A comparison between the original sample and the same sample incubated for 24 h using the SIFR^® technology was performed first, to assess the accuracy of the technology for mimicking in vivo conditions. A diverse range of in vivo-derived gut microbes endured over the 24 h incubation period, maintaining microbial diversity, both in terms of species richness and evenness (Figures 2A,B). Under control conditions (NSC), the total cell numbers even increased from 0 h (INO) to 24 h with a factor of 2.06 (±0.25) (Figure 2C). Importantly, at the end of the 24 h NSC incubation, the microbial composition reflected to the original inoculum (INO; Figure 2D). This accurate preservation of in vivo-derived microbiota for the entire duration of the experiment classifies the application of SIFR^® technology as an ex vivo study, which is a study that uses an artificial environment outside the human body with minimum alteration of natural conditions. Such sustained similarity is fundamentally different from consistent biases observed for the current generation of in vitro gut models (Rajilić-Stojanović et al., 2010; Van den Abbeele et al., 2010, 2013, 2020a; O’Donnell et al., 2018; Gaisawat et al., 2019; Biagini et al., 2020).

IN primarily increased relative abundances of Actinobacteria, due to the stimulation of Bifidobacteriaceae (Figures 3A,B). This is in line with findings of Vandeputte et al. (2017a) which demonstrated that the key impact of inulin intervention using next-generation sequencing (%) is an increase in Bifidobacteriaceae. While such in vivo observations were obtained upon 4 weeks of repeated daily administration (12 g/day), the SIFR^® study only lasted for 24 h, displaying the efficiency at which data predictive for in vivo outcomes can be obtained using an ex vivo technology.

As reviewed by Delcour et al., a key limitation of in vivo studies is the inability to analyze the most prominent functional changes induced by prebiotics, i.e., increased production of SCFAs, given their rapid absorption/consumption in vivo (Delcour et al., 2016). Using the SIFR^® technology, it was demonstrated that IN significantly increased SCFA production for each of the 6 donors tested (Figure 3F; Supplementary Figure S3). In line with in vivo findings (Boets et al., 2015), IN most strongly increased acetate. Further, also propionate and butyrate production was enhanced. Further, marked interpersonal differences were observed in terms of how IN stimulated propionate (Supplementary Figure S3D) and butyrate (Supplementary Figure S3E), suggesting a low selectivity of IN. By performing ex vivo studies in parallel to clinical studies, unique insights in the production of metabolites can be generated that could explain in vivo observations such as responders/non-responders caused by interpersonal differences.

While quantitative sequencing has been introduced for the analysis of in vivo samples as a means to further discriminate enterotypes (Vandeputte et al., 2017b), it is particularly useful in ex vivo studies where, unlike in vivo, simulated colonic volumes are exactly known, allowing to accurately account for changes in microbial loads (Van den Abbeele et al., 2020b). Here, quantitative sequencing removed the large bias caused by the drastically increased cell counts upon IN treatment, as visualized at the phylum level (Figure 3C). This revealed that, in contrast to aforementioned conclusions based on relative data that exclusively identified a bifidogenic effect, a broad spectrum of families benefited from IN administration (Figure 3D), including families that, based on relative data, were supposedly unaffected (e.g., Lachnospiraceae, Ruminococcaceae) or even inhibited by IN (e.g., Oscillospiraceae and Rikenellaceae). Likewise, 28 species significantly increased upon IN treatment (Figure 3E), including among others Bifidobacterium adolescentis that was responsible for the marked bifidogenic effect of IN.

Further, correlations were established that support the mode of action. Butyrate production upon IN treatment correlated with Faecalibacterium prausnitzii and Anaerobutyricum hallii [formerly known as Eubacterium hallii (Shetty et al., 2018)], species known to produce butyrate via the butyryl-CoA:acetate CoA-transferase route (Louis and Flint, 2017). Further, propionate production correlated with Bacteroides stercoris, a species known to produce propionate via the succinate pathway (Johnson et al., 1986; Louis and Flint, 2017).

By accounting for interpersonal differences, the SIFR^® technology not only generated predictive findings, but also allowed for the formulation of a hypothesis regarding the mode of action. Understanding the mechanism by which microbiome-targeted products exert beneficial effects is essential to increase success rates of clinical studies and provide understanding when such studies failed (e.g., when a specific species is absent, it could explain non-response).

Resistant dextrin (RD) is a soluble fiber that is well tolerated and exerts health benefits by increasing satiety and improving both insulin resistance and determinants of metabolic syndrome (Li et al., 2010; Guérin-Deremaux et al., 2011). RD highly and specifically stimulates Parabacterodes distasonis from 0.92% at baseline up to 7.2% of the gut microbiota upon repeated intake by human adults over 6 weeks [20 g/day during final 4 weeks; (Thirion et al., 2022)]. A single intake study using the SIFR^® technology (48 h) demonstrated that RD treatment markedly increased Lachnospiraceae and especially Tannerellaceae levels (Figure 4A). At species level, Parabacteroides distasonis (Figures 4B,C; Supplementary Figure S4A) increased from 0.97% in the NSC up to 9.3% for RD treatment, thus highly accurately mirroring aforementioned clinical data (Thirion et al., 2022). In line with metabolic capabilities of Parabacteroides distasonis to produce acetate and propionate (Gotoh et al., 2017), RD strongly increased these SCFA (Figure 4D; Supplementary Figure S5). The rCCA confirmed this link and additionally revealed a correlation between (i) Bacteroides uniformis and propionate, and (ii) Porphyromonas_u_s and butyrate. Bacteroides uniformis has indeed been identified as potent dextrin fermenting species (Reichardt et al., 2018), able to produce propionate via the succinate pathway (Louis and Flint, 2017). Further, the unclassified Porphyromonas species belongs to the Porphyromonadacae family which contains species able to produce butyrate via the acetyl-CoA route (Vital et al., 2017), thus further supporting the involvement of particular members of this family in butyrate production from RD.

A third case study was performed with 2′FL, an abundant HMO that exerts potent bifidogenic effects (related to stimulation of Bifidobacterium adolescentis) when dosed to human adults over 2 weeks [at 5, 10, or 20 g/day; (Elison et al., 2016)]. Furthermore, when administered to IBS patients (over 4 weeks at 5 or 10 g/day), bifidogenic effects were accompanied by an increase of Anaerobutyricum hallii (Iribarren et al., 2020). A single intake study using the SIFR^® technology (24 h) revealed that 2′FL markedly increased levels of Bifidobacteriaceae and Lachnospiraceae (Figure 5A) due to increases of B. adolescentis and A. hallii (Figures 5B,C). In line with the metabolic capabilities of B. adolescentis and A. hallii to, respectively, produce acetate (De Vuyst et al., 2014) and propionate/butyrate (Engels et al., 2016), 2′FL strongly increased the production of these SCFA (Figure 5D; Supplementary Figure S7). Two mechanisms have been identified that highlight the metabolic complementarity of B. adolescentis and A. hallii. First, Bifidobacterium species produce acetate and lactate (De Vuyst et al., 2014) that can be consumed by A. hallii to produce butyrate (Engels et al., 2016). Secondly, Bifidobacterium species can ferment sugars that A. hallii is unable to ferment (e.g., fucose) and in doing so produce 1,2-propanediol, which can be utilized by A. hallii to produce propionate (Schwab et al., 2017). While A. hallii is unable to ferment complex oligo- and polysaccharides (Scott et al., 2014), such metabolic compatibilities with B. adolescentis likely render A. hallii competitive against the 100’s of other species in the gut microbiota.

Overall, similar to case study 1 the SIFR^® technology enabled predictive observations as quickly as within days during case studies 2/3. This accurate mirroring of in vivo observations that required repeated daily administration over weeks was not restricted to primary substrate degraders (e.g., B. adolescentis) but also extended to secondary degraders (e.g., A. hallii).

At the end of incubation (24 h), the coefficients of variation (CV = standard deviation/average) across the two donor scenarios, each tested in n = 6 (one donor and pooled sample of 6 donors) were on average 2.4% for fundamental fermentation parameters (pH, acetate, propionate and butyrate) and 15.2% for quantification of the 50 most abundant microbial species. This covered variation due to reactor preparation, incubation, sampling, but also technical variation of SCFA or sequencing/cell counting analysis. E.g., when butyrate levels in the NSC would be 10.0 mM, the standard deviation would on average be 0.24 mM, meaning that univariate statistical tests with significance level of 0.05 would allow to identify a significant treatment effect when butyrate levels increase up to 10.48 mM (=average + 2 × standard deviation). Similarly, when a microbial species is present at 8.00 log₁₀ (cells/mL) in the NSC, a significant effect can be identified when it increases to 8.11 log₁₀ (cells/mL). Such high technical reproducibility renders the SIFR^® technology highly sensitive in identifying small but significant changes, while enabling studies that focus on biological rather than technical replicates.

PCAs based on microbial composition (Figure 6A) and metabolite production (Figure 6B) revealed that the pooled inoculum did not provide the same results as the average among individual donors. By combining samples of 6 donors, microbial diversity dramatically increased to levels that were not representative for the individual donors (Supplementary Figure S8). This resulted in an overestimation of the metabolic capabilities of the artificial community as reflected by, e.g., total SCFA levels that largely exceeded those of the individual donors (Supplementary Figure S9). Further, a key drawback of pooled samples but also of repetitions from a single donation was that no meaningful correlations could be established between metabolites and species (Figures 6C,D). Using single/pooled donations, all species stimulated by IN correlated with IN-mediated metabolites and vice versa thus impairing to obtain insightful correlations between specific bacteria and metabolites that could support mode of action.

On the other hand, working with a single/pooled sample in n = 6 provided great statistical power to identify changes within these samples, which revealed that IN stimulated a broad range of gut microbes. For the pooled sample, 77 (of the 100 most abundant) species significantly increased upon IN treatment (Figure 6D). This number is markedly higher compared to the number of species when accounting for 6 individual samples (Figure 3E). Such general stimulation of a broad range of species also strongly disputes the conclusions of current clinical studies that IN would be highly selective given the small amount of significantly affected taxa [i.e., Bifidobacteriaceae; (Vandeputte et al., 2017a)]. The present data shows that counterintuitively, it is not the high but rather the low selectivity of IN that causes only a limited number of taxa (e.g., Bifidobacteriaceae) to be significantly affected when accounting for interpersonal differences, especially when only considering proportional and not quantitative sequencing outputs. This finding supports recent hypothesis that IN could not be as selective as has been proposed (Cantu-Jungles and Hamaker, 2020). Overall, the current study highlights the importance of performing ex vivo studies in parallel to clinical studies to enable correct conclusions on fundamental mechanistic aspects such as the tentative selectivity of how substrates modulate the gut microbiota, a criterium at the core of the prebiotic consensus definition (Gibson et al., 2017).

Upon 24 h of incubation, the mucin gel was preferentially colonized with Lachnospiraceae, accounting for, on average, 53.7% of the simulated mucosal microbiota, largely exceeding luminal levels (15.2%; Supplementary Figure S10A). This is in line with human in vivo data (Lavelle et al., 2015) and findings from in vitro gut models run up to 3 days (Van den Abbeele et al., 2013). In contrast to the current generation of gut models running over weeks that do not preserve enrichment of mucosa-associated Lachnospiraceae (Van den Abbeele et al., 2021), the current SIFR^® study thus enabled the potential observation of treatment effects on these mucosal microbes. In contrast to the profound treatment effects of IN on the luminal microbiota (Figure 3), IN did not significantly impact mucosal microbes (Supplementary Figure S10B). Treatment effects of IN on the luminal microbiota were similar whether or not the mucin gel was added, both in terms of microbial composition and metabolite production (Supplementary Figures S10C,D). The absence of treatment effects of IN on the mucosal microbiota could potentially be attributed to the aforementioned low selectivity of how IN impacts the microbiota. While it would be relevant to more broadly apply the SIFR^® model with mucin gel to gain an understanding of how therapeutics affect this simulated mucosal microbiota, in absence of in vivo data on this research topic, it remains difficult to validate the predictivity of such findings.