stress responses. RH showed distinct compounds like 3,4-Dihydroxymandelic acid and dicarboxylic acids, indicating aromatic amino acid degradation and secondary fermentation. Distinct metabolite signatures were detected across the samples, corresponding to differences in microbial community composition and fermentation profiles among the beverages (Fig. 7a, Supplementary File: ST 5b). Ethanol content was highest in BR (8.98%), followed by RH (5.67%), JOU (5.46%), and SM (4.99%).

Fig. 7.

a Heat map analysis of major 59 primary metabolites from GC–MS/MS analysis in rohi modh (RH), black rohi modh (BR), sai mod (SM)and jou bidwi (JOU). b shared and unique primary metabolites within rohi modh (RH), black rohi modh (BR), sai mod (SM)and jou bidwi (JOU) shows 6 metabolites in BR sample. c Spearman r’s rank correlation of major aroma related metabolites with major species with abundance > 1. Network connections represent statistically significant associations identified through correlation analysis after adjustment for fermentation time and multiple testing

Flavor and aroma compound

From the metabolite pool, 18 aroma-associated compounds were identified across the samples. These included amino acids such as L-proline, L-methionine, L-phenylalanine, and L-threonine; organic acids including 2-butenedioic acid, 2-hydroxypropane-1,2,3-tricarboxylic acid, propanoic acid, butanoic acid, and acetic acid; aromatic acids such as benzoic acid, benzeneacetic acid, and 3,4-dimethylbenzoic acid; and sugar alcohols including myo-inositol, D-mannitol, and dulcitol. Spearman correlation analysis revealed significant positive associations between several of these metabolites and dominant microbial taxa, including Saccharomyces cerevisiae, Enterobacter cloacae, Enterococcus faecium, Lentilactobacillus hilgardii, and Leuconostoc pseudomesenteroides, indicating co-occurrence patterns between microbial abundance and metabolite profiles across the fermented beverages (Fig. 7c).

After controlling for fermentation time through partial Spearman correlation analysis, several statistically robust microbes–metabolite associations remained significant (|r|≥ 0.60, FDR-corrected p < 0.05), indicating that these relationships were not driven solely by differences in fermentation duration. However, given the limited number of independent fermentation batches analyzed in the present study, the statistical power of these models remains constrained. Therefore, the identified associations should be interpreted as exploratory indicators of potential microbial–metabolite relationships rather than definitive causal interactions, and future studies with larger sample sizes will be necessary to confirm their reproducibility. Saccharomyces cerevisiae exhibited strong positive correlations with short-chain fatty acids, particularly propanoic acid (r = 0.77), butanoic acid (r = 0.26), and acetic acid (r = 0.26), while fermentation-time–adjusted negative associations were observed with polyols such as D-mannitol (r = − 0.77) and dulcitol (r = − 0.77). Wickerhamomyces anomalus and Weissella confusa retained similar metabolite association patterns after adjustment, showing strong positive correlations with amino acids—including L-methionine and L-phenylalanine (r = 0.63 each) and with propanoic acid (r = 0.77), suggesting coordinated metabolic contributions independent of fermentation length. Among lactic acid bacteria, Pediococcus pentosaceus demonstrated fermentation-time–corrected negative correlations with aromatic acids such as 3,4-dimethylbenzoic acid (r = − 0.95) and strong positive associations with sugar alcohols (D-mannitol and dulcitol, r = 0.77), reflecting species-specific carbohydrate metabolism. In contrast, Lactiplantibacillus plantarum displayed positive correlations with polyols (r = 0.77) but negative relationships with short-chain fatty acids (propanoic acid, r = − 0.77) even after adjusting for fermentation duration, indicating distinct metabolic pathways unaffected by time-related confounding. A cluster of taxa including Acinetobacter spp., chytrid fungi such as Batrachochytrium dendrobatidis, and zygomycetes showed nearly identical fermentation-time–adjusted correlation signatures, characterized by perfect positive correlations with propanoic acid (r = 1.00) and moderate positive associations with several amino acids, while uniformly negative correlations were observed with acetic, benzoic, and butanoic acids (r = − 0.33). Given the limited number of independent samples (n = 4), the statistical power of partial correlation and mixed-effects models is low; therefore, these analyses should be interpreted as exploratory hypothesis generation rather than confirmatory evidence, and reported correlation coefficients should be considered with caution.

Discussion

Traditional rice-based alcoholic beverages such as sai mod, black rohi modh, rohi modh, and jou bidwi are popular among indigenous communities in Northeast India because of their availability and affordability (Nath et al. 2019). Traditional rice-based beverages like rohi modh, sai mod, and jou-bidwi are believed to gain functional benefits from native microbial communities that naturally produce fermentation-derived metabolites. Integrated metagenomic and metabolomic analyses reveal diverse microbes, particularly LAB and yeasts, contributing to fermentation and flavor. However, despite their rich cultural heritage, scientific insights into their microbial and metabolic profiles remain limited.

The physicochemical properties of the present fermented rice beverages (BR, RH, JOU, and SM) showed moderate acidity (pH 4.1–4.5), which is slightly higher than the values reported for sai and rohi mod (pH 3.72–3.86) in the study by Kumari et al. (2024). Although pH and titratable acidity showed directional relationships with certain microbial groups, the relatively small numerical differences and overlapping variability among samples suggest that these parameters likely act in combination with other environmental and microbial factors, including unmeasured variables such as dissolved oxygen and raw material batch variation, rather than serving as sole drivers of community separation. Similarly, the reported pH range for different JOU variants (3.75–4.32) were reported in the study by Deka et al. (2018). Antioxidant-Activity and Physicochemical Indices of the Rice Beer Used by the Bodo Community in North-East India aligns closely with our jou sample (4.1 ± 0.2), suggesting comparable fermentation-driven acidification. Alcohol content reported in earlier studies was considerably higher (26–35% v/v for sai and rohi; up to 332.92 mg/mL ethanol in distilled jou), whereas the present study focused primarily on the undistilled products. Protein levels in our samples (1.6–2.1 g/100 g) were substantially higher than those reported for Sai and Rohi (0.49–0.51 µg/mL) and Jou (0.35–0.79 mg/mL), likely due to differences in expression units, sample concentration, and analytical methods. Carbohydrate content in our beverages (19.8–23.2 g/100 g) was also higher than the soluble sugar values reported in earlier works, which measured residual sugars rather than total carbohydrate fractions (Deka et al. 2018; Kumari et al. 2024). Black rohi modh (BR), jou bidwi (JOU), rohi modh (RH), and sai mod (SM). HPLC analysis showed that BR had the highest alcohol content (8.13%), followed by JOU (5.55%), RH (5.44%), and SM (4.28%), there was no presence of methanol concentration in any 4 fermented beverages, whereas, via GC–MS/MS ethanol content was highest in BR (8.98%), followed by RH (5.67%), JOU (5.46%), and SM (4.99%). Ethanol levels for BR sample obtained using HPLC (8.13% v/v) and GC–MS/MS (8.98% v/v) differed slightly due to the inherent characteristics of the two analytical platforms. Antioxidant assays revealed functional differences among samples, with RH showing the highest DPPH activity (57.99 ± 1.21%) and SM the lowest (46.84 ± 1.39%). Interestingly, SM exhibited the highest FRAP value (96.93 ± 1.46 µmol/mL), suggesting the presence of other antioxidant compounds. While numerical differences in antioxidant activity and metabolite abundance were observed among samples, these comparisons should be interpreted as batch-specific observations rather than statistically generalizable biological differences due to the limited number of independent fermentation batches analyzed. A more detailed metagenomic analysis indicated an intricate structure of the microbial community.

The microbial profile of fermented rice beverages revealed 68 LAB and 55 yeast isolates, with RH showing the highest counts. Dominant LABs included Lactiplantibacillus plantarum, Lentilactobacillus hilgardii, and Levilactobacillus brevis, present across all samples. ITS sequencing identified Saccharomyces cerevisiae, Wickerhamomyces anomalus, and S. paradoxus as the most common yeasts. Saccharomyces cerevisiae was also found in zutho a rice beverage from Nagaland, India (Teramoto et al. 2002). Similarly, rice beverage like chubitchi by Garo, kyiad by Khasi and sadhiar by Jaintias tribes of Meghalaya also had a presence of Saccharomyces cerevisiae and Rhodotorula muciliginosa in abundance (Kumar Mishra et al. 2018). S. cerevisiae is widely utilized in baking and alcohol industries due to its robust sugar metabolism but lacks native amylolytic activity, meaning it cannot degrade starch directly without external enzymes (Lahue et al. 2020). Similarly, S. paradoxus is saccharolytic but not amylolytic. In contrast, W. anomalus exhibits both saccharolytic and weak amylolytic activities, with some strains reported to produce extracellular enzymes like α-amylase and glucoamylase, enabling limited starch breakdown (Boynton et al. 2021). This was also mentioned in the fermented rice beverage known as haria hailing from west Bengal (Tamang et al. 2012). Previous studies have identified key microbial taxa involved in the fermentation of traditional rice beverages like sai mod and jou bidwi. In sai mod, the presence of amylolytic microbes, along with Rhodotorula taiwanensis and Wickerhamomyces anomalus, suggests a community capable of breaking down complex carbohydrates and contributing to aroma and flavor (Buragohain et al. 2013; Tamang et al. 1988). In contrast, jou bidwi has been reported to harbor beneficial genera such as LAB, Roseburia, and Faecalibacterium sp., along with fermentative yeasts like Saccharomyces cerevisiae, Pichia burtonii, and Wickerhamomyces anomalus.

Metataxonomic analysis revealed a highly diverse microbial community across all fermented rice beverage samples, comprising 1,785 species distributed among 48 phyla. Variability in assembly metrics, including N50 and L50 values observed among samples, indicates differences in assembly continuity that may influence the detection sensitivity of genes and taxa, particularly low-abundance organisms. Assemblies with higher contiguity generally enable more complete gene prediction and functional annotation, whereas more fragmented assemblies may underestimate diversity or gene content. Therefore, comparisons of taxonomic richness and functional profiles in the present study were interpreted cautiously, focusing on overall community patterns rather than absolute quantitative differences. Ascomycota dominated in BR, JOU, and SM, whereas Bacillota was most abundant in RH. At the family level, Saccharomycetaceae prevailed in BR, JOU, and SM, while Lactobacillaceae dominated in RH. At the genus level, Saccharomyces was predominant across all samples, particularly in BR (81.54%) and JOU (79.62%). It should be noted that these values represent relative abundances derived from compositional metagenomic data rather than absolute microbial counts. Consequently, differences in percentage abundance among samples may reflect proportional changes within the microbial community rather than true differences in organism quantity. Without absolute quantification or external normalization, comparisons of relative abundance should therefore be interpreted cautiously as indicators of community structure rather than direct measures of microbial population size. Pediococcus was the second most abundant genus in BR, JOU, and RH, whereas Lentilactobacillus dominated in SM (28.29%). Comparable microbial patterns were reported by Yumnam et al. (2024) in a metagenomic study of rice-based fermented beverages from five ethnic tribes of southern Assam (Zeme Naga, Dimasa Kachari, Hmar, Karbi, and Tea tribes). Firmicutes, Proteobacteria, Actinobacteria, and Bacteroidetes were dominant at the phylum level, with genera such as Lactobacillus, Bacillus, and Pediococcus prevailing. However, PERMANOVA analysis showed no significant differences in phylum-level composition among tribes (p > 0.05), suggesting functional stability despite taxonomic variation. Similarly, a study on the alcoholic rice beverage Zufang by Lalbiakmawia et al. (2024) reported Firmicutes as the overwhelmingly dominant phylum, with Lactobacillales accounting for more than 99% of total relative abundance, indicating a highly specialized lactic acid bacteria–driven fermentation system. Further supporting these findings, Bhattacharjee et al. (2023) characterized the starter culture Bakhor and identified 18 phyla, 28 classes, 41 families, and 45 genera, with Firmicutes and Proteobacteria as dominant phyla. Abundant genera included Bacillus, Lactobacillus, Clostridium, and Lactococcus. From an ecological perspective, the dominance of Saccharomyces and lactic acid bacteria reflects a classic succession pattern in carbohydrate-rich fermentation environments, where rapid sugar fermentation lowers pH and selectively favors acid-tolerant microbes. Such ecological filtering has been widely documented in spontaneous cereal fermentations, where microbial communities stabilize through competitive exclusion and metabolic cross-feeding interactions between yeast and bacteria. The observed co-occurrence of fermentative yeasts and lactic acid bacteria in the present study therefore represents a predictable ecological adaptation to nutrient availability and acidic fermentation conditions.

Metagenomic functional analysis of the four traditional rice-based fermented beverages predicted significant potential metabolic pathways. KEGG-based annotation identified six major categories, with metabolism (40.57%) as the most dominant, followed by human diseases (18.43%) and genetic information processing (11.83%). Carbohydrate (9.83%) and amino acid metabolism (6.82%) pathways were particularly enriched. Comparable functional profiles have been reported in other fermented rice and cereal systems. For example, metagenomic analysis of rice-flavor Baijiu fermentation revealed that metabolism-related genes accounted for approximately 74.08% of total annotated functions, with carbohydrate metabolism representing 10.2% and amino acid metabolism 7.4% of predicted pathways (Yuan et al. 2025). Similarly, a traditional rice wine starter culture (Xaj-pitha) showed metabolism contributing 46.9% of functional genes, with carbohydrate metabolism ranging between 8.5 and 11.3% depending on fermentation stage (Bora et al. 2016). The relative abundance values observed in the present study therefore fall within the reported range for cereal-based fermentation systems, indicating functional consistency across geographically distinct fermentation ecosystems.

In comparison, pathways related to retinol metabolism were detected at substantially lower relative abundance and are interpreted as auxiliary metabolic functions rather than dominant fermentation pathways. The detection of retinol metabolism–related genes in the present metagenomic dataset likely reflects the gene presence prediction involved in lipid and carotenoid transformation rather than direct vitamin A biosynthesis. In microbial fermentation systems, cytochrome P450 enzymes and retinol dehydrogenases participate in oxidative metabolism of terpenoid and lipid substrates, contributing to membrane remodeling and stress adaptation under anaerobic or nutrient-limited conditions. These pathways have been reported in yeast and lactic acid bacteria as auxiliary metabolic systems associated with redox balance and detoxification rather than primary vitamin production (Cuamatzin‐garcía et al. 2022) Importantly, the relative abundance of retinol metabolism genes in the present study was considerably lower than carbohydrate metabolism (9.83%) and amino acid metabolism (6.82%), supporting the interpretation that retinol-related functions represent minor ancillary pathways rather than dominant fermentation mechanisms.

Similarly, in a metagenomic analysis of rice-flavor Baijiu fermentation, functional annotation against the KEGG database revealed that approximately 74.08% of the predicted genes mapped to metabolism at KEGG level 1, with carbohydrate and amino acid metabolism pathways being particularly enriched at KEGG level 2 (Yuan et al. 2025). Similar functional trends have been reported in other rice-based fermentation metagenomes. For instance, KEGG annotation of a traditional rice wine starter culture (Xaj-pitha) from Assam revealed that metabolism accounted for 46.9% of annotated functions, followed by genetic information processing (22.7%), environmental information processing (14.3%), and cellular processes (11.8%) (Bora et al. 2016).

GC–MS/MS-based untargeted metabolomics identified 167, 113, 145, and 123 metabolites in BR, SM, JOU, and RH, respectively, with 93 shared compounds, including 59 primary metabolites. BR was particularly rich in beta-alanine, DL-ornithine, and glutamic acid, along with unique compounds such as L-valine and L-Val-L-leu, predicting strong proteolytic and antioxidant potential (Wang et al. 2025). It should be noted that metabolite abundances in the present study were estimated using area normalization (% composition), which reflects relative signal intensity rather than absolute metabolite concentration. Consequently, statements regarding metabolite enrichment or functional importance should be interpreted as indicative of comparative metabolic patterns rather than quantitative differences in metabolite levels among samples. The observed metabolite diversity likely reflects substrate-specific microbial enzymatic activity during fermentation. Amino acid–derived metabolites such as L-valine and DL-ornithine suggest enhanced proteolysis mediated by lactic acid bacteria and yeast proteases, whereas sugar alcohols detected in JOU indicate osmotic stress adaptation and redox balancing mechanisms in fermentative microbes. Similar metabolite-driven functional diversification has been reported in traditional rice beer systems, where carbohydrate fermentation and amino acid catabolism were identified as dominant metabolic drivers of flavor and bioactivity (Das et al. 2019). SM featured serine, aromatic acids, and disaccharides (trehalose and cellobiose), may consequently suggest active carbohydrate metabolism and stress-protective functions (Albertorio et al. 2007; Handzlik and Metallo 2023). JOU contained sugar acids and sugar alcohols (D-mannitol and dulcitol), suggesting oxidative sugar metabolism (Meena et al. 2015). RH showed metabolic signatures linked to aromatic compound degradation and secondary fermentation processes. BR exhibited six unique biofunctional metabolites, while JOU contained ten distinctive compounds. In contrast, SM and RH mainly contributed to the shared metabolite pool without exclusive markers. In a similar study on Apong (Poro and Nogin), Xaaj and Joubishi by Das et al. (2019) metabolomic profiling of fermented rice beer revealed a remarkably diverse chemical landscape, with a total of 230 compounds detected across all samples. However, focusing on the 66 metabolites directly involved in metabolic and functional processes provided a clearer picture of the biochemical activities shaping these traditional beverages. These metabolites were dominated by saccharides (18) and organic acids (18), followed by sugar alcohols (11), amino acids (8), and a smaller number of specialized compounds, including a vitamin and several secondary metabolites. A total of 18 aroma-related metabolites were identified, playing key roles in the flavor, fragrance, and sensory attributes of the beverages. Amino acids such as L-proline, L-methionine, L-phenylalanine, and L-threonine serve as precursors for volatile compounds that impart sweet, floral, and sulphurous notes (Zhang et al. 2022). Organic acids, including 2-butenedioic acid, propanoic acid, and acetic acid, contribute sour, cheesy, and pungent aromas characteristic of fermentation (Sorathiya et al. 2025). Aromatic acids such as benzoic acid and 3,4-dimethylbenzoic acid add subtle bitterness and sharpness, while sugar alcohols like D-mannitol and dulcitol influence mouthfeel and aroma retention (Mäkinen 2010).

Despite exhibiting the highest FRAP value (96.93 μmol/mL), SM showed comparatively lower DPPH radical-scavenging activity (46.84%). This divergence reflects the distinct antioxidant mechanisms captured by the two assays. FRAP primarily detects electron-donating reducing agents, whereas DPPH measures the hydrogen-atom–donating capacity of free radical quenchers. Consistent with these assay principles, the GC–MS/MS profile of SM revealed relatively higher abundances of compounds associated with reducing power, such as organic acids, simple sugars, and fermentation-derived intermediates, supporting its strong FRAP response. In contrast, metabolites typically linked to higher DPPH scavenging activity, including aromatic phenolics, specific aldehydes, and higher alcohols, were present at comparatively lower levels in SM. Spearman correlation analysis revealed strong associations between these metabolites and key lactic acid bacteria and Saccharomyces cerevisiae. These results collectively demonstrate that microbial metabolism governs biochemical diversity through coordinated carbohydrate utilization, amino acid catabolism, and stress-adaptive metabolic pathways during fermentation. A limitation of the present study relates to the restricted availability of authentic household-fermented rice beverage samples, where each beverage type could be collected only once from its traditional preparation setting. Consequently, the dataset should be interpreted as a case-specific representation of existing fermentation practices rather than a statistically representative survey of all regional production systems. While triplicate analytical measurements were performed to ensure methodological reliability, biological replication across multiple households and fermentation batches was not feasible due to the declining prevalence and seasonal nature of traditional preparation. Therefore, caution should be exercised in generalizing these findings to broader populations. Future investigations incorporating larger sample sizes, repeated sampling across locations, and temporal monitoring of fermentation batches will be essential to strengthen statistical robustness and improve the generalizability of microbial and metabolomic patterns observed in traditional fermented rice beverages.

Conclusion

This study provides an integrated culture-dependent, metagenomic, and metabolomic characterization of four traditional fermented rice beverages, revealing diverse microbial communities dominated by lactic acid bacteria and yeasts that drive fermentation-related biochemical transformations. Functional annotation demonstrated that carbohydrate and amino acid metabolism were the principal pathways underlying fermentation, while metabolomic profiling identified distinct and shared metabolites associated with flavor development and microbial activity. These findings should be considered preliminary and warrant validation using independent sample batches collected across different regions and multiple years to ensure robustness and broader applicability. The combined multi-omics approach establishes mechanistic links between microbial composition and metabolite production, offering a comprehensive scientific framework for understanding fermentation ecology and supporting future development, standardization, and functional evaluation of traditional rice-based beverages.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Rohit Das: Original draft, Visualization, Validation, Methodology, Investigation, Formal analysis, Conceptualization. Madhu Chandana Medhi: Formal analysis, methodology. Buddhiman Tamang: Original draft, Visualization, Validation, Methodology, Investigation, Formal analysis, Conceptualization.

Funding

This study did not receive any kind of intramural or extramural funding.

Data availability

The data related to sanger sequencing is present in the form of accession number at the end of name of each organism and yeast in Fig. 3 and raw data for shotgun metagenomics is present in NCBI with BioProject id: PRJNA1284041.

Declarations

Competing interests

The authors declare no competing interests.