The human gut microbiota is a dense, complex community comprising up to 10¹⁴ microorganism belonging to a broad range of genera (500–1,000 across individuals) fit to thrive under the conditions of temperature, pH range, nutrient composition, and redox potential/oxygen concentration that define the colon environment (Falony et al., 2016; Poppe et al., 2021). Variation of these abiotic factors translates in changes in microbiota composition with potential impact on host health and wellbeing, opening perspectives for the design of targeted modulation strategies. Exploring these perspectives requires mechanistic insights in community dynamics (Vieira-silva et al., 2019; Vieira-Silva et al., 2020). To gain such essential insights, in vivo intervention trials remain the gold standard, despite having the disadvantages of being highly resource-and time consuming. Due to the high degree of inter-individual variation in microbiota composition, large numbers of volunteers are required. Moreover, avoiding more invasive sampling strategies, the effect of modulation approaches is mostly evaluated based on the fecal microbiome, representing the end-point of the colonic fermentation process and not allowing dissection of the preceding growth dynamics (Falony et al., 2018). Hence, in vitro fermentation strategies are considered an interesting alternative to bypass these issues. As such, a range of fermentation systems spanning from simple batch fermentations to more chemostat-like complex set-ups have been developed over the years (Venema and van den Abbeele, 2013; Vrancken et al., 2019). While more advanced simulators aim at mimicking the in vivo reality as accurately as possible, their complexity comes at the expense of throughput. In contrast, batch systems have a clear cost and throughput advantage, but are often more limited with respect to optimal process control in terms of pH and accumulations of metabolites. However, considering the fecal microbiota as the endpoint of a process of ecosystem maturation (rather than a sample of a continuously mixed population inhabiting the large intestine), dynamics as observed in batch systems can be hypothesized to resemble the plug-flow conditions that determine colon ecology (Falony et al., 2019).

Storage of starting material, cell density, and fermentation medium all have been described independently as factors affecting the outcome of fermentation experiments inoculated with single bacterial species as well as complex communities (Bircher et al., 2018; Atolia et al., 2020). When describing in vitro set-ups with fecal slurries, however, investigators rarely present the selection processes applied to identify the optimal fermentation conditions for their experiments. Such (often trial-and-error based) screening processes usually consist of a time-consuming and laborious series of small-scale incubations. Here, we present a structured way to investigate how fermentation conditions influence batch incubation dynamics. By providing insights in a process that often remains too obscure, we aim to facilitate the selection of appropriate incubation conditions for researchers planning to set up batch fermentation experiments with fecal material.

To determine optimal fermentation conditions for future batch experiments with fecal material, we designed a range of explorative incubations, evaluating the effects of storage (fresh vs. stored fecal material) and concentration (bacterial cells/ml) of the inoculum on fermentation outcome. Homogenized fecal material was obtained from a single healthy donor, aliquoted, and used to start up a set of 36 fermentations. To explore the effect of dilution and frozen storage (1 week at −80°C) of the starting material on fermentation outcome, inocula encompassed fresh fecal slurry as literature-established gold standard (processed within 2 h post sampling; 1/10 w/v in physiological solution [PS; 9 mg NaCl/ml]; inoculum cell concentration of 10¹¹ cells/ml), fractions obtained from a fecal dilution series of fresh feces (1/10 dilution series starting from 10¹⁰ to 10⁷ bacterial cells/ml), stored aliquots obtained from the same dilution series (10⁸ and 10⁷ cells/ml), and two suspensions prepared out of a 125 mg aliquot of stored fecal material (10⁸ cell/ml and a 1/10 dilution to 10⁷ cells/ml). For each fermentation, 1 ml of inoculum was added to 9 ml of medium. Per inoculum (n = 9), fermentations were performed in triplicate in Luria Broth (LB) supplemented with 1 g/L of pectin (citrus pectin; degree of methylation, 52.5%; galacturonic acid content, 78.1%). In addition, per inoculum, one fermentation was started up in LB without pectin as a control. Fermentations were incubated at 37°C in an anaerobic cabinet (80% N₂, 10% CO₂, and 10% H₂); samples were taken 24, 48, and 96 h after inoculation and subjected to 16S rRNA gene-based amplicon sequencing to determine the optimal time-range for subsequent batch fermentation experiments.

Across fermentations and time points, the largest effect sizes in genus-level community variation were observed for differences in starting cell densities (dbRDA on the Aitchison distance matrix, n = 108, R² = 0.13, p_adj = 2.0e⁻⁶; Supplementary Table S1), followed by the fresh/stored nature of the inocula used (R² = 0.11, p_adj = 2.0e⁻⁶). When comparing the incubations of fractions obtained from the fresh and stored dilution series overlapping in terms of starting cell density (10⁷–10⁶ cells/ml), frozen storage of inocula resulted in lower community diversity in 96 h fermentation outcomes (Inverse Simpson, Mann–Whitney U test, n_96h = 16, r = 0.630, p_adj = 1.04e⁻²; Figure 1A). Similarly, focusing on the fresh dilution series encompassing the broadest range of starting cell concentrations (including fecal slurry incubations), inoculum density was observed to correlate positively with diversity after 96 h of fermentation (Inverse Simpson, n_96h = 20, Spearman ρ = 0.79, p_adj = 4.95e⁻⁵; Figure 1B). The observed decrease in diversity correlated with increased relative abundances of Enterobacteriaceae (Spearman ρ = −0.72, p_adj = 4.58e⁻³; Figure 1C; Supplementary Table S2), reaching highest proportions in the 10⁶ and 10⁷ cells/ml starting cell count conditions, but remaining barely detectable in the >10⁸ cells/ml incubations.

Fermentation time had a significant albeit modest impact on genus-level community variation as a whole (dbRDA, n = 108, R² = 0.01, p_adj = 5.57e⁻³; Supplementary Table S1). The evolution of the relative abundances of taxa was inoculum-dependent. In fermentations inoculated with fresh material, the largest increase was observed for Peptostreptococcus (Friedman with pairwise Durbin-Conover test, n = 60, Kendall’s W = 0.72, p_adj = 1.6e⁻⁵; pairwise comparisons in Supplementary Table S3) followed by Dialister (Kendall’s W = 0.28, p_adj = 1.5e⁻²). Clostridium XI (Kendall’s W = 0.25, p_adj = 1.9e⁻²) and Blautia (Kendall’s W = 0.39, p_adj = 3.0e-3) relative abundances decreased. In incubations of stored material, Parabacteroides and Clostridium XIVa (n = 48, Kendall’s W = 0.40, p_adj = 9.3e⁻³ and Kendall’s W = 0.33, p_adj = 1.8e⁻², respectively) increased, while proportions of unclassified Enterobacteriaceae decreased (Kendall’s W = 0.36, p_adj = 1.4e⁻²). Relative abundances of Flavonifractor were observed to rise in incubations of both stored and fresh material (fresh, n = 60, Kendall’s W = 0.44, p_adj = 1.3e⁻³; stored, n = 48, Kendall’s W = 0.78, p_adj = 5.0e⁻⁵). Community diversities increased significantly over time after inoculation with stored material (Inverse Simpson, n = 48, Kendall’s W = 0.77, p_adj = 9.57e⁻⁶), an association that was absent in overall more diverse incubations of freshly collected feces (n = 60, Kendall’s W = 0.032, p_adj = 0.55; Supplementary Table S2).

In contrast with the observed changes in genus-level taxa relative abundances over time, larger fluctuations in overall metabolite profiles could be observed between the 24, 48, and 96 h time points (dbRDA on Euclidian distance matrix n = 108, R² = 0.11, p_adj = 1.33e⁻⁶; Supplementary Table S4). The most striking changes in metabolite class concentrations concerned a decrease in alcohols (Friedman with pairwise Durbin-Conover test, n = 108 Kendall’s W = 0.67, p_adj = 1.69e⁻¹⁰), combined with an increase in branched-chain fatty acids (Kendall’s W = 0.90, p_adj = 8.12e⁻¹⁵; Figure 1D; Supplementary Table S5). Mirroring microbiome observations, overall metabolite profiles were significantly impacted by starting cell densities and the fresh/stored nature of the inocula. Stored inocula mainly resulted in higher relative concentrations of sulfur-containing (Mann–Whitney U-test, n_96h = 24, r = 0.76, p_adj = 1.9e⁻⁴) and heterocyclic compounds (r = 0.68, p_adj = 1.85e⁻³; Supplementary Table S6), while proportions of branched-chain fatty acids (BCFAs) remained lower (r = −0.63, p_adj = 4.2e⁻³) when compared to fermentations inoculated with fresh material. Similarly, in incubations initiated with fresh material, proportions of short-chain fatty acids (SCFAs) and benzenoid compounds positively correlated with starting cell density (n_96h = 20, Spearman ρ = 0.74, p_adj = 6.03e⁻⁴ and Spearman ρ = 0.72, p_adj = 7.80e⁻⁴, respectively; Supplementary Table S6). A negative association was observed for sulfur-containing (Spearman ρ = −0.83, p_adj = 4.93e-5) and heterocyclic compounds (Spearman ρ = −0.80, p_adj = 1.01e⁻⁴).

The initial, explorative analyses described above indicated an overgrowth of the family Enterobacteriaceae and lower community diversity in fermentations inoculated with stored material and with lower inoculum cell counts. Fermentation time only had a limited impact on microbiome variation within the 24–96 h sampling window. Based on these observations, we designed a single-donor full factorial follow-up experiment, refocusing the sampling window and increasing frequency to 0, 6, 12, and 24 h. Variation in initial cell densities in the fermentation medium was narrowed to 10⁸–10¹⁰ cells/ml. Besides LB, we added a ½ diluted medium as well as a physiological saline (PS) arm to the design to evaluate the effect of nutrient density on the microbiota and metabolite profiles. Fermentations were inoculated with fresh or stored inocula, with or without supplementation of the medium with 1 g/L of pectin. Incubations were carried out in duplicate, resulting in a total of 72 experiments.

As expected, given the narrowed, more focused set-up of the full-factorial experiment, the explanatory power of fermentation conditions on microbiome variation increased compared to the exploratory trial. Fermentation time was observed to have the second largest effect size in microbiome variation (dbRDA on Aitchison distances, n = 288, R² = 0.08, p_adj = 5.0e⁻⁶), topped by starting cell count (R² = 0.22, p_adj = 5.0e⁻⁶) and followed by medium nutrient density (i.e., LB, LB_/2, or PS; R² = 0.06, p_adj = 1.2e⁻⁴), fresh/stored nature of the inoculum (R² = 0.03, p_adj = 1.74e⁻³; Figure 2A), and addition of pectin (R² = 0.02, p_adj = 2.2e⁻²; Supplementary Table S7). Across fermentations, the community diversity decreased in function of fermentation time (Inverse Simpson, Friedman with pairwise Durbin-Conover test, n = 288, Kendall’s W = 0.12, p_adj = 1.07e⁻⁵; Supplementary Table S8). This decrease was however mostly driven by microbiome variation as observed in the 10⁸ and 10⁹ cells/ml starting cell count conditions, since no significant association between diversity and time could be established in the 10¹⁰ cells/ml fermentations (10⁸ cells/ml, Friedman with pairwise Durbin-Conover test, n = 96, Kendall’s W = 0.57, p_adj = 2.0e⁻⁸; 10⁹ cells/ml, n = 96, Kendall’s W = 0.35, p_adj = 1.8e⁻⁵; 10¹⁰ cells/ml, n = 96, Kendall’s W = 0.107, p_adj = 0.052; Figure 2B; Supplementary Table S8). Excluding the 10¹⁰ cells/ml condition, we observed a decrease in diversity over time for both fresh and frozen material (fresh, Friedman with pairwise Durbin-Conover test, n = 96, Kendall’s W = 0.24, p_adj = 5.99 e⁻⁴; stored, n = 96, Kendall’s W = 0.72, p_adj = 6.14e⁻¹¹; Supplementary Table S8). Overall, the diversity observed after 24 h of incubation appeared to be tightly linked to Enterobacteriaceae relative abundances (n_24h = 72, Spearman ρ = −0.94, p_adj = 2.2e⁻¹⁶; Figure 2C; Supplementary Table S9).

While the evolution of community diversity allows to screen for major macro-ecological perturbations such as population overgrowth, it does not provide insights in the dynamics of individual taxa. One could even envisage a scenario where a microbial population maintains a constant level of diversity, but where all taxa present in the ecosystem get replaced by opportunistic colonizers. To quantify drift from starting profiles, we evaluated the Bray–Curtis dissimilarity distances between time points 0 and 24 h of all incubations. For all fermentations inoculated with stored material and for those with a starting cell density of 10⁸ cells/ml, dissimilarity was higher than 0.45, indicating a substantial drift away from the initial microbiota composition (Figure 2D; Supplementary Table S10). Lowest dissimilarities (<0.30) were observed for starting densities of 10¹⁰ cells/ml, incubated in either LB or LB_/2.

Fermentation metabolite profiles were most strongly impacted by starting cell density (dbRDA on Euclidian distance matrix, n = 288, R² = 0.28, p_adj = 2.5e⁻⁶;), followed by medium nutrient density (R² = 0.076, p_adj = 2.5e⁻⁶) and fermentation time (R² = 0.044, p_adj = 2.5e⁻⁶; Figure 3A; Supplementary Table S11). In contrast with microbiome observations, pectin addition did not have an effect on metabolite profiles (R² = 0.002, p_adj = 0.576), and the effect size of inoculum storage was smaller (R² = 0.01, p_adj = 0.016). Concentrations of all absolutely quantified metabolites increased over time, with the largest effect sizes observed for butyric acid (Friedman test, n = 288, Kendall’s W = 0.742, p_adj = 1.7e⁻³³), iso-valeric acid (Kendall’s W = 0.566, p_adj = 1.3e⁻²⁵), and indole (Kendall’s W = 0.464, p_adj = 1.3e⁻²⁵; Supplementary Table S12). Meanwhile, the putative deleterious protein fermentation metabolite para-cresol only showed limited variation over time (Kendall’s W = 0.0566, p_adj = 6.6e⁻³). Interestingly, para-cresol increased only in the 10¹⁰ cells/ml condition (Friedman with pairwise Durbin-Conover test, n = 96, Kendall’s W = 0.736, p_adj = 5.5e⁻¹¹), while decreasing in the 10⁸ and 10⁹ cells/ml incubations (10⁸ cells/ml, n = 96, Kendall’s W = 0.7, p_adj = 3.3e⁻⁷; 10⁹ cells/ml, n = 96, Kendall’s W = 0.457, p_adj = 9.8e⁻¹¹; Figure 3B; Supplementary Table S12), resulting in a significant difference at the end of the fermentation (Kruskal-Wallis with post hoc Dunn-test, n_24h = 72, eta²[H] = 0.69, p_adj = 3.0e⁻⁴; Supplementary Table S13). In contrast, indole increased in all conditions, but was significantly higher after 24 h of fermentation for the 10⁸ and 10⁹ cells/ml starting cell count conditions (n_24h = 72, eta²[H] = 0.21, p_adj = 8.8e⁻¹¹; Supplementary Table S13). The discrepancy between indole and para-cresol observations in the 10¹⁰ and the 10⁸–10⁹ cells/ml starting cell count conditions might, along with the difference in diversity dynamics and Enterobacteriaceae bloom, point to a difference in growth stage for the microbiota.

To evaluate the effect of pectin addition on the absolute concentrations of metabolites through time of fermentation, we modelled the experiments stratified per medium and starting cell density (linear modelling [LM]). Pectin addition had a significant effect on at least one of the metabolites’ concentration in 3 of the 9 combinations of conditions evaluated (Figure 3C). For the LB_/2 medium with a starting cell density of 10⁹ cells/ml, pectin addition significantly increased the concentrations of acetic acid (LM, n = 32, estimate = 0.695 mM/h, p_adj = 0.045), propionic acid (estimate = 0.230 mM/h, p_adj = 0.045), and indole (estimate = 0.0143 mM/h, p_adj = 0.045) produced during the fermentation.

Overall, based on microbiome observations, the results of the full factorial experiment indicate that in order to preserve the fecal microbiota composition over the course of batch fermentation experiments, the use of a freshly prepared inoculum, incubated in LB or LB_/2 at a starting cell count of 10¹⁰ cells/ml is recommended. Metabolome findings indicate that, in order to follow up the effect of the addition of a fermentable substrate (i.e., pectin), incubations should preferably be carried out in a diluted medium. To further validate the workflow proposed for the key endpoint of community representativeness - tightly coupled to controlling opportunistic colonizer proportions – we performed 12 additional fermentations inoculated with fecal material from different donors spanning all four commonly detected enterotypes (Vandeputte et al., 2017). Carried out in LB_/2 medium without pectin with a starting cell density of 10¹⁰ cells/ml, proportions of Escherichia were determined after 24 h of fermentation using 16S rRNA gene-based amplicon sequencing. Resulting proportions of Escherichia varied between 0 and 19% (median abundance of 3%) – comparable to the 6% in the corresponding condition of the full factorial experiment. This demonstrates that our workflow is generally applicable across community types, taking into account the variability of individual donors (Supplementary Table S15).

Assessing the impact of fermentation conditions on the outcome of incubations of fecal material, we observed starting cell density to be the major driver of end-point community as well as metabolite profile variation, followed by storage of inocula and medium nutrient density. A key factor for the interpretation of our findings is presumably the carrying capacity of the fermentation set-up, i.e., the maximum population density an environment can support (Nowak, 2006). Inoculation of the fermentation medium to a starting density of 10¹⁰ cells/ml appeared to approximate or even exceed the maximal load supported by the batch systems analyzed. On one hand, such a set-up confined the growth of opportunistic colonizers. On the other hand, while high-density incubations conserved the initial composition of the inoculum, they also limited exploration of the dynamics of ecosystem development. Lower starting cell densities (<10¹⁰ cells/ml) did provide opportunities for exponential growth of inoculated species, but the outcome of these fermentation experiments was characterized by lower microbial diversities and high proportions of Enterobacteriaceae. Representing a phylogenetic family of opportunistic pathogens, the latter mostly display short lag times and high growth rates (Pompan-Lotan et al., 2015; Milani et al., 2017). By lowering starting cell densities below the carrying capacity of fermentation set-up, an opportunity was created for such opportunistic colonizers to dominate the bacterial community (Monod, 1949; Jabot and Pottier, 2012) This effect was potentially exacerbated by the short but substantial oxygen and temperature shock the microbiota was exposed to after egestion and during inoculum preparation. Freezing the inoculum before fermentation compounded this stress, adding longer lag times to the equation (Atolia et al., 2020). This further benefited Enterobacteriaceae (and potentially other opportunists), pushing the endpoint community profile further from being representative for fecal ecosystem. Of note: while freezing stress mostly results from freeze/thaw cycles (leading to crystal formation and development of osmotic gradients), residual traces of unfrozen water at storage temperatures above −135°C could induce additional biological variation tightly linked to duration of frozen storage (Hubálek, 2003; Heylen et al., 2012). However, the latter was not investigated in the scope of the present study.

From an ecosystem development perspective, starting cell densities below or at system carrying capacity represent two distinct stages of fecal community development (Falony et al., 2018). Lower densities correspond with an immature configuration, where the relative fitness of individual taxa mostly depends of their ability to outcompete others in terms of nutrient consumption as well as growth rates (Costello et al., 2012). The associated ecosystem dynamics not only occur during initial colonic colonization of the chime matrix, but also determine the composition of the fecal microbiota of individuals with short transit times. In vitro, where the outcome of inter-taxon competition is not subjected to host control mechanisms such as immune surveillance, the end result of the community maturation process initiated by inoculum dilution unfortunately differed substantially from what has been observed in vivo, due to the unhampered overgrowth of opportunistic colonizers (Ho et al., 2017; Byndloss et al., 2018). Hence, the most effective way to generate an in vitro equivalent of a mature microbiota is to incubate (preferably freshly harvested) fecal material at cell concentrations approaching the maximal load supported by the set-up envisaged.

Regarding metabolite profiles, also the observed trade-off between indole and para-cresol production might reflect the difference in ecological development stage hypothesized to result from high-versus low-density inoculations. Indole production, mostly observed in our incubations with a starting cell density < 10¹⁰ cells/ml, has been associated with fast-growing Enterobacteriaceae (Lee and Lee, 2010). Para-cresol, observed in high-density inoculations potentially corresponding with more mature ecosystem configurations, has been traced back to more slow-growing clostridial clusters (Saito et al., 2018). Similarly, alcohol production has been linked to fast-growing saccharolytic species (Boumba et al., 2008; Elshaghabee et al., 2016) while branched-chain fatty acids are considered markers of protein fermentation (Macfarlane et al., 2006)– again aligning the observed proportional changes in metabolite production between low-and high-density inoculations with the expected dynamics of ecosystem maturation (Falony et al., 2018). Proportional depletion of the more readily degradable carbohydrate substrates with respect to the amount of amino acids and peptides present in the fermentation medium would indeed push the microbial community towards the energetically less favorable protein fermentation. The ratio of carbohydrates to amino acids and peptides in the medium also provides an explanation to the more pronounced effect of pectin addition in a more dilute medium. At a constant concentration of the added carbohydrate substrate (pectin), an increased concentration of peptides and amino acids lowers the Gibbs free energy of protein relative to that of carbohydrate fermentation, hence lowering the energetic benefit of carbohydrate fermentation.

Here, we optimized the fermentation conditions for in vitro fermentations with fecal inocula by systematically investigating the effect of starting cell density, medium nutrient density, and fresh inoculation versus inoculation after frozen storage on the metabolite and microbiota profiles. The aim of the optimization – here in the context of fermentations with fiber products - was to minimize drift from the starting inoculum in the condition without fiber product and to maximize the impact of the addition of fiber product on the fermentation characteristics. The optimization revealed that starting cell densities of 10¹⁰ cells/ml maximally lowered in vitro drift, but possibly limited exponential growth of taxa. Meanwhile, the addition of a more dilute medium (LB_/2) maximized the impact of fiber product addition on the metabolite profiles. Above all, we have provided a framework for other researchers to set up their own fermentations systems, and optimize for their specific set-up and research question.

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found at: https://www.ebi.ac.uk/ena, PRJEB58105.

The studies involving human participants were reviewed and approved by Ethics Committee Research UZ/KU Leuven (reference number S65767). The patients/participants provided their written informed consent to participate in this study.

JP, GF, and KV performed the conception and design of the study. JP performed the data collection and experimental work. JP and SV-S performed the data preparation. JP, SV-S, and GF performed the statistical and data analysis. JP, SV-S, JR, and GF performed the interpretation. JP, SV-S, GF, and KV drafted the manuscript. All authors contributed to the article and approved the submitted version.

JP is funded by a doctoral fellowship from Flanders Innovation & Entrepreneurship (VLAIO) (HBC.2017.0596), while GF is funded by the ReALity Innovation Fund, Research Initiative of the State of Rhineland-Palatinate, Germany.

JR is an advisor for MRM health.

The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.