To understand how glycans influence the growth of probiotic strains in monocultures and co-cultures, we characterized three bacterial species: AM BAA-835, BB JCM1192, and BL JCM1217. We chose brain-heart infusion (BHI) as the base media for our system since it can minimally support the growth of all species (allowing us to obtain sufficient samples for proteomics analysis), and their growth could be further enhanced by supplementing other carbon sources (Supplementary Figure S1A). We characterized their growth anaerobically in the base media containing either glucose and N-acetyl-D-glucosamine (Glc + GlcNAc) or mucin in monocultures and co-cultures (Figure 1A). Mucin is one of the host-derived glycans in the gut, serving as a continuous endogenous source of microbiota-accessible carbohydrates for resident microbes. While AM can utilize mucin as its sole carbon and nitrogen source (Kim et al., 2022), Bifidobacterium spp. have varying degrees of mucin degradation capabilities (Ruas-Madiedo et al., 2008; Abe et al., 2010). It is unclear how mucin affects the growth of AM in co-culture with Bifidobacterium spp. and how it impacts other cellular processes beyond the expression of mucin utilization-related enzymes.

Based on the growth profiles, all species reached maximum population size within 24 h (i.e., entered the stationary phase) (Figures 1B,C; Supplementary Figure S1B). By comparing the growth in the presence and absence of Glc + GlcNAc or mucin, we observed that all species can utilize Glc + GlcNAc. By contrast, mucin can be utilized by AM, but not by BL. While BB can also utilize mucin, the amount of growth was much lower than that of AM. Growth characterization via optical density (OD₆₀₀) measurements or CFU counting resulted in a consistent trend across all media conditions (Figures 1D,F; Supplementary Figure S1C). While OD₆₀₀ can be used to assess the total growth of bacteria in communities, CFU counting using selective plates enables growth measurement of individual species within the communities. In co-culture, AM displayed differences in growth profile with the two Bifidobacterium spp. (Figures 1E,G). After 24 h of growth, the relative abundance of AM in AM + BB co-culture was low in the presence of Glc + GlcNAc (~20%) and increased in the presence of mucin (~54%). The absolute abundance of AM (quantified by CFU counting) was lower in AM + BB co-culture compared to AM monoculture in the Glc + GlcNAc media (Figures 1D,E), whereas the abundance of AM was similar in AM + BB co-culture and AM monoculture in the mucin media (Figures 1F,G). By contrast, AM dominated BL in the AM + BL co-culture in the presence of both carbon sources after 24 h of growth (relative abundance of ~68 and 83% in Glc + GlcNAc and mucin, respectively). In both media, the absolute abundance of BL was lower in AM + BL co-culture compared to the BL monoculture (Figures 1D–G).

To investigate proteome alterations of AM, BB, and BL in co-culture vs. monoculture across the two carbon sources and how they are associated with changes in the growth profiles, we subjected them to proteomic analysis (See Methods, Supplementary Figures S2, S3). AM, BB, BL, AM + BB, and AM + BL were cultured for 24 h in media containing different carbon sources since this duration allows all species to reach maximum population size (Figure 1B), followed by protein extraction and LC–MS/MS analysis.

In monoculture, AM exhibited higher expression of many glycosyl hydrolases required for mucin degradation, such as GH20 (Xu et al., 2020; Chen et al., 2019), GH35 (Kosciow and Deppenmeier, 2019; Guo et al., 2018), and GH16 (Crouch et al., 2020) enzyme families in the presence of mucin compared to Glc + GlcNAc (Figures 2A,D,G; Supplementary Table S2). Other glycosyl hydrolases that were expressed higher by AM in the presence of mucin (e.g., GH2, GH13, GH33, GH43, GH89, GH95) have also been shown to be important for growth in mucin, as their deletion led to profound growth defects in mucin medium (Davey et al., 2023). In the presence of Glc + GlcNAc, AM displayed higher expression of transporter proteins and those that play a role in glycolysis, sulfur metabolism, and stress response. BB also exhibited higher expression of proteins involved in glycan breakdown and metabolism in the presence of mucin than Glc + GlcNAc (Figures 2B,E,H; Supplementary Table S3). By contrast, the expression of proteins for stress response, amino acid biosynthesis, and iron binding and transport was higher in the presence of Glc + GlcNAc. Since BL cannot utilize mucin, it did not show differential expression of enzymes for glycan breakdown in the presence of mucin compared to Glc + GlcNAc (Supplementary Table S4). The number of differentially expressed proteins (DEPs) of BL in mucin vs. Glc + GlcNAc was higher compared to AM or BB (Figures 2C,F,I). The DEPs for BL in mucin vs. Glc + GlcNAc are involved in cofactor biosynthesis, amino acid biosynthesis, translation, stress response, cell wall biosynthesis, cell division, DNA recombination, metabolic processes, and transporters.

When analyzing proteomics data of AM + BB and AM + BL co-culture (Figure 3A), there is a possibility that a peptide sequence belongs to the proteins of both AM and BB or BL (i.e., shared peptides). To evaluate the severity of this issue, we performed in silico tryptic digestion of the whole proteome of AM, BB, and BL using the Uniprot database and generated a list of all possible shared peptides between AM and BB or AM and BL (see Methods). The percentage of theoretical shared peptides across the proteome for AM + BB, AM + BL, and BB + BL is 0.222% (779/351,301), 0.218% (774/354,827), and 14.8% (46,083/310,822), respectively. This is consistent with the fact that AM is not closely related to BB and BL, whereas BB and BL are closely related. Most of these shared peptides between AM and BB or BL are short (~5 amino acids), which was the threshold we set in our database searching (Supplementary Figure S4). We then searched whether the peptides identified in the co-culture proteomic samples contain any of the shared peptides. There is a very low number of shared peptides detected across our proteomic datasets, ranging from 13 to 19 in each sample. To take a conservative approach, we assigned the spectra of the shared peptide to both species in the co-culture. Unconfident peptide assignment with a low number of sibling peptides will be removed by the Trans-Proteomics Pipeline (TPP) algorithm.

For the proteomics analysis, we first compared the proteome profile of the samples under different carbon sources to observe how resource environments impact protein expression. Then, we compared the proteome profile of each species in monoculture versus co-culture under the same carbon source to reveal the effect of the partner species on the protein expression of a particular strain. For co-culture samples, the expression of a protein was normalized with the expression of the whole proteins of the species that express that particular protein. In AM + BB co-culture, AM has higher protein identification in the presence of mucin (56%, 1,562 out of 2,777 total proteins) compared to Glc + GlcNAc (45%, 1,083 out of 2,429 total proteins) (Figure 3B). Similarly, the total protein expression of AM is also higher in mucin than Glc + GlcNAc when co-cultured with BB (57% vs. 23% of the total co-culture protein expression) (Figure 3C). By contrast, the protein identification and total protein expression of AM are similarly high when co-cultured with BL in Glc + GlcNAc and mucin (Figures 3D,E). The number of proteins identified for AM when co-cultured with BL in Glc + GlcNAc and mucin are 1,641 out of 2,105 (78%) and 1,663 out of 2,051 (81%) respectively, whereas the total protein expression of AM when co-cultured with BL in Glc + GlcNAc and mucin are 92 and 94% of the total co-culture protein expression, respectively. These results are consistent with the higher relative abundance of AM in co-culture with BB when cultured in mucin than in GlcNAc, and the higher fraction of AM in AM + BL co-culture both in the presence of Glc + GlcNAc and mucin (Figures 1D–G).

In the presence of Glc + GlcNAc, AM did not significantly affect the proteome of BB, whereas BB led to substantial alterations in the proteome of AM (~28% DEPs, Figure 3F; Supplementary Figure S5; Supplementary Tables S5, S6, S12, S13). Many of the up-regulated proteins were those involved in translation, amino acid biosynthesis, and carbohydrate metabolism, whereas the down-regulated proteins were glycosyl transferases and those that play a role in amino acid biosynthesis, carbohydrate metabolism, stress response, and cell wall biosynthesis. By contrast, BB and AM did not significantly alter the protein expression of each other in the presence of mucin (Figure 3H; Supplementary Tables S7, S8). In the presence of Glc + GlcNAc, BL did not significantly affect the proteome of AM, whereas AM led to substantial alterations in the proteome of BL (~34% DEPs) (Figure 3G; Supplementary Table S9). Similarly, BL did not affect the proteome of AM, and AM substantially altered the proteome of BL in the presence of mucin (~38% DEPs) (Figure 3I; Supplementary Tables S10, S11).

In sum, the two nutrient environments (Glc + GlcNAc and mucin media) affect the proteome expression of AM and BB in co-culture, but not AM and BL. While BB induced substantial alterations of AM’s proteome in Glc + GlcNAc medium, this was not observed in the mucin medium. On the other hand, AM induced proteome alterations in BL in both Glc + GlcNAc and mucin media.

We hypothesized that providing resources accessible only to BL would alleviate the reduced growth of BL in co-culture with AM compared to monoculture (Figures 1D–G). To this end, we screened for health-relevant diet-derived glycans that can be exclusively utilized by BL (Figures 4A,B; Supplementary Figure S6). We used inulin, xylan, arabinoxylan, arabinogalactan, and gum Arabic. Inulin (found in chicory root, artichoke, bananas, and onions) and acacia gum (arabinogalactan is a major component) have been demonstrated to increase both butyrate production in human subjects and health-relevant Bifidobacteria in vitro (Phillips et al., 2008; Kaddam and Kaddam, 2020; Belenguer et al., 2006). Xylan, an abundant component of cereal grains, has been shown to exhibit health-promoting properties (Smith and Melrose, 2022). Arabinoxylan (AX) is an important constituent of hemicelluloses in the endosperm and outer layers of cereal grains, including corn, wheat, rye, barley, oat, and rice. Of the five tested glycans, BL displayed a high degree of growth in the presence of AX (Figures 4A,B; Supplementary Figure S6) and exhibited a higher relative abundance than AM in co-culture (Figure 4C). By contrast, AM and BB cannot utilize AX in monoculture. Notably, in the presence of AX, BL increased the abundance of AM after 24 h of growth in co-culture compared to monoculture, likely via cross-feeding of sugars liberated from AX (Figure 4C).

While the proteome profile of AM was similar in the presence and absence of AX in monoculture (Figures 4D,F), BL up-regulated enzymes for AX degradation and arabinose/xylose utilization in the presence of AX (Figures 4E,G–H; Supplementary Table S15). In addition, we also observed some other changes in the biological processes of BL in the presence of AX (e.g., proteins for cell wall biosynthesis, stress response, metabolic processes, amino acid biosynthesis, and transporters), possibly associated with the cost of producing those enzymes (Figure 4H). In the presence of AX, BL displayed higher protein identification and total protein expression in AM + BL co-culture compared to the absence of AX (Figure 5A). Unlike in the Glc + GlcNAc or mucin media, AM did not alter the proteome of BL in the presence of AX (Figure 5B). On the other hand, BL led to substantial alteration in the proteome of AM (Figures 5B,C; Supplementary Table S16). Several proteins involved in carbohydrate metabolism, stress response, and cell wall biosynthesis were expressed at higher levels in AM monoculture, whereas proteins involved in translation and pili or type II secretion were expressed at higher levels when co-cultured with BL. These pili/type II secretion-related proteins were reported to be important for utilizing mucin (Davey et al., 2023), and thus, our results suggest that these proteins might be involved in other processes beyond mucin utilization. Notably, the membrane protein Amuc_1100 was up-regulated in the presence of BL (Figures 5B,C; Supplementary Table S16). Amuc_1100 is one of the most important proteins in AM that activates TLR-2 in IECs, improves gut barrier integrity, and was reported to replicate most of the beneficial effects of the bacteria (Plovier et al., 2017; Ottman et al., 2017).

Previous studies have shown that AM (Reunanen et al., 2015) and BL (Srutkova et al., 2015) enhance the barrier integrity of IECs through different mechanisms. AM adheres to the intestinal lining, and its outer membrane protein Amuc_1100 binds to TLR2 in IECs to activate the downstream signaling pathway, including the expression of tight junction proteins. By contrast, BL enhances barrier integrity by modulating immune responses and secreting extracellular vesicles (Nie et al., 2025). To compare the membrane integrity-enhancing activity of AM, BL, and AM + BL in the presence and absence of AX, we co-cultured the bacteria with IEC-6 cells and performed a permeability assay using a previously established protocol (Gildea et al., 2017; Fan et al., 2021; Fan et al., 2020). Although Caco-2 cells (derived from human colon cancer) have been used as a cell model in the studies of intestinal barrier function (Wilson et al., 1990), the transepithelial electrical resistance (TEER) value of the Caco-2 cell monolayer (~900 Ω cm²) is much higher than that of the intestinal epithelium (~40 Ω cm²) (Shi and Zhao, 2019). Thus, the practicability of Caco-2 cells as a cellular permeability model might have some limitations. For this reason, IEC-6 cells [originated from the rat small intestinal crypt (Quaroni et al., 1979)] have been used to study the proliferation and differentiation of IECs since they better replicate the extracellular transport characteristics of intestinal epithelial cells (Fan et al., 2021). The TEER value of the IEC-6 cell monolayer is lower than that of Caco-2 but closer to the actual situation of the small intestine.

AM, BL, and AM + BL displayed similar positive effects on membrane integrity (~14 to 17% decrease in leakiness), whereas supplementing AX to the AM + BL pairwise community during incubation with IEC-6 cells increased the positive effect (~33% decrease in leakiness) (Figure 5D). Supplementing AX alone did not affect the membrane integrity, and the reduction in permeability due to AM + BL + AX treatment was significantly larger than AM or BL alone. This could be because AX enhanced the growth of BL when incubated with IEC-6 cells (Figure 5E), which subsequently promoted the growth of AM and up-regulated the expression of Amuc_1100 as observed from our proteomics data. In sum, our results show that the AM + BL pairwise community, with AX supplementation, led to the highest improvement in the barrier integrity of IECs.

To investigate the changes in protein expression of IEC-6 cells upon co-culture with AM, BL, and AM + BL in the presence and absence of AX, we subjected the IEC-6 cells that were treated with bacterial samples for 6 h to proteomics analysis (Supplementary Figure S7A). Treatment of IEC-6 cells with different bacterial samples led to a similar degree of proteome alterations (3.1, 4.5, 3.0, and 2.6% DEPs for IEC-6 cells treated with AM, BL, AM + BL, and AM + BL in AX compared to untreated cells, respectively) (Supplementary Figure S7B; Supplementary Tables S17–20). However, our proteomics approach was unable to detect most of the tight junction proteins (peptide spectrum match (PSM) value of 0), such as Claudins (Cldn-1, Cldn-18, Cldn-19, Cldn-3, Cldn-16, Cldn-5, Cldn-7, and Cldn-11) and Occludin, although we detected Cldn-15 in low abundance (PSM values of 1–2 across samples). This suggests that their expression levels were low under our experimental conditions, and/or enrichment steps during sample preparation are needed to increase their abundance in the samples. Junctional adhesion molecules (JAMs) were also detected in low abundance (average PSM of 2 across all samples for JAM-C or JAM-3, and average PSM of 3 for JAM-A or JAM-1, whereas JAM-4 was not detected with PSM of 0 across all samples). We also analyzed other junction proteins, including adheren junction proteins and desmosome-related proteins, and gap junction proteins (Supplementary Figure S7C). Both AM and BL treatments led to the up-regulation of protein kinase C iota type (Prkci) by 2.1 and 2.8-fold for co-culture with AM and BL, respectively. Notably, the expression of Cytohesins, in particular Cyth-2, was uniquely high in the AM + BL + AX treatment group (up-regulated by ~5.6-fold) and was not observed in other groups. Further mechanistic study on how BL improves the efficacy of AM in the presence of AX and how this leads to the enhancement of gut barrier integrity, combined with further in vivo validation experiments, could have important therapeutic implications.

The strains used in this work were obtained from the sources listed in Supplementary Table S1. Two media formulations were tested to select the base media for this study: DeMan-Rogosa-Sharpe (MRS) broth (BD Difco™) and Brain Heart Infusion (BHI) broth (Thermo Scientific™ Oxoid™) (Derrien et al., 2004; Anand et al., 1984). BHI was chosen as the base media due to its ability to minimally support the growth of all species and allow growth enhancement when supplemented with specific additional carbon sources (Supplementary Figures S1A,B). For the experiments in this study, the base media was supplemented with either glucose (Glc) and N-acetyl-D-glucosamine (GlcNAc) as previously described (Plovier et al., 2017), 1% mucin, or 1% dietary fiber (inulin, xylan, arabinoxylan, arabinogalactan, or gum Arabic).

For all experiments, cells were cultured in an anaerobic chamber (Coy Lab products) with an atmosphere of 3.0 ± 0.5% H₂, 15 ± 1% CO₂, and balance N₂ at 37 °C. Starter cultures were inoculated by adding 100 μL of a single-use 25% glycerol stock to 3 mL of BHI broth supplemented with Glc + GlcNAc, and cultured at 37 °C without shaking. All strains used in this study displayed an OD₆₀₀ of ~0.8 or higher after overnight culture in this media condition (Supplementary Figure S1A).

Rat ileum epithelial IEC-6 cell line (CRL-1592) was purchased from the American Type Culture Collection (ATCC). The cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM, 4.5 g/L glucose) (Gibco, Thermo Fisher Scientific) supplemented with 3.7 g/L NaHCO₃, 10% (v/v) fetal bovine serum (FBS) (Gibco, Thermo Fisher Scientific), and 1% (v/v) Antibiotic-Antimycotic cocktail (penicillin, streptomycin, and Gibco Amphotericin B) (Gibco, Thermo Fisher Scientific). The cells were kept at 37 °C in 95% air/5% CO₂ in a humidified incubator.

Starter cultures of AM, BB, and BL were prepared. The cell pellets from starter cultures were collected by centrifugation at 3,000×g for 10 min, and then washed with the base media. The washed cell pellets were resuspended in the base media to a final OD₆₀₀ of approximately 1. These cultures were inoculated into individual culture tubes containing the base media supplemented with specific carbon sources (Glc + GlcNAc, mucin, or dietary fiber) to an initial OD₆₀₀ of 0.02. For monocultures, 60 μL of washed cultures was inoculated into 3 mL fresh media (50× dilution). For co-cultures, AM and either BB or BL were inoculated to an equal ratio (OD₆₀₀ of 0.01 each) by inoculating 30 μL of each strain into 3 mL fresh media. The cultures were incubated at 37 °C anaerobically, and cell growth was determined by monitoring the OD₆₀₀ every 3 h using BioTek Synergy H1 multimode reader (Agilent).

Absolute abundance could be calculated using OD₆₀₀ or CFU counting. While cellular traits such as cell adhesion, size, and shape can influence OD₆₀₀ measurements (Stevenson et al., 2016), CFU counting is biased by cell adhesion, dormant sub-populations, growth selection on solid vs. liquid media, and growth stage (Jansson and Prosser, 1997; Volkmer and Heinemann, 2011; Ou et al., 2017). To complement the measurement of growth profile (OD₆₀₀ using plate reader), we performed CFU counting to determine the absolute abundance of bacterial species. For pairwise communities, OD₆₀₀ measurement only provides information on the total growth of the co-culture, whereas CFU counting could distinguish the abundance of the two species using selective plates.

BHI agar (Thermo Scientific™ Oxoid™) was used for plating and counting the total CFU of all species in monocultures and co-cultures (total absolute abundance). Since AM is unable to grow in MRS, MRS agar plates were used as Bifidobacterium selective plates to calculate the abundance of BB and BL in the AM + BB and AM + BL co-cultures, respectively. The abundance of AM in the co-cultures was calculated by subtracting the total CFU obtained from the BHI agar from the CFU of either BB or BL obtained from the MRS agar. Three technical replicates were performed for each biological replicate during plating and CFU counting.

For proteomics analysis, monocultures and co-cultures of AM, BB, and BL were cultured anaerobically in the respective media for 24 h, since this growth period allows all species to reach maximum population size. For all proteomics experiments, three biological replicates were performed for each sample, including the control sample. The cell pellet was suspended in 300 μL of lysis buffer (8 M Urea, 50 mM Tris–HCl pH 8.0), frozen in liquid nitrogen, and sonicated for 5 min. The sample was centrifuged (16,000×g for 10 min) to remove cell debris and insoluble materials. An aliquot of the sample was taken for the BCA protein assay (Pierce™ BCA Protein Assay Kit). After protein quantification, the sample was reduced by dithiothreitol (DTT; 0.1 M final concentration) at 37 °C for 1 h. For shotgun proteomics, 150 μg of proteins were mixed with up to 250 μL of the exchange buffer (6 M Urea, 50 mM Tris–HCl pH 8.0, 600 mM guanidine HCl), transferred to an Amicon® filter device (Millipore, Darmstadt, Germany), and centrifuged (14,000×g for 20 min). The proteins in the filter device were alkylated with iodoacetamide (IAA, 50 mM in exchange buffer) in the dark for 20 min, and then centrifuged (14,000×g for 20 min). To reduce the urea concentration, 250 μL of 50 mM ammonium bicarbonate was added to the filter device and centrifuged (14,000×g for 20 min). This step was repeated once. Proteins were digested by sequencing-grade modified trypsin (1:50 w/w, Promega, Madison, WI) for 12 h at 37 °C. Then, the sample was acidified with 10% formic acid to a final concentration of 0.1% (v/v) and centrifuged at 16,000×g for 5 min. Finally, the samples were desalted by C18 reverse-phase ZipTip (Millipore, Darmstadt, Germany) and dried with SpeedVac (Eppendorf, Hamburg, Germany) for 30 min.

The samples were reconstituted in 25 μL water/acetonitrile/formic acid in a 97.9:2:0.1 ratio (v/v/v), and processed through Bruker nanoElute Ultra-High-Performance Liquid Chromatography (UHPLC; Bruker Daltonics, Bremen, Germany) coupled to a hybrid trapped ion mobility-quadrupole time-of-flight mass spectrometer (TimsTOF Pro, Bruker Daltonics, Bremen, Germany) via a nano-electrospray ion source (Captive Spray, Bruker Daltonics). A volume of 1 μL (approximately 200 ng of the protein digest) was injected into the UHPLC system and separated on an IonOpticks 25 cm Aurora Ultimate Series Emitter column with Captive Spray Insert (250 mm × 75 μm internal diameter, 120 Å pore size, 1.7 μm particle size C18) at a flow rate of 0.3 μL/min. The mobile phase composition is 0.1% formic acid in water for solvent A, and 0.1% formic acid in acetonitrile for solvent B. The gradient was applied from 2 to 5% of solvent B for 0.5 min, from 5 to 40% of solvent B for 26.5 min, and then from 40 to 95% of solvent B for 0.5 min. In the end, the mobile phase was kept at 95% of solvent B for 0.5 min, and then decreased to 2% of solvent B for 0.1 min. 2 min equilibration with 2% of solvent B was applied before the next injection.

A detailed description of the Bruker TimsTOF Pro mass spectrometer used in this work can be found in the literature (Meier et al., 2015; Meier et al., 2018). We set the accumulation and ramp time to 100 ms each and recorded mass spectra in the range from m/z 100 to 1,700 using the positive electrospray mode. The ion mobility was scanned from 0.85 to 1.30 Vs cm⁻². The quadrupole isolation width was set to 2 Th for m/z < 700 and 3 Th for m/z > 700, and the collision energy was linearly increased from 27 to 45 eV as a function of increasing ion mobility. The overall acquisition cycle of 0.53 s comprised one full TIMS-MS scan and four Parallel Accumulation-Serial Fragmentation (PASEF) MS/MS scans. Low-abundance precursor ions with an intensity above a threshold of 2,500 counts but below a target value of 20,000 counts were repeatedly scheduled and otherwise dynamically excluded for 0.4 min. The TIMS dimension was calibrated linearly using three selected ions from the Agilent ESI LC/MS tuning mix (m/z, 1/K₀: 622.0289, 0.9848 Vs cm⁻²; 922.0097, 1.1895 Vs cm⁻²; 1,221.9906, 1.3820 Vs cm⁻²) in positive mode.

The raw data were converted to mgf files by Bruker Compass DataAnalysis (version 5.2), and subsequently converted to mzML files by msconvert of the ProteoWizard [version 3.0.23233 (Kessner et al., 2008)]. The mzML files were searched using Comet [version 2024.01 rev.0 (Eng et al., 2013)] against the A. muciniphila BAA-835, B. breve JCM1192, or B. longum JCM1217 protein sequence database obtained from UniProt (downloaded December 2024). For co-culture samples, the protein sequence of all strains in the co-culture was combined into a single database. The sequences of common contaminants, such as trypsin and human keratins, and decoy sequences generated by shuffling amino acid sequences between tryptic cleavage sites were added to the database. The decoy sequences in the database are used for the false discovery rate (FDR) estimation of the identified peptides. The search parameters criteria were set as follows: 40 ppm peptide mass tolerance, monoisotopic mass type, fully digested enzyme termini, 0.02 amu fragment bin tolerance, 0 amu fragment bin offset, carbamidomethylated cysteine, and oxidized methionine as the fixed and variable modifications, respectively. The search results from Comet were processed by PeptideProphet (Keller et al., 2002), iProphet, and ProteinProphet of the Trans-Proteomics Pipeline [TPP (Deutsch et al., 2010)] in the decoy-assisted non-parametric mode. Every mzML run was analyzed independently. Protein identifications were filtered at a false discovery rate of 0.01 as predicted by ProteinProphet. The mass spectrometry proteomics data have been deposited to ProteomeXchange via the PRIDE repository with the dataset identifier PXD060954 (monoculture and co-culture growth of AM, BB, and BL in the presence of different carbon sources) and PXD061633 (IEC-6 cells co-cultured with AM, BL, or AM + BL in the presence and absence of AX).

Some of the identified peptide sequences in pairwise co-culture samples could match the proteome database of both species (shared peptides), and the probability would be higher for short peptide sequences. It is not possible to determine which of the two species (or both) these shared peptides belong to. To know what peptide sequence could be shared between AM and BB or AM and BL in co-culture, we performed in silico tryptic digestion of the whole proteome of AM, BB, and BL using the Uniprot database (parameters: minimum residue count 5, max missed cleavage 2, minimum fragment mass 400 Da, maximum fragment mass 7,000 Da), and generated a list of all possible shared peptides between AM and BB or AM and BL using an in-house python script. The percentage of theoretically shared peptides across the proteome of AM and BB or AM and BL (0.222 and 0.218%, respectively) is much smaller than that of BB and BL (14.8%). Most of the shared peptides between AM and BB or BL are short peptides (~5 amino acids, which is the threshold we used in our Comet database searching) (Supplementary Figure S4).

We then searched whether the peptides identified in the co-culture proteomic samples contain any of the shared peptides. There is a very low number of shared peptides detected across our 21 proteomic datasets, ranging from 13 to 19 in each sample. To deal with the shared peptides, we took a conservative approach by assigning the spectra of that peptide to both species in the co-culture. The TPP algorithm will only consider peptides with enough evidence of sibling peptides identification and remove unconfident peptide assignments with a low number of sibling peptides.

To analyze the proteomics data, we first compared the proteome profile of the samples between different carbon sources to observe how resource environments impact protein expression. Next, we compared the proteome profile of each species when cultured in monoculture versus co-culture under the same carbon source to reveal the effect of the partner species on protein expression of a particular strain.

The proteins identified in at least two out of three biological replicates were used for label-free quantification by spectral counting. Only proteins with average spectral counts (across all runs) of at least three were considered for quantification. The quantification of proteins was given by the normalized spectral abundance factor [NSAF (Paoletti et al., 2006)], where the number of peptide-spectrum matches (PSMs) for each protein divided by the length of the corresponding protein is normalized to the total number of PSMs divided by the length of protein for all identified proteins. Student’s t-test was employed on the NSAF values to detect differential expression between the two time points. Benjamini–Hochberg (BH) multiple testing correction was applied to the t-test p values to control the false-discovery rate (FDR) at 10% (Benjamini and Hochberg, 1995). To further reduce false discoveries and limit our attention to the more highly regulated proteins, only proteins with fold changes higher or lower than ±1.5-fold were considered differentially expressed in our subsequent analysis.

We visualize our proteomic data using principal component analysis (PCA) of the NSAF values using the PCA function from the sklearn package with centering and scaling in Python (Supplementary Figure S3). We added 95% confidence intervals by calculating correlation matrices for the three replicates of each sample and then adding these intervals to our plot using the matplotlib package in Python. To compare the protein expression profiles of a particular strain across different samples, we generated heatmaps of NSAF values of the proteins identified across all samples (Supplementary Figure S2). STRING version 12.0 (Szklarczyk et al., 2016) was used to predict the protein–protein interactions and to visualize the interactions among the differentially expressed proteins.

IEC-6 cells were cultured in transwells with 0.4 μm pore size inserts (SPL Life Sciences) in DMEM medium until fully differentiated. The transepithelial permeability assay was conducted as previously reported (Gildea et al., 2017; Fan et al., 2021; Fan et al., 2020). The integrity of the cell monolayer was determined by measuring the TEER using a Millicell ERS-2 voltohmmeter (Merck Millipore) every 24 h. The medium was changed every 2 days until the TEER value reached ~50 Ω cm². The cells in the inserts were washed twice with phosphate-buffered saline (PBS) to remove residual DMEM medium.

Co-culturing bacteria with IEC-6 cells was performed using a previously reported protocol (Reunanen et al., 2015). A previous study reported that there is no growth difference in AM under oxic and anaerobic conditions (Reunanen et al., 2015), whereas BL is a facultative anaerobe and thus could tolerate oxic conditions. Around 10⁷ CFU of AM, BL, or AM + BL (mixed in equal ratio of 5 × 10⁶ CFU each) were resuspended in either 1 mL of Hanks’ Balanced Salt Solution (HBSS) (Gibco, Thermo Fisher Scientific) or HBSS supplemented with 1% arabinoxylan. This bacterial seeding density is suitable for binding with the IEC-6 cells (Shi et al., 2022). The bacterial suspension was added to the apical side of the IEC-6 monolayer, whereas either HBSS or HBSS with arabinoxylan was added to the basolateral side. The IEC-6 cells and bacteria co-culture were incubated for 6 h at 37 °C in oxic conditions. Then, the bacterial suspension was removed from the IEC-6 cells, and the inserts were washed twice with fresh HBSS. For the lucifer yellow assay, 900 μL fresh HBSS was added to the basolateral side, and 200 μL of lucifer yellow dye (Sigma Aldrich) was added to the apical side of the IEC-6 cells and incubated for 2 h at 37 °C in oxic conditions. Transepithelial permeability was assessed by quantifying the % of dye that passed through the monolayer by fluorescence measurement of the basolateral side (485 nm excitation and 535 nm emission) and comparing the values with lucifer yellow standard with known concentration values.