Study population.

This cross-sectional analysis included the baseline data of 48 healthy, normoglycemic lean-to-overweight Caucasian women and men aged 20–55 yr prior to participating in a human intervention study at the Maastricht University Medical Centre+ (MUMC+) aiming to improve colonic transit time by means of a prebiotic fiber supplementation (29). Normoglycemia was determined by oral glucose tolerance test (OGTT) according to the diagnostic criteria of the American Diabetes Association, 2010 (1). Initial recruitment was based on a defecation frequency of ≤4 per week. Participants with colonic transit time >35 h determined by radiopaque marker method were included. Exclusion criteria were related to outcomes of the interventional study (29): two or more symptoms of constipation according to ROME III criteria, diagnosis of T2DM, prior abdominal surgery, cardiovascular or gastrointestinal diseases, or the use of antibiotics, laxatives, prebiotics, or probiotics 3 mo before the study or during study participation. Participants did not use β-blockers, lipid- or glucose-lowering medication, or corticosteroids. All participants gave written consent before inclusion. The study protocol was approved by the MUMC+ Medical Ethical Committee and was conducted in accordance with the Declaration of Helsinki (revised version, October 2008). During an initial screening, the participant’s medical history and anthropometrics were assessed and an OGTT was performed. On clinical investigation day 1 (CID1), (segmental) colonic transit was measured. A week after CID1, participants came fasted to the university for blood sampling and to deliver stool samples for CID2.

Segmental and total colonic transit time.

On 6 consecutive days, participants ingested 10 radiopaque markers (Colonic Transit, P. & A. Mauch, Münchenstein, Switzerland) in the morning. On day 7, abdominal radiographs were taken in the morning of CID1, 24 h after ingestion of the last marker. Based on the visible number of markers on the abdominal X-ray, total colonic transit time was calculated as follows: sum of markers on X-ray/[24 (time in hours between ingestions) × 10 (daily dosage of markers)]. Segmental transit times of ascending, descending, and rectosigmoid colon were assessed using anatomical landmarks as follows: ascending colonic transit time was calculated by the sum of markers in the area between an imaginary line right of the vertebral spinous processes from the fifth lumbar vertebra to the pelvic outlet. Descending colonic transit time was calculated by the sum of markers in the area between an imaginary line left of the vertebral spinous processes and the imaginary line above the fifth lumbar vertebra and the left anterior superior iliac crest. Rectosigmoid transit time was calculated by the sum of markers in the area under the imaginary line from the pelvic brim on the right to the superior iliac crest on the left (Fig. 1) (5).

Dietary intake and stool consistency and frequency.

Participants were instructed by a dietician to weigh and document food intake on two weekdays and one weekend day. Energy and macronutrient intake were analyzed using the Dutch food composition database (Nederlands Voedingsstoffenbestand 2016/5.0, National Institute for Public Health and Environment, Ministry of Health, Welfare and Sport, The Hague, The Netherlands). Participants completed a 7-day gastrointestinal questionnaire reporting on stool frequency and stool consistency using the Bristol stool scale (BSS).

Blood sampling and biochemical analysis.

One week after CID1, participants were instructed to come to the university after an overnight fast (10–12 h) for CID2. After a cannula was inserted into the antecubital vein, blood samples were taken during the fasted state. Blood was collected in prechilled EDTA tubes (0.2 mol/L EDTA; Sigma, Dorset, UK) for SCFA and LBP. SCFA in plasma were measured using liquid chromatography-mass spectrometry, as reported previously (41). For the detection of LBP, plates (Greiner Mocrolon 600 high binding; Sigma Aldrich, St. Louis, MO) were coated with polyclonal anti-human LBP antibodies. Diluted plasma samples (1:5,000) and a standard dilution series with recombinant LBP were added to the plate. Detection occurred with a biotinylated polyclonal rabbit anti-human LBP IgG, followed by peroxidase-conjugated streptavidin and substrate. The detection limit for the LBP assay was 200 pg/mL.

Fecal short- and branched-chain fatty acids and calprotectin.

Fecal samples were collected at home and stored in the participants’ freezer at −20°C, transported on dry ice, and stored on arrival at the university at −80°C on CID1. Fecal SCFA (acetate, propionate, butyrate, valerate, caproate) and BCFA (isobutyrate, isovalerate) were measured by gas chromatography-mass spectrometry (Dr. Stein and Colleague Medical Laboratory, Mönchengladbach, Germany), as described previously (18). Fecal calprotectin was determined using a commercial enzyme-linked immunosorbent assay (BÜHLMANN fCAL, Basel, Switzerland).

Microbiota composition.

DNA was isolated from 0.25 g feces with repeated bead beating followed by automated isolation and purification using a Maxwell 16 Tissue LEV Total RNA Purification kit (Promega, Madison, WI). The V4 region of the 16S rRNA gene was amplified with a double barcoded primer pair 515F (5′-GTGCCAGCMGCCGCGGTAA)-806R (5′-GGACTACHVGGGTWTCTAAT) as previously described (34). Each sample was amplified in triplicate using Phusion hot start II high fidelity polymerase (Thermo Scientific). The 40-μL PCR reactions contained 28.4 μL nucleotide-free water (Promega, Madison, WI), 0.4 μL of 2 U/μL polymerase, 8 μL of 5X Phusion Green HF Buffer, 0.8 μL of 10-μM stock solutions of each of the barcoded forward (515F) and reverse (806R) primers, 0.8 μL of 10-mM 2-deoxynucleotide 5′-triphosphates (dNTPs; Promega), and 0.8 μL of template DNA. Cycling conditions were as follows: reactions were held at 98°C for 30 s with amplification proceeding for 25 cycles at 98°C for 10 s, 50°C for 10 s, 72°C for 10 s, and a final extension of 7 min at 72°C. PCR products were checked for correct size on a 1% agarose gel and subsequently combined and purified with the CleanPCR kit (CleanNA, Alphen aan den Rijn, The Netherlands). Purified PCR products were quantified with Qubit using the dsDNA BR Assay kit (Invitrogen, Carlsbad, CA), and a composite sample for sequencing was created by combining equimolar amounts of amplicons (200 ng) from the individual samples. The resulting library was sent to GATC Biotech AG for 2X150nt sequencing on an Illumina Hiseq2500 instrument. Sequence analysis was performed in NG-Tax using default settings (34). Paired-end libraries were demultiplexed using read pairs with perfectly matching barcodes. Amplicon sequence variants (ASV) were picked as follows: sequences were ordered by abundance per sample, and reads were considered valid when their cumulative abundance was ≤0.1%. Taxonomy was assigned using the SILVA reference database version 128 (6). ASVs are defined as individual sequence variants rather than a cluster of sequence variants with a shared similarity above a specified threshold such as operational taxonomic units.

The total bacterial counts in samples was assessed by quantitative PCR (qPCR). The qPCR measurements were performed in triplicates in 25-μL reactions in an iQ5 iCycler (Bio-Rad, Veenendaal, The Netherlands) using the iQ SYBR Green Supermix kit (Bio-Rad), with universal bacterial primers targeting the 16S rRNA gene: Eub341F(5-CCTACGGGAGGCAGCAG-3′), Eub534R (5′-ATTACCGCGGCTGCTGGC-3′) (30). The thermal cycling conditions included an initial DNA denaturation step at 95 °C for 10 min, followed by 40 cycles of denaturation at 95 °C for 20 s, annealing at 60°C for 30 s and extension at 72°C for 30 s. Standard curves for the assays were prepared using serial dilutions of purified full-length PCR amplicons amplified with universal primers 27F (5′-GTTTGATCCTGGCTCAG) - 1492R (5′-GGTTACCTTGTTACGACTT) from fecal genomic DNA. Gene copy numbers were calculated per gram of feces.

Statistical analysis.

Normality of data was assessed with the Gaussian distribution and Shapiro-Wilk test procedure, and not normally distributed variables were square root-transformed. Significant Spearman correlations (P < 0.05) between colonic and segmental colonic transit time with microbiota-related variables were included in multiple linear regression analysis. Age, sex, BMI, and protein and fiber intake were identified as covariates (i.e., changing standardized β-coefficient of the simple linear regression by more than 10%) of the association colonic transit, SCFA and BCFA, and therefore included as covariates in the linear regression. Multicollinearity was checked using variance inflation factor index, which was <1.5 for all analyzed variables. Statistical analyses were performed using SPSS 22.0 (IBM, Armonk, NY).

For the gut microbiota composition, all analyses were performed in R version 3.4. Distance-based redundancy analysis (db-RDA) with pairwise weighted UniFrac, unweighted UniFrac distance, and Bray Curtis dissimilarity were performed to determine the multivariate effects of the environmental variables on the microbiota composition using the capscale function from the vegan package (32). RDA is a technique summarizing the linear relationships between a set of variables, i.e., gut microbiota composition explained by a set of explanatory variables, or host variables. Missing values were imputed using the K-nearest neighbor algorithm as implemented in the bnstruct package (17). All environmental variables were scaled (mean set to zero) before the RDA was performed. To determine which set of environmental variables resulted in the most parsimonious model for each distance matrix, we performed forward and reverse automatic stepwise model selection for constrained ordination methods using permutation tests with the ordistep function from the vegan package, which bases the term choice on Akaike’s information criterion. For robustness, we performed the same automatic model selection using the ordiR2step function, which only performs forward model building and bases term choice solely on adjusted R² and P value. Both methods were in concordance with regard to the selection of significant terms. To determine the overlap and relationship in microbiota variation explained by the significant environmental variables, we used variation partitioning using the varpart function as implemented in vegan. Inverse Simpson, Shannon, ASV richness, and Faith’s phylogenetic diversity were calculated to define microbial α-diversity for each participant as implemented in vegan and picante (22). To determine whether α-diversity was dependent on sequencing depth or absolute bacterial count, we calculated correlations with each diversity metric and the sequencing depth and absolute bacterial count, respectively. Based on normality of the data, we used Spearman or Pearson correlation.

UniFrac and unweighted UniFrac, which are based on the phylogenetic relatedness of the ASVs, were used as estimators of microbial β-diversity between the participants. Whereas weighted UniFrac considers the abundance of each ASV, unweighted UniFrac provides equal weight to all ASVs, thereby focusing on presence or absence of low abundance ASVs. Additionally, we calculated Bray-Curtis dissimilarity, which maximizes the pairwise distance between individual samples between 0 and 1 (0 meaning two sites have the same composition, i.e., sharing all the species, and 1 meaning two sites do not share any species). Univariate partial Spearman correlations (corrected for age and sex) were calculated using the ppcor package, and multiple testing correction was performed using the Benjamini–Hochberg procedure (23). Corrected P values (Q values) of < 0.1 were considered significant.

Descending colonic transit correlates with microbiota α-diversity.

In total, 42 participants were included in this analysis (Table 1). Sequencing-based microbiota profiling was successful for n = 42 participants and yielded 470,928 ± SD 171,136 reads per sample. Spearman rank correlations showed that increased α-diversity was significantly associated with longer descending colonic transit time but not to total colonic transit time, BSS, or stool frequency (Fig. 2). These associations remained after correction for age and sex using partial Spearman correlations (Supplemental Table S1; all supplemental material is available at https://doi.org/10.6084/m9.figshare.9861134.v6). Importantly, technical covariates as well as absolute bacterial counts did not have an impact on α-diversity estimates (Supplemental Table S2).

Age, sex, diet, and stool consistency, but not colonic transit, are associated with microbiota variation.

Furthermore, we analyzed whether GI transit parameters as well as host parameters were linked to microbiota composition using distance-based redundancy analysis (db-RDA). We included (segmental) colonic transit time, stool consistency, age, sex, BMI, dietary intake, and microbial and gut-related parameters as listed in Table 1. We first selected the most parsimonious model using automatic model selection. Subsequently, to determine the relationship between these variables as well as their effect on the microbiota composition, we calculated their simple effects (i.e., the effect of the variable in isolation) and their conditional effects, i.e., the effect of the variable when all other significant variables were kept constant (Supplemental Table S3). Weighted UniFrac-based db-RDA showed that age, sex, and total fat intake were significantly related to microbiota variation. Total fat intake was inversely correlated with the relative abundance of Bifidobacterium. Age correlated mainly with the relative abundance of Prevotella 9 and Methanobrevibacter, and inversely with Bifidobacterium and two Blautia ASVs (Fig. 3A). Using unweighted UniFrac-based db-RDA, microbiota variation was explained by age, plasma LBP, protein intake, and fecal calprotectin (Fig. 3B). Total protein intake and fecal calprotectin correlated with the presence of Blautia and Bifidobacterium ASVs. Consistent with their relative abundance in weighted UniFrac, age correlated with the presence of Methanobrevibacter and was inversely correlated with the presence of Blautia and Bifidobacterium ASVs. Thus, not only the presence of these ASVs is associated with age but also their relative abundance. Using Bray-Curtis dissimilarity, only BSS significantly explained microbiota variation. BSS correlated with ASVs from the genus Faecalibacterium, Prevotella 9, and an unclassified member of the Lachnospiraceae family (Fig. 3C). Other GI transit parameters did not significantly explain microbiota variation using any of the metrics. Univariate testing of single ASVs or genera with single GI transit parameter partial Spearman rank correlations adjusted for age and sex showed no significant associations after adjustment for multiple testing (Supplemental Table S4).

Segmental colonic transit is associated with plasma acetate and fecal SCFA.

Next, we determined the relationship between the microbial metabolites listed in Table 1 and colonic and segmental colonic transit time. Multiple linear regression analysis showed that fecal butyrate, propionate, valerate, and caproate were consistently negatively associated with rectosigmoid transit time before and also after adjustment for age, sex, BMI, protein, and fiber intake (Table 2). Fasting plasma acetate was negatively associated with descending colonic transit. Additionally, fecal caproate was negatively associated with total colonic and ascending transit. Fecal BCFA isovalerate and isobutyrate were not related to (segmental) colonic transit time.