DSS treatment does not affect murine colonic microbiota in absence of the host

ABSTRACT

The prevalence of inflammatory bowel disease (IBD) is rising globally; however, its etiology is still not fully understood. Patient genetics, immune system, and intestinal microbiota are considered critical factors contributing to IBD. Preclinical animal models are crucial to better understand the importance of individual contributing factors. Among these, the dextran sodium sulfate (DSS) colitis model is the most widely used. DSS treatment induces gut inflammation and dysbiosis. However, its exact mode of action remains unclear. To determine whether DSS treatment induces pathogenic changes in the microbiota, we investigated the microbiota-modulating effects of DSS on murine microbiota in vitro. For this purpose, we cultured murine microbiota from the colon in six replicate continuous bioreactors. Three bioreactors were supplemented with 1% DSS and compared with the remaining PBS-treated control bioreactors by means of microbiota taxonomy and functionality. Using metaproteomics, we did not identify significant changes in microbial taxonomy, either at the phylum or genus levels. No differences in the metabolic pathways were observed. Furthermore, the global metabolome and targeted short-chain fatty acid (SCFA) quantification did not reveal any DSS-related changes. DSS had negligible effects on microbial functionality and taxonomy in vitro in the absence of the host environment. Our results underline that the DSS colitis mouse model is a suitable model to study host–microbiota interactions, which may help to understand how intestinal inflammation modulates the microbiota at the taxonomic and functional levels.

1. Introduction

The prevalence of inflammatory bowel disease (IBD) is increasing globally and is expected to grow even further until 2025.¹ The increasing global prevalence is a consequence of the number of newly diagnosed cases (incidence) in Western countries and the rising incidence in developing countries.² Despite extensive research, the etiology of IBD is not fully understood. Patients’ genetics, immune system, personal environment, and associated gut microbiota are the most important factors contributing to IBD.³ Further research is needed to understand the contribution of individual factors in more detail, and the underlying mechanisms in the broader context of the disease.

Mouse models are valuable tools for investigating the pathophysiological mechanisms underlying IBD. Several models exist, such as IL-10 knock-out mice,⁴ adoptive T cell transfer colitis model,⁵ and chemically induced colitis models using oxazolone, trinitrobenzene sulfonic acid, or dextran sulfate sodium (DSS).⁶ The DSS colitis model⁷ is probably the most widely used among the chemically induced colitis models⁸ because it is very versatile and can be used to study different aspects of intestinal inflammation, such as acute, chronic, and remitting-relapsing gut inflammation, depending on the molecular weight of DSS, its concentration, and its administration schedule.⁹ For colitis induction, the water-soluble DSS is commonly administered to the animals ad libitum via the drinking water at concentrations ranging from 1% to 5%.^8–10 Although the exact mode of action of DSS remains unclear,^8,11 the induction of colonic inflammation has mostly been attributed to the epithelium-damaging effect of DSS.⁹ DSS affects zonula occludens-1 (ZO-1) synthesis, a tight junction protein, on day 1 after the first administration, leading to epithelial leakiness.¹² Consequently, luminal content induces inflammation in the underlying tissues.¹³ Inflammation, in turn, affects a variety of physiological parameters such as mucin production,¹⁴ strength of the mucus layer¹⁵ and colon length.¹⁶ DSS-induced colitis is dependent on the genetic background,¹⁷ the intestinal microbiota, and the severity of colon inflammation has been shown to depend on microbiota composition.¹⁸ DSS-induced colitis is accompanied by changes in the gut microbiota.¹³

The intestinal microbiota has a high genetic repertoire¹⁹ that allows it to adapt to dietary changes,²⁰ drug²¹ or xenobiotic exposure²² very quickly. Thus, it is still unclear whether DSS itself alters the gut microbiota composition and functionality, thereby contributing to colitis. To address this, we analyzed the effect of DSS on the composition and functionality of murine intestinal microbiota in vitro using a chemostat bioreactor system for continuous cultivation. This setup permits us to investigate the effects of chemical agents or drug candidates on the microbiota directly to understand the potential microbiota-modulating properties of DSS without host interference. In this study, we aimed to ascertain the suitability of the DSS-induced colitis murine model for investigating colitis-associated microbiota effects, that is, to determine whether changes in the microbiota are directly linked to DSS or are associated with host colitis.

2. Results

To assess the microbiota-modulating properties of DSS, we cultivated a complex murine microbiota in continuous bioreactors. Three bioreactors, A, C, and E, served as control bioreactors and were treated with PBS, while bioreactors B, D, and F were treated with DSS to a final concentration of 1% starting on day 14 directly after sampling until the end of the experiment on day 20. We assigned all samples to a treatment group: before DSS treatment, Bioreactors B, D, and F, day 11 to day 14; DSS treatment, bioreactors B, D, and F, day 15 to day 20; before PBS control bioreactors A, C, and E, day 11 to day 14; and PBS control, bioreactors A, C, and E, day 15 to day 20.

2.1. The cultured microbiota represents the murine fecal microbiota both in terms of taxonomy and functionality

To determine whether the cultivated microbiota represented the murine colonic microbiota, we compared the cultivated microbiota from days 1, 11, 12, and 13 with the fecal microbiota from four female C57BL/6JRj wild-type mice from the same mouse facility using metaproteomics. All the samples were measured in one batch. Due to the significantly lower number of protein groups identified, we excluded the sample from bioreactor E day 13 from the analysis. The number of protein groups detected in fecal and cultivated microbiota was similar (feces = 8747 ± 694, day 11 = 8315 ± 376, day 12:8247 ± 505, day 13:8562 ± 341), but slightly lower on day 1 (5935 ± 595). In total 12,674 protein groups were identified with 85.4% detected in both bioreactor and feces samples, while 10.5% only detected in bioreactor samples and 4.1% only detected in fecal samples. At the taxonomic level, the protein groups were annotated to 70 bacterial genera (Figure 1a). Of these, 98.57% were shared between the cultivated and fecal microbiota. In the respective non-metric multidimensional scaling (NMDS) analysis, based on the number of protein groups annotated to genera, the cultured microbiota at day 1 formed a separate cluster, whereas the cultured microbiota from day 11 to day 13 and the fecal microbiota clustered together (Figure 1b) (PERMANOVA P = 0.002). On taxonomic level of genus, pairwise comparison of fecal samples with the samples from the different bioreactor days did show significant differences, although the magnitude of difference decreased over the course of cultivation (PERMANOVA: feces vs day1: F = 33.9, P = 0.008; feces vs day11: F = 21.9, P = 0.009, feces vs day12: F = 17.4, P = 0.09, F = 19.3; feces vs day13: F = 2.5, P = 0.008).

Figure 1.

Cultivated microbiota (day 1, day 11 and day 12 each n=6; day 13 n=5) reflect taxonomic and functional features of murine fecal microbiota (murine fecal samples n=4) based on metaproteomics. Total numbers of bacterial genera identified in the cultivated microbiota and fecal microbiota (a) and non-metric multidimensional scaling (NMDS) analysis of bacterial genera (b) based on the numbers of protein groups assigned to each genus. Number of KEGG orthologous (c) and number of KEGG pathways (d) identified in the cultivated microbiota and fecal microbiota. NMDS analysis of KEGG pathways (e) abundances based on the numbers of protein groups assigned to each pathway.

At the functional level, we identified 2056 different protein group functions based on Kyoto Encyclopedia of Genes and Genomes (KEGG) orthologs (KOs), i.e. specific protein functions. About 92.95% (1911) of these were shared between the cultivated and the fecal microbiota. 6.52% (134) were only detected in the cultivated microbiota, whereas 0.54% (11) were only detected in the fecal microbiota (Figure 1c). The detected KOs were assigned to 86 KEGG bacterial pathways, all of which were present in both the cultured and fecal microbiota (Figure 1d). Although all KEGG pathways were shared between fecal and cultured microbiota (Figure 1d), the NMDS analysis based on KEGG pathways revealed that the cultured microbiota at day 1 as well as the fecal microbiota clustered slightly away from microbiota cultures from day 11 to day 13 (Figure 1e). On the functional pathway level, pairwise comparison of fecal samples with the samples from the different bioreactor days did show significant differences, although the magnitude of difference decreased over the course of cultivation and was non-significant when compared to day 13 samples (PERMANOVA: feces vs day1: F = 36.9, P = 0.023; feces vs day11: F = 10.3, P = 0.032, feces vs day12: F = 9.3, P = 0.026, F = 19.3; feces vs day13: F = 6.9, P = 0.07). Depending on the ecosystem (bioreactor or gut) and time in culture (early/late), metabolic pathways are expressed at slightly different proportions. Nevertheless, the bioreactor samples represented the fecal microbial community at the taxonomic and functional levels to a large extent.

2.2. DSS does not affect the number of protein groups in the metaproteome of murine microbiota

We re-measured the samples from days 11 to 14, with samples from days 15 to 20, which were treated with either DSS or the PBS-control in one batch. A total of 11,017 protein groups were identified across all samples, with an average number of protein groups per sample of 7057 ± 480 SD. The mean number of protein groups before DSS treatment (6819 ± 783 SD) remained unaffected by DSS treatment (DSS treatment: 7154 ± 273 SD; Student’s t-test P = 0.286). Similarly, no change in the number of protein groups was observed in the control bioreactors (before PBS control: 7202 ± 233 SD; PBS control: 7065 ± 285 SD; Student’s test P = 0.241).

2.3. DSS treatment maintains the taxonomic structure of murine microbiota

We summed the relative protein group intensities per sample to determine the relative taxa abundance. Bacillota was most abundant phylum, accounting for ~50% of the microbiota (Figure 2a). Other abundant phyla were Pseudomonadota, Bacteroidota, and Verrucomicrobiota in decreasing order. Approxi-mately 25% could not be assigned to a single phylum. Upon DSS treatment, the relative taxa abundance did not change. At a higher taxonomic resolution, that is, at the genus level, we could not identify any shifts in taxonomic distribution associated with DSS treatment (Figure 2b). Hierarchical clustering based on z-scores did not reveal clustering of microbiota by treatment, but rather by the bioreactor. In addition, with hierarchical dendrogram clustering per bioreactor, we did not observe any clustering by treatment (Supplemental figure S1). To remove the bias resulting from the slightly different development of microbial communities in each bioreactor, we analyzed the relative abundance at the genus level as log₂ fold change per bioreactor to a baseline. As a baseline, we chose the microbial communities on day 13 as a proxy for the stabilization phase.

Figure 2.

Taxonomic composition of microbial communities. a: mean relative abundances of bacterial phyla based on relative protein group intensities for each day with either exposure to DSS or PBS (n= 3). b: clustered heatmap of bacterial genera abundances (z-scores). Top horizontal annotation bar depicts treatment, lower horizontal annotation bar depicts bioreactor of origin. Vertical annotation bar depicts phylum.

NMDS analysis of these log₂ fold changes did not reveal any segregation of samples due to DSS treatment (Figure 3a), nor could significant changes be observed in the corresponding PERMANOVA (P = 0.11) when compared to the PBS-treated controls. To identify potentially more nuanced changes induced by DSS, we performed a further NMDS analysis only on the DSS-treated bioreactors B, D, and F. Additionally, this analysis did not reveal any distinction to the control-treated bioreactors or to the community before DSS was applied (P = 0.619). To follow this up in more detail, we compared all bacterial genera between the treatment groups using the Kruskal–Wallis test. However, no single genus was altered significantly in relative abundance when comparing the four treatment groups (Figure 3b). Taken together, metaproteome analysis showed that the taxonomic composition of the cultured microbiota was not affected by DSS.

Figure 3.

Analysis of taxonomic change compared to the baseline (day 13) for each bioreactor. a: NMDS analysis of the log₂ fold changes of bacterial genera to baseline. b: significance analysis based on log₂ fold changes between the four treatment groups: before DSS-, DSS-treatment, before PBS and PBS-treatment calculated by Kruskal–Wallis test corrected for multi-testing using the Benjamini–Hochberg method.

2.4. The functional structure of murine microbiota is not influenced by DSS

Protein groups were assigned to KEGG orthologs (KOs) to determine their function. These were further annotated to KEGG subroles and pathways. Protein groups were assigned to 2182 KOs, 19 KEGG subroles, and 87 KEGG pathways. At low functional resolution, the most abundant KEGG subrole was carbohydrate metabolism in all bioreactors, independent of treatment and cultivation time (Figure 4). Other abundant sub-roles were translation, amino acid metabolism, and energy metabolism. No significant changes in abundance were observed in the KEGG subroles. We evaluated the KEGG pathways based on z-scores (Supplemental figure S2). None of the pathways were significantly altered after DSS treatment. Similar to the taxonomic structure, KEGG pathways (Supplemental figure S2) clustered by bioreactor and not by treatment. In addition, hierarchical clustering of the relative abundance of the KEGG pathway for each bioreactor did not show any clustering according to treatment (Supplemental figure S3). No KEGG pathways changed significantly in relative abundance during DSS treatment (Kruskal–Wallis test, followed by post-hoc Dunn test) (Supplemental figure S4). Thus, our data demonstrated that DSS had no effect on pathway abundance.

Figure 4.

Functional structure of the microbial communities, before and during exposure to DSS or PBS. Mean distribution of KEGG functional subroles based on summed relative intensities of protein groups for each day with exposure to DSS or PBS control from day 15 (n= 3).

Since the taxonomic analysis was performed on log₂ fold-changes, the same analysis was applied to the relative abundances of KOs. The NMDS analysis and PERMANOVA of these values did not reveal a global change (P = 0.579) in the functionality of the microbiota after DSS treatment (Figure 5). No clustering of samples nor significant changes in KEGG pathway log₂ fold changes (P = 0.433) before and during DSS treatment after based on Kruskal–Wallis test combined with Dunn post-hoc correction was observed (Figure 5, supplemental figure S2).

Figure 5.

NMDS analysis of functional change based on log₂ fold changes of KEGG orthologous (KO) compared to the baseline (day 13).

2.5. DSS does not impact the global metabolome

Generally, metabolites have a comparably high turnover compared to other molecule classes²³ and the introduction of xenobiotic compounds itself, as well as their metabolization, may alter the metabolite profiles of microbial communities²⁴. Thus, we analyzed whether DSS alters the global metabolome in extracellular medium during cultivation, independent of the lack of changes in KO pathways derived from the meta-analysis. We based our analysis on peaks measured in either the negative or positive ionization mode, with at least one possible identification. This resulted in the identification of 378 putative metabolites. We did not observe a clear clustering of metabolite profiles by bioreactor (Figure 6a), although the metabolite profiles from bioreactors A and F were separated from the other bioreactors. Moreover, the comparison before and during DSS treatment in the individual bioreactors did not result in the discrimination of these groups. Hierarchal clustering of metabolite profiles for each bioreactor did not show clustering according to the treatment (Supplemental figure S5). Similar to the metaproteome analysis, we eliminated bioreactor bias by data normalization to log₂ fold changes relative to day 13 per bioreactor. For a more comprehensive visualization of the global changes in metabolite profiles, NMDS was performed (Figure 6b), focusing on the before and during DSS and PBS control exposure. Metabolite profiles were separated from sample groups before and during DSS treatment (PERMANOVA, F = 5.76, P  = 0.001). However, we also observed a separation before and during PBS treatment (PERMANOVA: F = 2.55, P  = 0.032), indicating a shift in the global metabolome that was dependent on the cultivation duration. Thus, we tested whether metabolite profiles changed in a time-dependent manner. For this analysis, we extracted the Bray-Curtis (BC) similarities per bioreactor compared to day 13 and compared the mean BC similarities of DSS- and PBS-treated bioreactors over time (Figure 6c). On day 14, that is, before treatment, the mean BC similarity was highest for both DSS- and PBS-treated bioreactors. Over time, in both treatment groups, we observed a steady decline in BC similarity compared to day 13 (multiple t-tests with Welch correction in GraphPad Prism v9.4), indicating that not the exposure to DSS but the cultivation itself modified the metabolite profiles over time. Taken together, we showed that the global metabolome changes over time and that these changes occurred independently of DSS treatment.

Figure 6.

Intercellular metabolomics of the bacterial communities. a: clustered heatmap of normalized intensities (z-scores) of those peaks with putative identification. b: NMDS analysis of the log₂ fold changes of intercellular metabolites to day 13 baseline. c: Bray-Curtis (BC) similarity of untargeted metabolomics profiles relative to the day 13 baseline.

2.6. The short-chain fatty acid profiles are not influenced by DSS-treatment

Since the global metabolome comprises more than 350 putative metabolites, we also determined the effect of DSS treatment on a smaller subset of relevant metabolites, short chain fatty acids (SCFAs), the major fermentation products of intestinal microbiota.²⁵ We quantified nine SCFAs (acetate, propionate, butyrate iso-butyrate, 2-methylbutyrate, valerate, iso-valerate, caproate, and iso-caproate). Acetate, butyrate, and propionate were by far the most abundant SCFAs, with the other six SCFAs being less abundant. The mean SCFA profiles of the DSS- and PBS-treated bioreactors were very similar (Figure 7a). Thus, we analyzed the SCFA profiles as relative abundances to day 13. In line with these findings, NMDS analysis of these SCFA profiles confirmed the high similarity of SCFA profiles across all bioreactors, independent of DSS treatment (Figure 7b).

Figure 7.

Short-chain fatty acid analysis of the intercellular medium. a: mean relative abundance of short chain fatty acids. b: NMDS analysis of the log₂ fold changes of intercellular short-chain fatty acids concentrations to day 13 baseline.

3. Discussion

The DSS colitis model is the most frequently utilized murine model in IBA research. Nevertheless, it is not clear whether the microbial dysbiosis observed in colitic mice is induced by the DSS itself or results from the tissue damage and/or the subsequent inflammation.

The current understanding of the DSS mode of action is that after oral uptake, the negatively charged DSS polymer penetrates the intestinal mucosae. There, DSS immeadiately alters the expression of tight junction proteins and thereby destabilizes the epithelial layer.⁹ This potentially leads to DSS entering the circulation and, in acute colitis, distributing to various organs, such as the liver, the spleen, and the kidney.¹⁰ In a further study, it was shown that DSS polymers with a molecular weight of 5 kDa were depolymerized due to the acidic pH in the murine stomach. Since DSS has been detected in various organs, a rapid absorption of DSS monomers and distribution to liver and spleen has been proposed.²⁶ Interestingly, in the large intestine DSS was mainly detected in the original polymerized form. The authors proposed that though DSS was depolymerized and absorbed, the longer chain DSS induced the colitis in the large intestine.²⁶ Excretion occurs either from the kidney via urine¹⁰ or for intestinal DSS by excretion via the feces.²⁷ It is thought, that DSS affects the structure of the mucus layer and allows microbial penetration as soon as it reaches the intestine as a first event during colitis development.¹⁵ Subsequent epithelial injury by invading intestinal microbiota then induces immune cell infiltration and intestinal inflammation,¹⁷ which is accompanied by microbial dysbiosis.

The DSS-induced colitis model may represent a valuable tool to study the effect of intestinal inflammation on the resident microbiota, if DSS does not directly affect the microbiota composition and functions. Thus, we determined the direct effects of DSS on murine colonic microbiota in the absence of the host in vitro by continuous cultivation of murine fecal microbiota. In vitro models have proven suitable for investigating the consequences of various treatments such as drug exposure,^21,28–30 chemical exposure,^31–33 nutrient exposure,^34,35 or other perturbations^36,37 on intestinal microbiota. An advantage in our case, is that this experimental setup excludes the host-derived factors tissue damage and inflammation,³⁸ a downside is that it does not capture effects that derive from the changes and the loss of the mucus layer upon DSS exposure. We first analyzed, whether and to what extent the cultivated microbiota from our in vitro model represented the original fecal microbiota at the taxonomic and functional levels. The cultivated and fecal microbiota in our study shared >98% of taxa, indicating the presence of similar taxa in all the microbiota, resulting in a highly similar functionality of the bioreactor cultivated microbiota compared to the functionality of the fecal microbiota. Although we cultivated fecal microbiota, mucus-associated members of the microbiota have been cultured due to the presence of mucin in the culture medium, such as Akkermansia muciniphila and Bacteroides thetaiotaomicron.²⁸

In the present study, we exposed the microbiota to DSS at a final concentration of 1%. We assumed that a concentration of 1% DSS in vitro equals the administration of 2% DSS in vivo, as in the in vivo situation the orally administered DSS is 1:2 diluted with the food.³⁹ DSS with 36–50 kDa was used, since this type of DSS has been described to induce the most severe colitis in mice.^8,10,27 Exposure was limited to 6 d, as within this timeframe, clinical symptoms have been reported.⁴⁰ The severity and manifestation of colitis both depend on the utilized concentration of DSS, which ranges between 1% and 5% in the literature, and the duration of application, but also other factors, such as genetic background and intestinal microbiota.²⁷ Although the colitogenic effect of DSS was more pronounced at higher DSS concentrations, already at the low concentration of 1% DSS, clinical symptoms have been reported after 6 d of oral DSS administration.^{9,10,16,40–43} In addition, changes within the intestinal microbiota have been observed as early as day 2 after in vivo DSS exposure at 1%, 2%, and 3%.⁴⁰ In this study, until day 6 the murine microbiota changed independent of the DSS concentration, while at time points beyond day 6 a dose-dependent effect on the microbiota has been observed.⁴⁰ Biton et al. (2018) showed via magnetic resonance imaging that 1% DSS induced the most severe colitis with lowest animal mortality, resulting in the highest reproducibility of the assessed colitis symptoms¹⁶. Thus, the application of 1% DSS, as in our experimental setup, should induce a microbiota shift within a short time frame of 6 d, as has been demonstrated in previous in vivo studies.^16,40,43

In this study, we assessed the potential taxonomic and functional changes at the protein and metabolite levels. Surprisingly, the application of DSS did not affect the microbial communities at the taxonomic and functional levels in the respective bioreactors. The most prominent differences in taxonomic and functional levels were observed between individual bioreactors. This suggests that the bioreactors, in which the microbiota were cultivated, had a more dominant effect on community composition than DSS treatment. Our metaproteomic analysis revealed that DSS at 1% final concentration did not affect the number of protein groups or the relative protein abundances of common taxa and metabolic pathways in the murine communities. This is in strong contrast to a recent study using metaproteomics that investigated the taxonomic and functional changes upon DSS treatment (concentration: 1–2% DSS) in vivo.⁴⁴ Haange et al. (2019) showed that upon DSS treatment, the taxonomic composition of the intestinal microbiota was altered significantly, which was accompanied by changes of metabolic pathways.⁴⁴ Further, the intestinal microbiota of DSS-treated mice showed lower alpha diversity, a commonality of disease-associated, dysbiotic microbiota.⁴⁰ The authors observed that changes in the murine microbiota paralleled the pathophysiological alterations in the mouse intestines.⁴⁰ In line with these findings, Schwab et al. (2014) showed that host inflammatory markers and community changes in a DSS colitis model occurred concomitantly.⁴⁵ Therefore, it was proposed that during DSS-induced colitis, inflammation and not DSS initially promotes the observed changes in both host- and microbiota-related factors.^40,45 Since DSS treatment did not influence the microbiota composition or their functionality in our study, our results support the hypothesis of Park et al.⁴⁰ and Schwab et al., ⁴⁵in which the negatively charged polymer DSS penetrates the intestinal mucosae, alters the expression of tight junction proteins and destabilizes the epithelial layer.⁹ Following DSS – mediated epithelial injury, the intestinal microbiota infiltrates the underlying tissue and induces immune cell infiltration and intestinal inflammation,^8,17,40,45 which is accompanied with microbial dysbiosis. Although, the microbial dysbiosis may already result from the tissue-damage independent of the progressing inflammation. Colitis induction in germ-free mice proved that DSS-induced tissue damage and inflammation occur microbiota-independent, although the grade of inflammation is clearly higher in conventional mice harboring a complex microbiota.¹⁸ In our study, we did not detect any significant changes in the microbiome associated with DSS exposure, suggesting that changes in the microbiota observed in vivo DSS colitis models are associated with inflammation and tissue damage, and not directly attributed to DSS.

Mucus changes and epithelial breakdown may, in turn, allow the translocation of microbiota and microbial metabolites, leading to inflammation and microbial dysbiosis, upon which the intestinal microbiota essentially contributes to intestinal inflammation in DSS colitis models.^15,18

The major limitations of our study are the limited number of replicate bioreactors, which were inoculated with fecal microbiota derived from pooled feces of only female animals, which primarily represent the luminal microbiota. Further, we tested only one DSS concentration. Thus, we cannot exclude the possibility that higher DSS concentrations may directly affect the microbial communities, nor can we exclude that DSS does not alter the mucus-associated microbiota. However, our primary focus was to assess the effects of DSS at a typical concentration used in murine colitis models. The use of a multi-stage bioreactor setup, combining aligned bioreactor vessels that simulate not only the colonic environment but also the stomach and small intestine, similar to SHIME⁴⁶ may be advantageous. Further, fine-tuning of colonic vessel pH more closely mirrors the murine colonic pH 5⁴⁷ and a more rapid turnover rate could be applied.⁴⁸ Since male and female gut microbiota differ,⁴⁹ although mainly only on the strain level, repeating this experiment with male murine derived microbiota would be beneficial.

Using continuous cultivation of murine fecal microbiota, we showed that DSS exposure of intestinal microbiota does not affect microbial taxonomy or functionality in our setup. This supports the suitability of the DSS-induced colitis model for investigating host–microbiota interactions during colitis, particularly how intestinal injury drives the vicious cycle of dysbiosis and intestinal inflammation.

4. Methods

4.1. Microbiota preparation

The microbiota was isolated from four female C57BL/6JRj wild-type mice at the age–8–10 weeks that were originally purchased from Janvier Laboratories (Saint-Berthevin Cedex, France). During acclimatization for 2 weeks, animals were housed in the animal care facility of the Fraunhofer Institute for Cell Therapy and Immunology (Leipzig, Germany) in a temperature- and humidity-controlled room (23°C, 50% humidity) under specific pathogen-free conditions with 12 h/12 h of light/dark cycle and free access to pelleted standard rodent chow and water ad libitum. The experiment was conducted according to the European Communities Council Directive (86/609/EEC) and approved by the local authorities (registration no. T 10/17; Landesdirektion Sachsen, Leipzig, Germany). All efforts were made to minimize animal suffering. After sacrificing the mice with flow-controlled carbon dioxide (1 L/min), the intestinal tract was extracted. Fecal pellets were aseptically isolated from the colon. Immediately after collection, fecal pellets were stored under anaerobic conditions using an Anaerocult oxygen reducer system (Oxoid, Thermo Fisher Scientific, Waltham, USA) and kept at 4°C. To increase the microbial biomass for bioreactor inoculation, murine microbiota pooled from the four animals were enriched in 100 mL complex intestinal medium (CIM,³⁶ overnight at 37°C and 175 rpm shaking under anaerobic conditions.

4.2. Experimental setup of bioreactor cultivation

Prior to cultivation, a Multifors 2 bioreactor (Infors, Bottmingen-Basel, Switzerland), equipped with six independent 250 mL culture vessels (bioreactors A-F), was set up as described previously.³⁶ The bioreactors were filled with deionized water, assembled completely, and steam sterilized. The water was then replaced with 250 mL sterile culture medium.^36,50 To prove sterility, we simulated experimental conditions (37°C, 150 rpm, no medium feed) for 24 h prior to inoculation. After the sterile run, we inoculated all bioreactor vessels with 10 mL murine microbiota from an overnight culture to increase the microbial biomass for inoculation and reproducibility among replicate bioreactors. To prevent washing out of slower dividing bacteria, murine bacteria were cultured for 48 h in batch to establish before the medium feed was turned on at day 1.

The culture temperature was 37°C, resembling the body temperature of mice⁵¹ and the pH was adjusted to 6.5 by automatic addition of 1 M sodium hydroxide (NaOH). The system retention time was 24 h, after which the culture medium was fully exchanged. To prevent the settling of medium components and bacteria and to distribute nutrients in the culture, the cultures were stirred constantly at 150 rpm. The bioreactor system was maintained under anaerobic conditions by continuously gassing the bioreactor vessels and reservoir bottles containing the medium and NaOH with sterile nitrogen.

The murine microbiota was cultivated in chemostats from day 1 to day 13, that is, 12 × medium exchanges, to allow the microbiota to adapt to the in vitro culture conditions and to reach a constant community state. We defined a control phase before treatment, including days 12–14 before DSS treatment. During the data analysis, we normalized our data to day 13, which served as the baseline.

4.3. DSS-exposure

For our experiments, we chose an exposure to 1% DSS. This choice was based on the following assumptions: In the in vivo situation water and food are taken up in nearly equal amounts³⁹ leading to a 1:2 dilution of DSS in animal experiments, e.g. 2% DSS are diluted to 1% in the murine intestinal tract. Thus, we assumed that a direct spike-in of 1% DSS into an in vitro system, which will not be diluted 1:2, simulates the administration of 2% DSS via drinking water in vivo. At 2% DSS, clinical symptoms have been reported in C57BL/6 mice on day 5,¹¹ while changes in microbiota composition have already been observed on days 2.⁴⁰ When comparing the onset time of clinical symptoms across mouse strains, C57BL/6¹¹ and Balb/c⁵² developed clinical symptoms within the same timescale. Thus, we assumed that (i) DSS exposure for 6 d may be sufficient to investigate the microbiota-modulating properties of DSS, and (ii) the effects of DSS may be independent of the selected genotype.

Immediately after sampling on day 14, the DSS bioreactors B, D, and F were spiked with 10% DSS (36–50 kDa, colitis grade, MP Biomedicals, Santa Ana, CA, USA) in 25 mL PBS to reach a final concentration of 1% DSS in 250 mL total reactor volume. Simultaneously, the feed medium was exchanged to CIM + 1% DSS, and a concentration of 1% DSS were maintained until day 20, i.e. seven medium exchanges. Control bioreactors A, C, and E were spiked with the same volume of PBS and further supplied with CIM until day 20. Samples were taken on day 1 and as a multiple of 24 h during the control phase (day 12–14) and the early (day 15) and late treatment phases (days 18–20).

4.4. Metabolome analyses in culture supernatants

Upon sample centrifugation (3200 × g, 10 min, 4°C), cell-free culture supernatants were stored at −80°C for metabolomic analyses.

4.4.1. Untargeted metabolome analysis

In brief, culture supernatants were thawed, and 100 µL was mixed with 500 µL methanol:acetonitrile (ACN):water (2:3:1), followed by vortexing and sonication. After centrifugation, 100 µL of supernatant was dried in a SpeedVac vacuum concentrator (Eppendorf, Hamburg, Germany). Prior to measurement, the extract was resuspended in 100 µl 1% ACN and 0.1% formic acid in water.

For LC-MS/MS analysis, 10 µL of each sample was injected into a UPLC system coupled online with a 6540 UHD Accurate-Mass QTOF (both Agilent Technologies). Extracts were loaded on a C18 pre-column (ACQUITY BEH C18 1.8 µm, 2.1 × 50 mm), and separation was achieved on a C18 column (ACQUITY UPLC HSS T3 1.8 µm, 2.1 × 100 mm). Metabolites were eluted at a constant flow rate of 0.3 mL/min with a binary solvent system of A (0.1% formic acid in water) and B (0.1% formic acid in ACN). The gradient was as follows: 0–5 min 1% B, 5.1–20 min 1%–100% B, 20.1–25 min 1% B. All samples were acquired in positive and negative ionization mode. The QTOF was set up in the centroid mode with a scan range of 60–1000 m/z. After each full scan, the five most intense ions (threshold of 200 counts) were subjected to fragmentation.

Raw data were obtained (. d-files) were imported into the Progenesis QI® software (version2.1, Waters Corporation). Each ionization mode was processed separately using generic workflow. The adduct ions involved [M+H], [M+H-H₂O], [M+H-2 H₂O], and [M+ACN+H] in the positive mode and [M-H], [M-H2O-H], and [M-H+ACN] in the negative mode. Chromatograms were aligned in the t_R direction. A reference chromatogram was automatically chosen from the dataset. The following software-guided peak-picking tool resulted in a data matrix that included the retention time, mass-to-charge ratio, and corresponding normalized peak area. A subsequent database search was performed using the built-in ChemSpider plug-in. The fecal metabolome (6738 compounds), E. coli metabolome (755 compounds), and KEGG compound (19,090 compounds) databases were used for identification. The precursor and fragment mass tolerances were set at 10 ppm. The genesis and fragment scores were set to ≥40 and ≥5, respectively. The resulting feature matrix was exported for further analysis. Features without any putative identification were removed, as well as features with higher peak areas in solvent blank samples than the mean peak area in the dataset.

4.4.2. SCFA analysis

The SCFA analysis was performed as previously described.⁵³ In brief, samples were mixed with ACN to a final concentration of 50%. SCFAs were derivatized with 200 mM 3-nitrophenylhydrazine and 120 mM N-(3-dimethylaminopropyl)-N‘-N-ethylcarbodiimide hydrochloride in pyridine and then diluted in 10% ACN. The diluted SCFA derivatives were injected into an RSLC UltiMate 3000® system coupled online with a QTRAP 5500® mass spectrometer. Chromatographic separation of the SCFAs was performed using a Waters ACQUITY UPLC BEH C18 column (2.1 × 100 mm, 1.7 µm). LC was performed at a constant flow rate of 0.35 mL/min with solvent A (0.01% formic acid in water) and solvent B (0.01% formic acid in ACN). For identification and quantitation, a scheduled MRM method was used with specific transitions for every SCFA. Data acquisition and peak integration were performed using the Analyst® software (version 1.6.2). The concentration was calculated using external calibration curves, and statistics were performed using the R software program.

4.5. Metaproteome analysis

As described previously³¹, thawed bacterial pellets were dissolved in 1 mL of lysis buffer (10 mM Tris-HCl, 2 mg/mL sodium chloride, 1 mM PMSF, 4 mg/mL SDS). For cell disruption following steps were applied: 1) bead beating (FastPrep-24, MP Biomedicals, Santa Ana, USA: 5.5 ms, 1 min, 3 cycles), 2) 15 min incubation at 60°C (Thermomixer comfort 5355, Eppendorf, Germany) and 3) Ultra-sonication (UP50H, Hielscher, Germany; cycle 0.5, amplitude 60%). Protein concentration was determined using a bicinchoninic acid assay according to the manufacturer’s instructions (Pierce™ BCA Protein Assay Kit, Thermo Fisher Scientific, Waltham, USA). Protein (100 µg) was precipitated in acetone 1:5 (v/v) at −20°C overnight and then centrifuged (10 min 14,000 × g). The precipitate was dissolved in Laemmli buffer and used for SDS-PAGE analysis, in-gel digestion, and protein purification with ZipTip® treatment.⁴⁴

From each sample, 5 µg of peptide lysate was injected into a nanoHPLC system (UltiMate 3000 RSLCnano, Dionex, Thermo Fisher Scientific, Waltham, USA). The peptides were separated on a C18 reverse-phase trapping column (C18 PepMap100, 300 µm × 5 mm, particle size 5 µm, nano viper, Thermo Fischer Scientific), followed by a C18 reverse-phase analytical column (Acclaim PepMap® 100, 75 µm × 25 cm, particle size 3 µm, nanoViper, Thermo Fisher Scientific). Solvents for nanoHPLC gradient were solvent A (0.1% formic acid in MS-grade Water) and solvent B (80% acetonitrile, 0.08% formic acid in MS-grade water). A linear gradient was run from 4% solvent B for 5 min to 55% solvent B for 125 min. Mass spectrometric analysis of the peptides was performed on a Q Exactive HF mass spectrometer (Thermo Fisher Scientific) coupled to a TriVersa NanoMate (Advion, Ltd., Harlow, UK) source in the LC chip coupling mode. The LC gradient, ionization mode, and mass spectrometry mode have been described elsewhere.⁵⁴ Briefly, eluting peptides were measured in positive ionization with MS parameters set to FWHM 12 s, MS resolution 120,000, AGC target 3e,⁶ maximum injection time 120 ms, and scan range 350 m/z to 1550 m/z. For M/MS parameters, the following values were set: resolution 30,000, AGC target 2e,⁵ maximum injection time 150 ms, isolation window of precursors 1.2 m/z and TopN set to 15. Raw data were processed using Proteome Discoverer (vr2.2, Thermo Fisher Scientific). The search settings for the Sequest HT search engine were set as follows: trypsin (full), max. Missed cleavage: 2; precursor mass tolerance: 10 ppm; fragment mass tolerance: 0.02 Da. Spectrum identification was performed as described.⁵⁴ For protein identification, a previously described target database (Haange et. al, 2019) containing 7,495,542 protein sequences was used. To calculate the false discovery rate (FDR) of identification, a decoy database was used by reversing all sequences. The FDR for peptide identification was set at 1%. The peptides were annotated as possible proteins. Finally, all the possible proteins identified with the same set of peptides were clustered into protein groups. Only the protein groups with at least one unique peptide were used for further analysis.

4.6. Data analysis and statistics

For metaproteomics data handling as well as annotating functions and taxonomy to protein groups, in-house written R-scripts were used.⁵⁴ For functional annotation to protein groups, all sequences of all possible protein identifications were first uploaded to the web application GhostKOALA from the Kyoto Encyclopedia of Genes and Genomes (KEGG) to assign KEGG orthologs.⁵⁵ Then, functions were annotated to the protein groups if only a single KEGG ortholog was assigned to the proteins belonging to the protein group and less than 30% of proteins had unknown functions. The protein groups were then annotated to KEGG pathways and subroles. Only pathways with at least five assigned protein groups and a functional coverage of at least 15% were further investigated. Taxonomic annotation of the protein groups was performed by assigning the lowest common ancestor determined from all proteins belonging to the protein group.

Statistics and visualization of data were performed in R. Briefly, non-metric multidimensional scaling analysis (NMDS) was performed using basic R functions and the vegan package.^56,57 For the complete sample data analysis, PERMANOVA was performed using the adonis function in the vegan R package. For single variables, the Kruskal-Wallis group test followed by a post-hoc pairwise Dunn test was applied if not stated differently. Where appropriate (number of tests >20), P-values were corrected for multi-testing using the Benjamini–Hochberg method.⁵⁸ Heatmaps were constructed using the pheatmap R package (version 1.10)⁵⁹ and all other figures were constructed using the R package ggplot2.⁶⁰

Funding Statement

Furthermore, we thank the German Federal Environmental Foundation for the financial support from Jannike Lea Krause. Stephanie Schäpe is grateful for the support from a DFG grant within Priority Program 1656. Beatrice Engelmann is grateful for funding from the Novo Nordisk Foundation (grant number NNF21OC0066551). Hyun-Dong Chang was supported by Dr Rolf Schwiete Foundation and DFG Project-ID 375876048 – TRR 241. Lehmann acknowledges partial funding from the Fraunhofer Cluster of Excellence Immune-Mediated Diseases (CIMD). Martin von Bergen acknowledges partial funding from the DFG Priority Program of 2002.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Author contributions

JLK, SSS, NJ, and SH conceptualized the study. JLK, BE, and SH wrote the first draft of this manuscript. MH, US, and JL provided the murine microbiota. JLK performed microbiological experiments. BE and URK were responsible for the untargeted metabolomics, SCFA measurements, and data analysis. SSS, NJ, and SH were responsible for the metaproteome measurements and data analysis. ACZ, HDC, NJ, JL, GH, JL, SR, and MvB provided helpful discussions and revised the manuscript accordingly.

Availability of data and materials

The metaproteome and metabolome datasets supporting the conclusions of this article are available at ProteomeXchange with identifier P×D038429 and Metabolomics workbench with Study IDs ST002394 (SCFA) and ST002393 (untargeted metabolomics).

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/19490976.2023.2297831