Essentially, experiments were carried out as described previously (Herrmann et al., 2017b). In brief, fresh fecal pellets from eight healthy C57BL/6J mice were collected within 5 h of defecation. Since the aim of our study was to establish RNA-SIP for identification of murine fecal bacteria able to utilize RS instead of determining inter-individual variation in the populations of these bacteria, we opted for a pooling strategy. Animals were bred and housed at the animal facility of the University of Ulm in a specific pathogen-free (SPF) environment and given a standard laboratory diet with water ad libitum. Mice were routinely bred to maintain the colony of C57BL76 mice at the animal facility of the university and are not used for animal experimentation unless under a specific ethical approval for other studies. Collection of fecal samples from the cages does not constitute an invasive treatment, physical distress or even pain to animals and is thus not subject to ethics approval. Collected fecal pellets were pooled (total weight 1.9 g) and mixed with 12.7 mL of sterile, deoxygenated, and pre-warmed (37°C) M9 minimal medium (Smith and Levine, 1964) without glucose supplemented with 2 mg L⁻¹ thiamin and 1 g L⁻¹ casamino acids resulting in a 15% (w/v) fecal slurry (0 h, control). Deoxygenation of the medium prior to experiments was achieved by incubation in airtight jars under anaerobic conditions generated using AnaeroGen sachets (Merck, Darmstadt, Germany) 24 h prior to experiments. For each incubation, 1 mL of slurry material was diluted 1:2 with M9 minimal medium containing either 14.4 mg [¹²C] or 15 mg of uniformly (98%) labeled [U¹³C] potato starch, yielding a final concentration of 40 mM glucose unit equivalents and 7.5% (w/v) of fecal material. Both [¹²C] and [U¹³C]starch were obtained from the same manufacturer (IsoLife, Wageningen, Netherlands), were isolated from the same potato variety, and therefore presumably contained similar levels of RS. Diluted fecal material was incubated with the native [¹²C] or [U¹³C]potato starch in single 15-mL reaction tubes at 37°C for 2 h or 4 h in airtight jars under anaerobic conditions generated using AnaeroGen sachets (Merck, Darmstadt, Germany). AnaeroGen sachet will reduce the oxygen level in the jar to below 1% within 30 min (information by the supplier). Incubation times of 2 and 4 h were chosen on the basis of recent experiments from a previously published RNA-SIP study of our group with murine fecal slurries and labeled glucose as substrate (Herrmann et al., 2017b). The results of these preliminary experiments showed that even with an easily accessible substrate such as glucose, sampling times shorter than 2 h after addition of the substrate yielded only negligible amounts of labeled RNA. This indicates that the earliest time point for detection of bacteria that have incorporated the isotope label into RNA (and thus represent glucose utilizing bacteria) is 2 h. Based on these observations, 2 and 4 h were considered reasonable time points to identify RS-utilizing bacteria. Incubations for each condition (0 h, 2 h-¹²C, 2 h-¹³C, 4 h-¹²C, 4 h-¹³C) were carried out in technical duplicates using the same fecal slurry.

Total RNA from each sample was extracted and residual genomic DNA was removed as described before (Herrmann et al., 2017b). After elution and quantification, absence of DNA was verified by PCR and total RNA was subsequently loaded into a cesium trifluoroacetate (CsTFA) centrifugation solution for density-dependent resolution by gradient ultracentrifugation following a previously published protocol (Herrmann et al., 2017a,b). Afterwards, gradients were fractionated and the buoyant density (BD) of each fraction was determined as previously described (Herrmann et al., 2017b). Briefly, CsTFA (average BD = 1.793 g mL⁻¹) was loaded with ~700 ng RNA, filled into 8 mL Quick-Seal polypropylene tubes (BeckmanCoulter Inc., Krefeld, Germany) which were subsequently sealed. Sealed tubes were placed in an MLN-80 rotor in an Optima MAX-XP bench-top ultracentrifuge (both BeckmanCoulter Inc.) and subjected to ultracentrifugation at 123,100 × g (45,000 rpm) and 20°C for 67 h to establish density gradients. After centrifugation, density gradients were fractionated into 16 equal fractions (each 0.5 mL) by displacement with water from the top of the tube under a consistent flow rate of 1 mL min⁻¹ using a syringe pump (World Precision Instruments, Berlin, Germany). BD of each fraction was determined by measuring the refraction of the solution using a refractometer (Reichert, Depew, NY, USA) and correlation to a previously established calibration curve. For subsequent analysis, RNA from each fraction was precipitated, washed and re-dissolved in nuclease-free water.

The amount of bacterial 16S rRNA in each gradient fraction was determined by quantitative reverse transcription polymerase chain reaction (RT-qPCR) (Lueders et al., 2004) in a two-step assay. Total RNA in 10 μL of each gradient fraction, was reverse transcribed to complementary DNA (cDNA) using the SuperScript VILO cDNA Synthesis Kit (Life Technologies, Darmstadt, Germany) according to the manufacturer's protocol. The cDNA was then used as template in qPCR reactions with universal bacterial primers F_Bact 1369 (5′-CGG TGA ATA CGT TCC CGG-3′) and R_Prok1492 (5′-TAC GGC TAC CTT GTT ACG ACT T-3′) (Furet et al., 2009) on a Light-Cycler® 480 system (Roche, Mannheim, Germany). A PCR reaction of 25 μL contained 12.5 μL of 2x Maxima SYBR Green/ROX qPCR Master Mix (Life Technologies), 0.15 μL of each primer (50 μM; Metabion, Planegg/Steinkirchen, Germany), 0.25 μL bovine serum albumin (BSA; 20 mg mL⁻¹; Roche), 9.95 μL of nuclease-free water and 2 μL of cDNA template. Amplification was carried out with the following thermal profile: 95°C for 10 min followed by 40 cycles of 95°C for 15 s (denaturation), 60°C for 30 s (annealing), and 72°C for 30 s (elongation). Serial dilutions of purified 16S rRNA gene amplicons of Escherichia coli K12 were used as an internal standard. Nucleic acid concentrations were calculated against the standard curve using the Light-Cycler® 480 software (version 1.5). To allow comparison between different gradients and samples, the 16S rRNA content in each fraction was normalized according to a procedure used by us (Herrmann et al., 2017b) and others (Lueders et al., 2004; Hatamoto et al., 2007). RNA content in different fractions is expressed as proportion (%) of the RNA content in the fraction of the same sample with the highest RNA concentration, which was set to 100%.

In order to identify the bacterial populations involved in RS assimilation, 16S rRNA gene amplicon libraries were constructed from the density-resolved and reverse-transcribed rRNA. Figures 1B–D shows the density fractions which were selected for sequencing. Gradient fractions 8–10 (Figures 1B–D) contained RNA in all preparations. Thus, they presumably represented unlabeled RNA and were designated “light” SIP fractions. A discernable amount of RNA in gradient fractions 3–5 (Figures 1B–D) were only obtained from the [U¹³C]starch incubations. These fractions presumably contained heavy, [¹³C]-labeled RNA of bacteria that actively metabolized labeled [U¹³C]starch and were designated “heavy” SIP fraction. Sequencing libraries of the respective density fractions were constructed from all incubations.

Amplicon libraries representing the V3 and V4 regions of the bacterial 16S rRNA gene were prepared as previously described (Herrmann et al., 2017b). Briefly, 16S rRNA genes were amplified from cDNA of gradient fractions in a first PCR step using the locus-specific primer set S-D-Bact-0341-b-S-17 and S-D-Bact-0785-a-A-21 (Klindworth et al., 2013), to which overhang adaptor sequence tails were added. After purification of the obtained PCR products a second PCR step was performed to anneal unique dual-index barcodes with Illumina sequencing adaptors (Nextera XT index kit; Illumina, Eindhoven, Netherlands) to the amplicon target. The obtained 16S amplicon libraries were subjected to bead-purification using Agencourt AMPure XP beads (BeckmanCoulter Inc.) and size of the amplicons (~630 bp) was verified with a Bioanalyzer DNA 1000 chip (Agilent Technologies GmbH, Waldbronn, Germany). Quantities of each library were fluorometrically assessed using the Qubit dsDNA HS assay kit (Life Technologies). Finally, the uniquely indexed 16S amplicon libraries were pooled in equimolar amounts (4 nM) and sequenced in technical duplicates on an Illumina MiSeq platform (Illumina, Eindhoven, Netherlands) by Eurofins Genomic (Eurofins Genomic, Ebersberg, Germany) using the Illumina MiSeq 600 cycle Kit_V3 chemistry (2 x 300 bp PE; Illumina).

Sequencing data were processed with QIIME 1.8 (Caporaso et al., 2010) as previously described (Herrmann et al., 2017b). To measure the bacterial diversity in the different density fractions, diversity analyses were performed using the core_diversity_analyses.py script. Faith's phylogenetic diversity estimate was calculated using the minimum number of reads (6397) across 10 iterations.

Statistical analyses of the sequencing data were performed using R 3.3.2 (R Core Team, 2015) to analyze the community structure represented by the cDNA of the 16S rRNA species present in the “heavy” and “light” fractions per time point and treatment; all presented results on the taxonomic composition were first averaged across the sequencing duplicates to obtain the community per fraction. Respective fractions were then averaged across both incubation duplicates (except for fraction 8, which is represented by only one incubation) and finally across the designated different density fractions (“heavy” and “light”), resulting in a total of seven different observation groups, each represented by three fractions (0 h “light,” 2 h [¹²C] “light,” 4 h [¹²C] “light,” 2 h [¹³C] “light,” 2 h [¹³C] “heavy,” 4 h [¹³C] “light,” and 4 h [¹³C] “heavy”).

Differences in the prevalence of bacterial taxa in the “light” SIP fractions during the course of the ¹²C-control incubations were assessed using non-parametric permutation ANOVA with time as factor. Differences in the prevalence of bacterial taxa between “heavy” and “light” SIP fractions of the [U¹³C]starch treatments were assessed using two-factor permutation ANOVA with time and density as factors. Both tests were implemented using the perm.anova function in the RVAideMemoire package for R (Hervé, 2015) with 1000 permutations. Adjustment of P-values for multiple testing was performed using the Benjamini & Hochberg false discovery rate (FDR) method, with FDR < 0.05 considered significant. Differences in alpha diversity were analyzed using two-factor ANOVA with incubation time and density as factors. Differences with a P-value < 0.05 were considered significant. Hierarchical cluster analysis of bacterial profiles was performed using distances calculated from centered Pearson's correlation and average linkage clustering.

All sequencing data were submitted to GenBank and are publicly available under the accession number PRJNA376059.

Total concentration and ¹³C-enrichment of lactate, acetate, propionate, butyrate and isobutyrate stemming from bacterial fermentation in the fecal slurries were monitored using HPLC-C-IRMS (Thermoquest, Bremen, Germany) as described before (Conrad et al., 2007; Liu and Conrad, 2011).

Concentrations and retention times of acetate, propionate, butyrate and isobutyrate were determined by comparison with unlabeled standards. The isotopic signal of the ¹³C/¹²C ratio detected in the IRMS was calibrated with a CO₂ gas standard, which was referenced against a methyl stearate working standard and calibrated at the Max Planck Institute for Biogeochemistry, Jena, Germany (courtesy W.A. Brand). The proportion of labeled SCFA was calculated as atom percent excess (APE). This was achieved by measuring concentration of ¹³C-labeled SCFA in untreated and [¹²C]starch-treated samples. Consistent with the natural abundance of ¹³C, a stable proportion of 1.08% of all SCFA in these samples was ¹³C-labeled. This value was subtracted from the percentage of labeled SCFA in samples of [U¹³C]starch incubations to obtain APE.

In order to identify bacteria able to utilize RS, slurries prepared from fresh murine fecal pellets were incubated with [U¹³C]potato starch and incorporation of the ¹³C-label into bacterial RNA was determined. The fractionated centrifugation gradients showed decreasing average BDs (Figure 1A) ranging from 1.844 g mL⁻¹ (fraction 1) to 1.726 g mL⁻¹ (fraction 16) which indicated an adequately linear gradient formation.

In gradients containing only unlabeled [¹²C]RNA species originating from fresh fecal slurry (0 h; Figure 1B) and from slurries incubated with [¹²C]starch for 2 h (Figure 1C) and 4 h (Figure 1D), an almost identical RNA distribution pattern was observed: The bulk of the unlabeled RNA species was present in low density fractions (≤1.796 g mL⁻¹; ≥fraction 8) with peak amounts detected in fraction 9 (1.792 g mL⁻¹) after 0 h and 2 h of incubation and in fraction 10 (1.787 g mL⁻¹) after 4 h of incubation. In gradients containing RNA extracted from fecal slurries incubated with the [U¹³C]starch, RNA content slightly shifted toward higher density fractions (≥1.796 g mL⁻¹; ≤fraction 8) with the highest concentration accumulated in fraction 8 and 7 (1.796–1.801 g mL⁻¹) after 2 and 4 h of incubation respectively (Figures 1C and D). Moreover, in these samples a second, smaller peak in RNA quantity was observed in fraction 3–5 (1.812–1.825 g mL⁻¹).

Sequencing of the selected RNA-SIP fractions generated a total of 2,407,933 paired end sequences with an average of 32,985 sequences per sample (minimum 6,397, maximum 109,654, standard deviation 20,421). The obtained 16S rRNA gene sequences in the entire dataset were affiliated with nine phyla, 16 classes, 25 orders, 52 families and 98 genera.

“Light” fractions collected from the unlabeled [¹²C]starch fermentations were analyzed to show how the overall fecal community changed during the course of incubation. In line with a previous study using fecal slurries of animals housed in the same facility (Herrmann et al., 2017b), Firmicutes was the dominant phylum in fresh fecal content followed by Bacteroidetes and Proteobacteria (Table 1, 0 h). Verrucomicrobia, Actinobacteria, Tenericutes and Deferribacteres were detected at low frequencies, and 0.4% of the sequences remained unclassified. Clostridia was the most prevalent class with sequences affiliated to unclassified Lachnospiraceae, unclassified Clostridiales, Dorea and unclassified Ruminococcaceae dominating the community (Figure 2A, 0 h “light”). In almost equal relative frequencies, Bacilli and Bacteroidia were the second most dominant classes with Lactobacillus and unclassified Porphyromonadaceae as their most abundant representatives (Table 1 and Figure 2A). Structural changes in the bacterial community were observed over the course of incubation in the ¹²C-control fermentations (Table 1 and Figure 2A). A significant drop in Firmicutes (FDR = 0.02), mainly caused by reduction of the abundant unclassified Lachnospiraceae, was observed. This drop was accompanied by an increase in the relative abundances of the majority of the other taxa detected. The most prominent increase in abundance was observed for Bacteroidetes (FDR = 0.02) and Proteobacteria (FDR = 0.064). However, community profiles were still dominated by Firmicutes.

As a first step toward the identification of bacterial populations involved in RS fermentation, community profiles in “heavy” and “light” density fractions of the ¹³C-fed cultures were analyzed for proportional differences of individual taxa (Wüst et al., 2011; Young et al., 2015). Previous studies have shown that the amount of ¹³C in RNA is directly linked to the bouyant density and RNA molecules with different proportions of ¹³C can be separated by density centrifugation and fractionation (Manefield et al., 2002b). Moreover, nucleic acid samples from RNA- and DNA-SIP experiments with increased BD also had an enrichment in ¹³C-contents, i.e., higher density fractions indeed contained ¹³C-labeled RNA/DNA, as measured independently by IRMS (Manefield et al., 2002a; Shao et al., 2014). We therefore considered it reasonable to assume that “heavy” fractions from samples incubated with [U¹³C]starch contained isotope labeled RNA. Sequencing profiles revealed a complex bacterial community structure consisting of many taxa in the “heavy” fractions after 2 and 4 h of incubation in the presence of [U¹³C]starch (Figure 2B). However, in these fractions alpha diversity was significantly (P < 0.01) lower compared to the corresponding “light” fractions at both time points (Figure 2C). Moreover, hierarchical clustering analysis showed bacterial profiles from the “heavy” fractions were clearly separated from the profile of the “light” fractions as well as a delineation between the “heavy” fractions of both time points (Figure 3).

At the phylum level, the relative abundance of Bacteroidetes was considerably increased in “heavy” SIP fractions (FDR = 0.003) compared with their “light” counterpart. Similarly, proportions of Proteobacteria were also enriched in “heavy” gradient fractions (FDR = 0.017), whereas a significant decrease in Firmicutes was detected (FDR = 0.005) (Table 2 and Supplementary Figure S1).

Among the 40 most abundant taxa, identified at the lowest taxonomic level available, 23 showed a significantly increased prevalence in the “heavy” SIP fractions (FDR < 0.05) compared to the “light” fractions. After 2 h of incubation, the taxa that dominated the community were assigned to unclassified Ruminococcaceae, unclassified Prevotellaceae and Prevotella spp. (FDR = 0.005; Table 2 and Supplementary Figure S1). In addition, a diverse array of sequences affiliated with Butyricicoccus, Bacteroides, Bacillus (all FDR = 0.006), Oscillibacter (FDR = 0.007), Barnesiella (FDR = 0.015), Helicobacter (FDR = 0.025) and, to a lesser extent, Ruminococcus spp. (FDR = 0.051) showed increased relative abundances in the “heavy” fractions (Table 2). Moreover, unclassified Ruminococcaceae (FDR = 0.035) and, to a lesser extent, Ruminococcus (FDR = 0.109) increased in relative proportions between 2 and 4 h of incubation (Table 2).

Based on a significant interaction P-value obtained between density and time (all P ≤ 0.044), Allobaculum (FDR = 0.013), unclassified Porphyromonadaceae (FDR = 0.014), unclassified Clostridia (FDR = 0.05) and, to a lesser extent, unclassified Desulfovibrionales (FDR = 0.053) were detected in increased proportions in the “heavy” fractions after 4 h of incubation in the presence of the [U¹³C]starch (Table 2). The increase of these taxa was associated with a decrease in the relative abundance of unclassified Clostridiales, unclassified Firmicutes (both FDR = 0.005), unclassified Bacteria (FDR = 0.013), unclassified Lachnospiraceae (FDR = 0.021) and some minor abundant taxa (Table 2 and Supplementary Figure S1).

HPLC-IRMS analysis was used to trace the ¹³C-label derived from the [U¹³C]starch into fermentation metabolites produced by the murine fecal microbiota from the same incubations as used for the RNA isolation. An almost identical pattern was observed in fermentations with [U¹³C]starch and [¹²C]starch with acetate (~12 mM), propionate (~4 mM), and butyrate (~2 mM) being the most abundant SCFA. The branched-chain SCFA (BCFA) iso-butyrate was also detected, albeit in lower amounts (Table 3). Interestingly, lactate (0.61 mM) was only detected in fresh fecal slurry at the start of the fermentation process. However, during the course of fermentation, no considerable changes in concentrations of SCFA could be observed (Table 3). Nevertheless, incorporation of the isotope-label from the [U¹³C]starch into the SCFA, expressed as atom percent excess (APE), could be detected consistently (Table 3). After 2 h, 0.95, 1.46, and 0.39% of labeled acetate, propionate and butyrate, respectively, was derived from RS. These proportions increased to, 2.00, 2.75, and 1.11%, respectively, after 4 h of incubation.

The 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.