Global Growth Phase Response of the Gut Bacterium Phocaeicola vulgatus (Phylum Bacteroidota)

Within the microbiome of the human colon, members of the genus Bacteroides have long been recognized as the most predominant anaerobes [e.g., 1, 2]. Their capacity to fermentatively decompose complex food-derived polysaccharides via a plethora of enzymes to valuable end products such as short chain fatty acids (SCFAs), e.g., acetate or succinate, suggested them as natural commensals aiding the host’s nutrition and well-being [3‒6]. However, the role of Bacteroides in the gut ecosystems goes beyond provision of fermentation end products and is likely multilayered, including a probiotic role as well as association with inflammatory bowl and other diseases [7‒9]. Moreover, Bacteroides contribute to complex metabolic interactions with the host, e.g., the maturation of the gut metabolome of infants [10] and the potential influence on chemoradiotherapy of rectal cancer via their nucleoside synthesis [11]. Next to these diverse biomedical roles, Bacteroides have more recently attracted attention in the area of sustainable chemical industry for their potential to produce bulk chemicals such as SCFAs from renewable instead of fossil starting materials [e.g., 12, 13]. For example, succinate is used in large quantities as an intermediate chemical for the production of a wide range of commodity chemicals (e.g., 1,4-butanediol as solvent and raw material for plastics) and speciality chemicals (e.g., feed additives) [14].

Phocaeicola vulgatus (basonym Bacteroides vulgatus [15]) is a well-known constituent of the human colon microbiome, with members of the genus Phocaeicola isolated from different human and animal feces [16]. P. vulgatus has been implicated in a variety of interactions with human health [e.g., 17] and with other gut microbiome members [e.g., 18, 19]. Most recently, application of a CRISPR-based approach to reduce the abundance of a given strain from a defined 13-membered Bacteroides-strain community as well as gene mutagenesis revealed intricate relationships between P. vulgatus and other community members [20]. Several recent studies aimed at gaining a deeper understanding into the basic physiology of P. vulgatus. For example, Pudlo et al. [21] investigated the potential of almost 3 dozen strains for their nutritional capacities across a broad range of carbon sources. Clausen et al. [22] studied the dynamics of growth stoichiometry, fermentation network, uptake systems, and respiratory potential of P. vulgatus across 14 different carbohydrates. Moreover, expansion of the toolbox for generating markerless gene deletions in P. vulgatus recently improved its genetic tractability [23]. Only few bioreactor studies investigated the potential of P. vulgatus for biotechnological applications so far. Cultivation in 0.5 L batch bioreactors at 15°C fed with cattle slurry investigated H₂ production [24], a research area that will benefit from a recently reported online H₂ measurement approach [25]. Other recent bioreactor-based studies with P. vulgatus focused on the effect of CO₂ and O₂ traces [26] as well as pH, acids, and buffer concentrations [27] on growth and product formation.

This study integrates proteomic and metabolomic profiling to elucidate the growth phase-dependent global response of P. vulgatus during anaerobic growth with glucose in process-controlled bioreactors. In particular, we focused on the dynamics of the fermentation stoichiometry and network, the stress response, and the complexity of the exometabolome that forms the molecular background for few fermentation products.

P. vulgatus ATCC 8482 (DSM 1447) [28] was obtained from the German Collection of Microorganisms and Cell Cultures (DSMZ; Braunschweig, Germany). The anaerobic cultivation was performed in a defined minimal medium as previously described [29, 30], with the following special features: Na₂S₂O₃ (5 mm) and Na₂S (6 mm) were added as sulfur source and reductant, respectively, instead of l-cysteine. Glucose (15 mm) was provided as sole source of carbon and energy. Pre-cultivation via 5 passages in 500-mL flasks was carried out as recently described [22]. The sterility of the medium was controlled by 2 flasks cultivated in parallel to each bioreactor run: the first with inoculum but without substrate, and the second, without inoculum but with substrate. For purity control of the used cultures, cells were streaked on YM-agar and Nutrient Agar plates at the beginning and the ending of every experiment. Used chemicals were of analytical grade.

For both types of experiments, i.e., stoichiometry versus harvesting for omics samples, the same bioreactor process design was used. It was modified from a previously described setup [31] for anaerobic cultivation as described in the following. Four 2-L double-jacket glass vessels (Labfors 5; Infors AG, Bottmingen, Switzerland) with working volumes of 1.5 L were operated in parallel via a touchscreen including the software Multifors 2 (version 3.0; Infors AG). The same mineral medium as described above was used. Optimal mixing of the culture broth with minimal shear forces was achieved by applying two Rushton turbine stirrers and one marine propeller rotated at 200 rpm. The head space had an atmosphere of N₂:CO₂ (90:10 [vol/vol]). Anoxic conditions were monitored using an O₂ sensor (InPro 6800 Series O₂ Sensor 12/25 mm; Mettler-Toledo AG, Urdorf, Switzerland) and further ensured by constant streaming of N₂/CO₂ at a flow rate of 1 L/h via a drilled-hole sparger at the bottom of each bioreactor. Temperature was continuously adjusted to 37°C (Pt-100; Infors HT) and the pH at 7.1 ± 0.1 (easyFerm Plus PHI Arc 225, pH/OrP Arc Sensor; Hamilton, Bonaduz, Switzerland). Aforementioned process parameters were continuously controlled via the process control system eve (version Q2 2019; Infors AG).

Retrieval of samples for various types of analysis relied on sterile, N₂/CO₂-flushed syringes to avoid contamination and to maintain anoxia. While in the case of stoichiometric analyses multiple samples were retrieved during the course of growth, sampling for proteomic and metabolite/exometabolome analyses was confined to ½OD_max, OD_max, and OD_stat. In all of these cases, 4 individual bioreactor runs provided the four biological replicates.

For stoichiometric analysis and monitoring of growth, several types of samples were retrieved at varying intervals from the bioreactors. Close sampling was conducted for OD measurements and HPLC-based determination of fermentation products. For this purpose, 3 mL samples were retrieved, of which 1 mL was directly used to measure OD, and the remaining 2 mL were immediately centrifuged (20,800 g, 10 min, 4°C) and the resultant supernatant was transferred into a new reaction tube and stored at −20°C until further analysis. Coarse sampling was applied for the determination of CDW and cellular ATP-content. In the case of CDW, 2 mL culture broth were retrieved. In the case of ATP-content, a 8 mL sample was put into 15-mL screw-cape tube (Sarstedt AG & Co. KG, Nümbrecht, Germany), from which 4 aliquots á 1.8 mL (technical replicates) were immediately transferred each into a 2-mL reaction tube using clean and sterile pipette tips and then centrifuged (20,000 g, 5 min, 14°C). After discarding the supernatant, the remaining pellet was placed on ice and covered with 300 μL ice cold quenching buffer (10 mL of 0.5 m Trizma-Base, 200 μL of 0.5 m EDTA, 2.5 mL of 20% [wt/vol] DTAB; ad 40 mL H₂O_mq) and immediately vortexed until complete solution. The sample was then shock frozen in liquid nitrogen and stored at −80°C until further analyses.

For proteomic analysis, 200 mL samples were centrifuged (14,300 g, 30 min, 4°C) and the remaining pellets were resuspended in 100 mL washing buffer (100 mm Tris/HCl, 5 mm MgCl₂ × 6 H₂O, pH 7.5). After anew centrifugation, the pellets were resuspended in 1 mL washing buffer and transferred to micro reaction tubes, which were then again centrifuged (20,800 g, 10 min, 4°C). The resultant pellets were shock frozen in liquid nitrogen and stored at −80°C until further analyses.

For targeted metabolite and global exometabolome analyses, strict anoxia was maintained including the use of gas impermeable tubes. Per sampling time point, 50 mL culture broth were centrifuged (12,000 g, 10 min, 4°C). From the supernatant, 2 mL were transferred into a micro reaction tube, immediately shock frozen in liquid nitrogen and stored at −80°C until targeted exometabolite analysis. The remaining supernatant was transferred into a separate tube and stored at −20°C until global exometabolome analysis. The cell pellet was resuspended in 1.5 mL of the sterile, defined minimal medium, transferred in a pre-weighted reaction tube, and centrifuged (12,000 g, 10 min, 4°C). Then, the tube was weighted, shock frozen in liquid nitrogen, and stored at −80°C until targeted endometabolite analyses.

OD was measured at 660 nm using a UVmini-1240 spectrophotometer (Shimadzu, Duisburg, Germany). For CDW determination, each sample was centrifuged (25,200 g, 10 min, 4°C) and the remaining pellet was washed in 5 mL of 50 mm ammonium acetate-buffer. After anew centrifugation, the pellet was resuspended in 300 μL of the same buffer, transferred into a pre-weighted tube, and incubated at 60°C until constant weight.

After thawing, samples were diluted 1:1 with water (membrane-purified) and filtered (0.2 μm pore size, regenerated cellulose; Chroma Globe, Kreuzau, Germany). Depletion of the growth substrate glucose and formation of the fermentation products propanoate, formate, succinate, and acetate were determined via HPLC analysis using an Ultimate 3,000 system (Thermo Fisher Scientific, Germering, Bavaria, Germany). Separation was achieved with a Phenomenex Rezex-ROA-Organic Acid++(8%) column (300 × 7.8 mm, 10-μm bead size; Knauer, Berlin, Germany), operated at a column temperature of 20°C and with 0.5 mm H₂SO₄ as eluent administered at a flow rate of 0.5 mL/min. Detection was achieved with a RI-detector (Shodex RI-501; Showa Denko GmbH, München, Germany), except for fumarate (UV detector DAD-3000; Thermo Fisher Scientific). The retention times were (alphabetical order): acetate at 23.7 min, formate at 22.0 min, glucose at 14.7 min, propanoate at 27.8 min, and succinate at 20.0 min. For all analytes, the detection limit was at 25 μm and the dynamic range extended to 10 mm. For each time point, three biological replicates with two technical replicates each were measured.

The ATP concentration was determined using the BacTiter-Glo™ Assay (Cat. #G8231; Promega, Fitchburg, WI, USA) essentially as recently described [32]. After thawing at room temperature, the quenched samples were mixed in a dilution buffer (10 mL of 0.5 m Trizma-Base, 200 μL of 0.5 m EDTA; ad 50 mL H₂O_mq). The level of dilution depended on the predetermined ATP concentration and ranged from 1:10 to 1:200. Twenty-five µL of each diluted sample as well as a calibration series of 0.1–1,000 nm ATP (Cat. #R0441; Thermo Fisher Scientific, Waltham, MA, USA) were applied to a 96-multiwell plate, immediately followed by addition of 25 μL BacTiter-Glo™ reagent to each well. Then, the plate was inserted into the CLARIOstar^® Plus reader (BMG Labtech, Ortenberg, Germany) and incubated (5 min, 25°C) under orbital shaking (10 s, 400 rpm) prior to luminescence measurement at 545–550 nm.

For optimal proteomic coverage, the soluble and membrane protein-enriched fractions of all analyzed samples were differentiated. For this purpose, the thawed cell pellets were resuspended in 500 μL of shotgun lysis buffer (7 m urea, 2 m thiourea, 30 mm Tris/HCl, pH 8.5) and transferred in matrix tubes filled with 0.25 g of an even mixture of 0.1- and 1-mm zirconium beads. Then, cell breakage was achieved by 3 rounds at 6 m/s for 30 s (Fast-Prep-24 5 G; MP Biomedical, Eschwege, Germany) with cooling for 90 s on ice between the beating. The disrupted cells were then twice ultra-centrifuged (104,000 g, 60 min, 10°C). The supernatants including the soluble proteins were shock frozen in liquid nitrogen and stored at −80°C until further analysis. For preparation of the membrane protein-enriched fraction, the obtained pellets were resuspended in 300 μL of membrane lysis buffer (MLB; 100 mm Tris/HCl, pH 7.5, 2 mm MgCl₂ × 6 H₂O, 10% (wt/vol) glycerol, 0.5 mm DTT) and treated with 35 mL sodium carbonate solution (100 mm) by stirring for 1 h on ice. The solution was then ultra-centrifuged (200,000 g, 60 min, 4°C) followed by resuspending the resultant pellet in 400 μL MLB and anew ultra-centrifugation. The remaining pellet was resuspended in 150 μL 1% (wt/vol) SDS and incubated at 95°C for 10 min with shaking (600 rpm), yielding the membrane protein-enriched fraction. This was immediately shock frozen and stored at −80°C until further analyses. The protein concentrations were determined according to Bradford [33] for the soluble proteins and with a RCDC protein assay (Bio-Rad Laboratories, Munich, Germany) for membrane protein-enriched fractions [34].

The preparatory steps for shotgun proteomic analyses of the two subcellular fractions were as followed. In the case of the soluble protein fraction, 50 μg protein were diluted with urea buffer (8 mm urea, 0.4 mm NH₄HCO₃) to a total volume of 50 μL and 45 mm DTT was added prior to an incubation at 55°C for 30 min in the dark (Thermomixer comfort; Eppendorf AG, Hamburg, Germany). Then, 100 mm iodoacetamide were added and the incubation was repeated under the same conditions for 15 min. After adding 143 μL water (HPLC grade) and 1 μg trypsin (Serva, Heidelberg, Germany), overnight incubation was conducted at 37°C, generating the final peptide mixtures. In the case of the membrane protein-enriched fraction, at first decomplexing was achieved by 1D SDS-PAGE (12% SDS mini gels, 10 × 7 cm; Bio-Rad) with a 10 μg protein load per lane. After electrophoresis, gels were stained with Coomassie Brilliant Blue and divided into 7 lanes, each of which was then cut in ∼1 mm³ pieces. Tryptic in gel digestion was conducted as previously described [35]. The generated peptide mixture of both subcellular fractions were shock frozen in liquid nitrogen and stored at −80°C until the MS analyses.

For the MS analyses, the tryptic peptide mixtures (three replicates per fraction) were firstly separated by nanoLC (Ultimate 3000 nanoRSLC; Thermo Fisher Scientific) with the following applied setup: trap column (C₁₈, 2 cm × 100 μm, 5-µm bead size; Thermo Fisher Scientific), separation column (C₁₈, 25 cm × 75 μm, 2-µm bead size; Thermo Fisher Scientific), and 280-min (shotgun) versus 90-min (membrane, per gel fraction) linear acetonitrile (0–80% (vol/vol)) gradients. The analysis was then performed by an online-coupled 3D ion trap mass spectrometer (amaZon speed ETD; Bruker Daltonics GmbH, Bremen, Germany) with the settings as described by Zech et al. [34]. For the identification of the proteins, the ProteinScape platform (version 3.1; Bruker Daltonik GmbH), an in-house Mascot server (version 2.3; Matrix Science Ltd., London, UK), and the genome of P. vulgatus [36] were used. For the used target decoy strategy, the settings were applied as described in [37].

Methanol (100 μL per 20 mg wet weight) was added to the frozen samples, which were then resuspended using an ultrasonic bath for 5 min at room temperature followed by a mixing step using a vortex. Then, 500 μL of the resuspended samples with 100 μL methanol containing 4% ribitol stock solution (0.2 g/L in water), cell lysis was achieved in an ultrasonic bath (15 min, room temperature). Then, 500 μL of water was added and vigorously mixed for 1 min using a vortex, followed by an addition of 500 μL chloroform and a further mixing step for 1 min. The resultant mixture was centrifuged (10,000 g, 5 min, 4°C). After a transfer of the polar phase into a 2-mL reaction tube, 100 μL of the polar phase were dried under a vacuum for endometabolomic analysis. For exometabolomic analysis, 100 μL of the respective samples were spiked with 100 μL methanol containing 4% ribitol stock solution (0.2 g/L in water) and dried under vacuum. All samples for polar metabolites were derivatized and analyzed by GC-MS (7890B mass spectrometer coupled to a 5977 GC; Agilent Technologies, Waldbronn, Germany) as described previously [38]. The software metabolite detector processed the gained data [39]. The results were normalized against the ribitol standard. The CoA-thioesters were identified via LC-MS/MS (6545 qTOF mass spectrometer coupled to a 1290 series HPLC, Agilent Technologies). The abovementioned resuspended samples were treated and analyzed as described previously [40].

Fourier-transform ion cyclotron resonance mass spectrometry (FT-ICR-MS) was applied to study the molecular composition of the P. vulgatus exometabolome on a global, untargeted level. Again, the growth states ½OD_max, OD_max, and OD_stat were analyzed, in all three cases based on four biological replicates each with two technical ones. Cell-free culture supernatants were initially acidified to pH 2 by adding ∼500 μL 25% (vol/vol) HCl (analytical grade). Then, remaining sulfide (from Na₂S reductant) was removed by bubbling each sample with N₂ for ∼15 min at room temperature. Subsequently, samples were filtered through 0.7-µm and 0.2-µm pore sized PES membranes (Minisart; Sartorius, Göttingen, Germany), subjected to solid-phase extraction using styrene divinylbenzene copolymer cartridges (100 mg PPL; Agilent, Santa Clara, CA, USA) modified as described by Dittmar et al. [41], and stored at −18°C.

Prior to FT-ICR-MS analysis, samples were diluted by a 100-fold in a 1:1 (vol/vol) mixture of methanol and ultrapure water. Then, direct infusion electrospray FT-ICR mass spectra were acquired in negative mode on a 15 T Solarix XR FT-ICR-MS (Bruker Daltonics, Billerica, MA, USA) using an autosampler (CTC Analytics AG, Zwingen, Switzerland). The ESI source temperature was 200°C, capillary voltage was 4.5 kV, and the hexapole accumulation time was 0.65 ms in negative mode. The flow rate was 360 μL/h with a nebulizer gas pressure of 14.5 psi and accumulated mass spectra with 200 scans per sample within a mass window ranging from 92 to 1,000 m/z. The spectra were calibrated as described in Riedel and Dittmar [42] using an internal calibration list within the Bruker Daltonics Data Analysis software package resulting in a mass error <0.1 ppm. Then, the mass lists were recalibrated using the publicly available software ICBM-OCEAN [43]. The implemented method detection limit (ICBM-OCEAN) accounts for signal noise and reduces systematic error to increase the precision of formula attribution. Attribution of masses with maximum abundances of C_nH_nO_nN₇S₂P₁ were allowed. The “Likeliest matches” dataset generated in ICBM-OCEAN was used for further analysis. All peaks that occurred in the medium blanks were considered as contaminants and removed from the dataset. All masses that occurred in at least two replicates within a given growth phase were considered. We used anvi’o [44] to perform a hierarchical clustering analysis of the intensity values of each formula using Euclidian distance and Ward linkage algorithm. Data analysis and graphical representation were carried out using R Studio Version 2023.06.1 + 524 and Anvi’o version 8.

The calculation was based on the consumed or generated ATP molecules as well as reducing equivalents of the glucose fermentation pathway (Fig. 3), essentially as recently described by Clausen et al. [22]. The proportion of use of the product forming pathways was inferred from the ratio of fermentation products as determined by HPLC analyses (online suppl. Table. S1; for all online suppl. material, see https://doi.org/10.1159/000538914). Excess reducing equivalents (not consumed during fermentation) but feeding respiratory energy conservation were included as previously described by Clausen et al. [22]. From this calculated ATP gain, the ATP requirement for biomass formation was subtracted. For this purpose, the determined CDW and the elemental composition (C₁H_1.79O_0.44) of a closely related bacterium, Segatella copri (formerly Prevotella copri) as described by Franke and Deppenmeier [29], were considered.

To visualize the localization of the genes within the genome of P. vulgatus, the software Artemis (version 18.2.0) [45] was used. For the analyses of the monomer biosynthesis of the anabolic modules, the pathway predictions from KEGG [46] and BioCyc [47] were used for P. vulgatus and E. coli. If incomplete or no information about a given biosynthesis pathway for P. vulgatus was provided by the mentioned databases, the predicted pathways for E. coli were analyzed and equivalent proteins were manually searched in the genome of P. vulgatus. Pairwise alignment of the UspA proteins from P. vulgatus, B. thetaiotaomicron, and E. coli was conducted using “Global alignment, Needle (EMBOSS)” [48]. The illustration of the heatmaps containing the peptide counts of the specific proteins were performed with MetaboMaps [49].

To determine the growth stoichiometry of P. vulgatus with glucose as sole source of carbon and energy, we operated 4 process-controlled bioreactors in parallel. Growth curve and substrate/product profiles are illustrated in Figure 1 and growth parameters summarized in Table 1.

P. vulgatus reached an OD_max of 1.8 after 8 h of incubation (Fig. 1a), coinciding with a ∼98% depletion of the initially provided 14.29 mm glucose. The glucose concentration dropped below the detection limit after another hour of incubation, paralleled by stable relative shares of the fermentation products: 2.23 mm formate, 8.71 mm acetate, 0.39 mm propanoate, and 5.25 mm succinate (Fig. 1b). During growth, P. vulgatus reached a doubling time of 1.74 h and a growth rate of 0.13 g/(L × h). At ¼OD_max, the calculated ATP yield and measured cellular ATP-content were considerably lower (1.35 mmol_ATP/mmol_S and 0.92 × 10⁻⁹ mmol_ATP/g_CDW) compared to the ½OD_max ‒ OD_max range (2.58–3.00 mmol_ATP/mmol_S and 1.85–1.92 × 10⁻⁹ mmol_ATP/g_CDW). This suggests that P. vulgatus devotes a higher share of consumed glucose for anabolic purposes or operates at an overall lower metabolic efficiency due to adaptation processes during the early stage of growth compared to the later ones. As evident from the aforementioned fermentation product shares, highest yields were observed for acetate and succinate with 0.61 and 0.36 mmol_P/mmol_S, respectively. Repeated cultivation in bioreactors for omics analyses (online suppl. Fig. S1) was highly reproducible with respect to the course of growth (green dots in Fig. 1a) and the fermentation product yields (Table 1).

The here-determined bioreactor-based growth parameters of P. vulgatus are in parts rather similar to the ones previously reported for its growth in 500-mL flasks (400 mL culture volume), e.g., an identical biomass yield of 0.06 g/mmoL_Glc (Table 1; [22]). However, the calculated ATP yields were considerably higher in the bioreactors (2.60 vs. 2.06 mmol_ATP/mmol_S), which agrees with a higher acetate yield (0.61 vs. 0.51 mmol_P/mmol_S) and higher apparent abundance of pyruvate-flavodoxin oxidoreductase Pfo (69 vs. 17 peptide counts at ½OD_max) reflecting a higher degree of substrate level phosphorylation attainable for P. vulgatus during cultivation in bioreactors. A recent flask-based study (250-mL flasks with 50 mL culture volume) that focused on online monitoring of CO₂ and O₂ and used a different type of mineral medium yielded essentially the same fermentation product spectrum, i.e., predominated by succinate followed by acetate [26]. Thus, growth stoichiometries based on flask versus bioreactor cultivation are partially comparable. However, bioreactor-based studies need to consider the effect of general cultivation conditions (e.g., temperature, pH, and stirring) on product formation, as previously reported for Bacteroides fragilis [50].

The generated proteomic dataset is illustrated in Figure 2. The genes encoding the 784 identified proteins are scattered over the chromosome of P. vulgatus (Fig. 2a) and represent 19.5% of the predicted proteome (Fig. 2b, left panel). Notably, 44.8% of the identified proteins had no clearly assigned function, which is even surpassed by the 65.0% of the predicted proteome without a functional prediction (Fig. 2b, left panel) [36]. This high proportion of genes with unclear functional assignment hampers interpretation of proteomic data and is contrasted by the considerably higher share of proteins with assigned function (at least 86%) in the well-studied standard organism Escherichia coli [37, 51]. Identified proteins are associated with the investigated metabolic modules (glucose fermentation, monomer biosynthesis, and stress response) account for 17.3% of the proteomic dataset (Fig. 2b, right panel). The principal component analysis displayed in Figure 2c shows clear clustering (distinction of states) of the proteomes determined per studied growth phase. The Venn diagram (Fig. 2d) shows the growth phase-specific versus -shared subsets of identified proteins and highlights the proteome dynamics when P. vulgatus progresses from active to arrested growth.

The pathway of glucose fermentation via glycolysis to the products acetate, formate, succinate, and propanoate is displayed in Figure 3, representing the recent reconstruction by Clausen et al. [22]. In addition, the profiles of glucose depletion, fermentation product formation, as well as abundances of involved enzymes and metabolites across the studied growth phases are overlaid. This pathway is composed of 24 proteins and 24 metabolites, 23 and 16, respectively, of which could be identified (95.8% and 66.6% coverage).

Three divergently dynamic sections of the fermentation network became evident. First, the profiles of glucose-6-phosphate and fructose-6-phosphate (F-6-P) parallel the rapid decrease of glucose from ½OD_max to OD_max, while those of the enzymes constituting the upper part of glycolysis remain fairly stable. Second, metabolites of the lower part of glycolysis consistently show highest abundances at OD_max, reflected also in the abundance profiles of triosephosphate isomerase, phosphoglycerate kinase, and enolase. Third, in the product-specific fermentation branches, metabolite abundance was highest at OD_stat, whereas the abundance profiles of involved enzymes (expect for Pfo) were more balanced. Thus, the protein and metabolite profiles overall are in accord with the bulk formation of fermentation products occurring from ½OD_max onwards. Such a shift in protein abundance from early steps of glucose conversion to the final steps of fermentation in response to progression through growth stages has also been documented for Lactobacillus rhamnosus GG cultivated in batch bioreactors [52].

Since monomer biosynthesis for the predominant cellular biomacromolecules is generally essential for growth, we inspected the dynamics of the anabolic pathways for amino acids, nucleotides, and fatty acids (Fig. 4a–c; online suppl. Table S2). We retrieved the individual pathways from BioCyc [47] and manually closed remaining gaps as far as possible. According to current knowledge from E. coli, the studied anabolic network should be encoded by 216 non-paralogous genes. Here, 190 genes of P. vulgatus could be assigned including 39 paralogs, leaving 65 biosynthetic steps related to E. coli unaccounted. Nevertheless, the anabolic pathways of P. vulgatus considered in the present study could be completely reconstructed from the genome (N-acetylglutamate synthase as sole exception) and covered in total by 68% with the generated proteomic data (Fig. 4d). For the sake of simplicity, we selected the enzyme with the highest abundance in the generated proteomic dataset for each metabolic pathway to display abundance profiles across the studied growth phases (Fig. 4a‒c).

In the case of amino acid syntheses (Fig. 4a), the detection of most proteinogenic amino acids (except for arginine, histidine, and lysine) by targeted metabolomics indicates that P. vulgatus should have the capacity for their synthesis. This is also evident from the growth media composition that was devoid of any amino acids. While the abundance pattern of the selected proteins was not uniform across the studied growth phases, the detected amino acids displayed a clear abundance maximum at OD_max. The subsequent decrease of amino acids reflects their strongly reduced requirement in the absence of net-growth. The most striking exceptions were tryptophan and methionine, showing continuously increasing intracellular abundance until OD_stat. A reasonable explanation could be their involvement in the synthesis of quorum sensing (QS) signaling compounds, e.g., tryptophan-derived indole derivatives [53, 54] and methionine-derived autoinducers-2/3 (AIs-2/3) [55, 56], see below section QS. The likewise high level of proline at OD_stat may reflect its additional role in protein stabilization and stress resistance known for many types of organisms [57, 58]. In the well-known gut bacteria E. coli and Salmonella typhimurium proline also acts as an osmo-protectant [59]. Since P. vulgatus lacks the put uptake system available for E. coli and S. typhimurium, the observed proline accumulation may compensate for the incapability of its uptake.

In the case of nucleotide synthesis (Fig. 4b), the pathways for 5-aminoimidazole ribonucleotide biosynthesis II and the pyrimidine deoxyribonucleotide de novo synthesis could be predicted only partially. All other biosynthetic routes could be completely reconstructed. Overall, most enzymes were detected with low peptide counts only. Proteins with high enough abundances for profiling peaked at ½OD_max agreeing with the lower demand of DNA/RNA synthesis upon entering the stationary growth phase. A conspicuous exception was orotate phosphoribosyl transferase (PyrE), the abundances of which consistently increased from ½OD_max to OD_stat, contrasted by CarB likewise involved in UMP biosynthesis II. The PyrE protein was previously indicated in biofilm formation of the human oral microbiome member Streptococcus sanguinis [60]. Accordingly, Bechón et al. [61] proposed that the prominent gut bacterium Bacteroides thetaiotaomicron uses biofilm formation as stress response. One may speculate that the PyrE protein may serve a similar function also in P. vulgatus.

In the case of fatty acid and cell wall synthesis (Fig. 4c), the expected pathways could only be partially reconstructed (<50%) compared to the gene repertoire of E. coli and only a few proteins involved were detected, often serving multiple synthetic pathways. Two divergent abundance profiles were observed. First, 3-hydroxy [acp] dehydratase (FabZ) showed highest abundances at ½OD_max, markedly decreasing thereafter. Second, the 3-oxoacyl-ACP reductase (FabG) displayed continuously increasing abundances from ½OD_max until OD_stat. It should also be considered that the FabZ and FabG proteins are not only involved in fatty acid synthesis, but also play a role in QS [62, 63], see below section QS.

Transition of P. vulgatus into stationary growth phase was tightly correlated with the depletion of the sole substrate glucose (Fig. 1a), leading to a starvation condition and therefrom-arising stress. To mitigate the latter, P. vulgatus harbors 106 genes for stress-related proteins, 27 of which could be identified in the present study (Fig. 5a, b). Marked abundance increases at OD_max and OD_stat were observed for proteins belonging to the functional categories repair, oxidative stress, and universal stress response (Fig. 5c); all identified proteins assigned to stress response are compiled in online supplementary Table S3.

Disruption of the cellular proteome homeostasis by stress conditions is generally countered by an arsenal of repair enzymes [64], some of which were observed with P. vulgatus. The complex chaperonin GroEL and its co-chaperonin GroES, which generally assist in protein folding [65], were observed at all studied time points showing slight abundance increases toward OD_stat. By contrast, the more simply structured chaperonin DnaK (Hsp70), which remodels misfolded proteins [66], showed a marked abundance increase at OD_max and OD_stat. The latter was also observed for the OmpH protein, which was previously implicated in assisting extracytoplasmic folding of proteins [67]. Moreover, abundance of ClpB increased toward OD_stat; ClpB is an oligomeric AAA+ chaperone that acts in cooperation with DnaK in protein disaggregation [68]. Thus, P. vulgatus apparently is capable of coping with the increasing demand for protein remodeling during starvation. Despite its anaerobic lifestyle, P. vulgatus is capable of microaerotolerant growth [18] and was recently shown to form an O₂-reducing cytochrome oxidase [22]. Furthermore, P. vulgatus is likely to be exposed to oxygen via the host’s epithelial cells in its native gut environment [69], necessitating that the bacterium be capable of responding to oxidative stress. Proteins related to this function showed clear abundance increases toward OD_max/OD_stat: (i) the iron-containing ruberythrin (Rbr), which is well known to scavenge, e.g., exogenously produced H₂O₂ in aerobic bacteria [70]; (ii) the iron storage protein ferritin A, which is known in E. coli to scavenge iron from Fe-S clusters upon their disruption due to H₂O₂ stress [71]; and (iii) the thiol peroxidase Tpx, which in E. coli reduces alkyl hydroperoxides rather than H₂O₂ [72].

Growth arrest due to carbon or other nutrient starvation requires downregulation of the protein synthesizing machinery as well as upregulation of the stationary growth response. In E. coli, this function is performed by the universal stress protein (UspA) [73]; notably, expression of the uspA gene is metabolically controlled by the pool size of F-6-P already prior to entry into stationary phase [74]. In accord, the abundance of the UspA protein in P. vulgatus already increases at OD_max, when F-6-P is still readily detectable, and reaches highest levels at OD_stat. Since, however, the sequence similarity between UspA from P. vulgatus and other Bacteroides species compared to that of E. coli is rather low (23.3%), regulatory/mechanistic differences may have to be considered.

Besides the aforementioned stress proteins, several metabolites with concurrent growth phase-dependent abundance profiles and likewise association with nutrient deprivation and stress response [e.g., 75] were detected in P. vulgatus in the present study (Fig. 5d). While the biosynthetic pathway for the polyamine putrescine in P. vulgatus is currently unclear, a strong increase toward OD_stat was observed. Polyamines are known in E. coli to influence gene expression related to stationary phase and stress responses [76]. Similarly notable but unclear, is the accumulation of ethanolamine by P. vulgatus. Furthermore, several homoserine lactone derivatives increased in abundance toward OD_stat. These molecules have been suggested to serve as universal signals in response to starvation [77], in the context of antibiotic resistance [78], or QS (see following section).

QS describes inter-bacterial communication via chemical signals, so-called autoinducers, in response to reaching cell density thresholds [79]. The genome of P. vulgatus encodes several components of QS systems (online suppl. Table S4). Of the canonical QS systems LuxI/R (known from squid symbiont Vibrio fisheri) and TraI/R (known from the plant pathogen Agrobacterium tumefaciens), only homologs of the response regulator LuxR (also identified) and the autoinducer synthase TraI could be predicted from the genome of P. vulgatus. The LuxI/R and TraI/R systems are known to respond to the autoinducers N-(3-oxohexanoyl)- and N-(3-oxooctanoyl)-homoserine lactone [80], respectively. Notably, the former homoserine lactone could be identified in the cell-free supernatant of P. vulgatus, showing high relative abundances at OD_max and even more so at OD_stat, while for the latter only traces could be detected at OD_stat. Their synthesis requires acyl-acyl carrier proteins [81], several potential candidates for which were also identified in the present study (online suppl. Table S4). In addition, P. vulgatus possesses the AI-2-forming LuxS-pathway with the two key enzymes polyadenylation factor (Pfs) and AI-2 production protein (LuxS); both were identified (Fig. 5c). This is in accord with previous studies demonstrating that culture supernatants of P. vulgatus stimulated the QS system-2 as monitored with a Vibrio harveyi reporter system [82] and that LuxS is functional [83]. Taken together, these data suggest that P. vulgatus may engage in QS in stationary growth phase; however, the formed AIs may also serve other hitherto unknown functions. For example, the closely related B. fragilis possesses nine predicted LuxI/R paralogs, of which four were found to promote expression of genes related to antibiotic resistance upon addition of homoserine lactone [77]. The role of the predicted yet unidentified TraI in P. vulgatus remains elusive at present. In A. tumefaciens TraI produces N-(3-oxooctanoyl)-DL-homoserine lactone, in response to which TraR activates expression of the tra operon responsible for the mobilization of the Ti plasmid.

We investigated the molecular composition and diversity of the P. vulgatus exometabolome. It was isolated through solid-phase extraction of cell-free supernatants obtained at each growth stage. Ultrahigh resolution mass spectrometry analysis via the Fourier-transform ion cyclotron resonance technique (FT-ICR-MS) revealed a complex and molecularly highly diverse exometabolome. The number of detected molecular formulas increased with incubation time of the bioreactor cultures: on average 2,317 at ½OD_max, 4,258 at OD_max, and 4,501 at OD_stat (Fig. 6a). Singlets, i.e., molecular formulas that were detected only once in all replicates and growth stages, are not considered. Over the entire period of incubation, a total of 7,918 molecular formulas were detected at all three growth phases. Notably, 126, 157, and 376 molecular formulas occurred exclusively at growth phases ½OD_max, OD_max, and OD_stat, respectively. The number of molecular formulas containing one or two sulfur atoms increased over time (Fig. 6b), which is likely the result of biosynthesis and secondary abiotic sulfurization reactions in the sulfide-rich medium. The H/C versus O/C ratio of the molecular formulas that contribute to the exometabolome showed no clear trend over time (Fig. 6c, d). We used anvi’o to perform a hierarchical clustering analysis of molecular formulas across samples (Fig. 6e). At ½OD_max, the molecular formulas group closely together. Thus, most of the molecular formulas occurred in all four replicate experiments. This initial cluster of molecular formulas remained essentially unmodified throughout the incubation period. The most prominent increase in molecular formula counts occurs between ½OD_max and OD_max, where the detected number of molecular formulas nearly doubled. These newly formed molecular formulas form a separate cluster that persists at OD_stat. While the numeric increase in detected molecular formulas was small between OD_max and OD_stat, the count of unique molecular formulas increased consistently over the incubation period. A relatively small subset of molecular formulas disappeared, potentially indicating continuous reworking and diversification of the exometabolome by P. vulgatus.

The present bioreactor-based study with P. vulgatus provides a solid basis for further assessment and development of this prominent Bacteroidota member as a platform organism for renewable production of commodity chemicals. In particular, cultivation in bioreactors proved highly reproducible with respect to general growth parameters as well as product yields. Furthermore, it became evident that growth performances determined by means of more convenient flask-based batch cultivation can only be conditionally transferred to the bioreactor setup. Future studies may address the topic of increasing product yields not only by optimizing general cultivation conditions or integrating genetic engineering to enhance metabolic flux, but also by considering the marked effect of the type of applied feed substrate on the composition of the fermentation products, e.g., rhamnose/fucose-specific formation of propane-1,2-diol [22], a capacity, which was to date apparently not reported for the well-studied B. fragilis [84]. A further application relevant asset of P. vulgatus could be its broader repertoire in PULs and CAZymes compared to B. fragilis (67 vs. 55 and 351 vs. 266) [85, 86]. Future process optimization with P. vulgatus may also be facilitated by developing a metabolic model as recently reported for B. fragilis [84].

While the stationary phase/stress response of P. vulgatus is essentially in accord with current knowledge from standard bacteria, several aspects remain unclear. This involves, e.g., the anabolic network for monomer and QS system biosynthesis. An improved functional prediction of encoded proteins would be essential for designing a predictive metabolic model, which will benefit biotechnological as well as gut health-related research. A better understanding of the QS system of P. vulgatus is required to decipher the true function of the identified autoinducers, which may be involved in symbiotic or pathogenic interactions in the gut environment.

A remarkable observation of the present study was the complexity and dynamics of the exometabolome of P. vulgatus as resolved by FT-ICR-MS. This unprecedented perspective on the highly diverse molecular background against which fermentation products and secondary metabolites are formed is valuable in at least two respects: First, the product purity in biotechnological applications can be assessed in much greater detail. Second, the gut metabolome is apparently highly complex, implicating multifaceted and interwoven effects on the physiology of individual microbiome members as well as epithelial cells of the host.

An ethics statement was not required for this study type, no human or animal subjects or materials were used.

The authors have no conflicts of interest to declare.

This study was supported by the Federal Ministry of Education and Research (Bundesministerium für Bildung und Forschung) within the framework of the collaborative BaPro project.

R.R. conceived the study; S.T.V., U.C., and P.L. conducted the growth experiments; S.S. conducted HPLC analysis; M.N.-S. and J.F. conducted the metabolite analysis; M.G. and L.W. performed the proteomic analysis; S.T.V. and U.C. did mathematical calculations and the illustration of figures and tables; R.R., S.T.V., and U.C. wrote the manuscript with contributions of M.N.-S., J.F., and T.D. All authors have agreed to the final version of the manuscript.

Sören-Tobias Vital and Urte Clausen contributed equally to this study.

The proteomic and metabolomic data sets as well as calculations were deposited at FAIRDOMHub (https://fairdomhub.org/data_files/7101?version=1). Further inquiries can be directed to the corresponding author.