Metabolic Responses to Butyrate Supplementation in LF- and HF-Fed Mice Are Cohort-Dependent and Associated with Changes in Composition and Function of the Gut Microbiota

Animals were maintained and handled in accordance with protocols approved by the Institutional Animal Care and Use Committee (University of California, Davis, CA, USA) (18520, Approved Feb-2015). Male C57BL/6J mice (6 weeks old) were purchased from Jackson Laboratory (Sacramento, CA, USA) in two separate cohorts (cohort 1 and cohort 2; n = 6/group/cohort), with a 2-week interval for assessment of reproducibility of experimental paradigm. After acclimation for a week, mice were randomly assigned into 4 groups and fed either a low-fat diet (LF/CON; 10% kcal as fat; D12450J; Research Diets, New Brunswick, NJ, USA), a high-fat diet (HF/CON; 45% kcal as fat; D08091803B; Research Diets), or an LF or HF diet with mono- and diglycerides of butyric acid (MB; ProPhorce^TM SR 710, Perstorp, Malmö, Sweden; LF/MB or HF/MB) at 0.25% (w/v) in drinking water for 6 weeks (Table S1). Mice were group-housed (2–4 mice/cage) at 22 °C with 12 h: 12 h light–dark cycles and had ad libitum access to food. Body weight and water intake were measured once a week. At week 5 before feeding behavior testing, fresh fecal samples were collected from each mouse for profiling of microbiota and metabolites. After 6 weeks on the respective treatments, mice were fasted for 6 h and euthanized using pentobarbital (Fatal Plus, Vortech Pharmaceuticals, Dearborn, MI, USA; 300 mg/kg; ip). Blood was collected via cardiocentesis in K3-EDTA tubes and centrifuged at 1000× g for 10 min at 4 °C for plasma collection. Visceral fat pads for adiposity assessment and gut tissues for determination of intestinal permeability (see below) were collected at necropsy from n = 3/group/cohort. The remaining mice were perfused with paraformaldehyde for obtaining fixed tissue for immunocytochemistry (data not included). Plasma and tissues were snap-frozen and stored at −80 °C until analysis.

After 5 weeks on the respective diets, sensitivity to the satiating effect of CCK was tested. Mice were singly housed and fasted in wire-bottom cages for 6 h during the light phase. CCK (octapeptide, sulfated, Bachem, Torrance, CA, USA, 100 µL at 3 µg/kg; ip) or saline (100 µL; ip) was administered and individual food was placed in the cage and food intake was recorded every 20 min for 60 min.

Fecal homogenates were prepared by adding pre-weighed fecal pellets (1–2) to a 2 mL screwcap tube with 0.1 mm diameter silica beads and 4-mm diameter glass bead and 500 mL fecal extract buffer (phosphate buffered saline + 0.1% Tween-20) and by homogenizing for 40 s at 6.0 M/s at room temperature in the MPBio Fastprep-24 5G homogenizer, and then centrifuging for 5 min at >12,000× g at 4 °C. The supernatant was retained for assay. The levels of fecal lipocalin-2 (Lcn-2; R&D Systems, Minneapolis, MN, USA) were detected via enzyme-linked immunosorbent assay (ELISA) according to the manufacturer’s protocol using the homogenate supernatant.

Colon tissue was opened along the mesenteric border, and duplicate samples were mounted in Ussing chambers (Physiologic Instruments, San Diego, CA, USA), exposing 0.1 cm² of tissue surface area to 2.5 ml of oxygenated Krebs-glucose (10 mM) and Krebs-mannitol (10 mM) at 37 °C on the serosal and luminal sides, respectively. The paracellular pathway and transcellular pathway were measured as the flux of FITC-Dextran 4000 (FD-4, Sigma Aldrich, St. Louis, MO, USA) and horseradish peroxidase (HRP Type VI, Sigma Aldrich, St. Louis, MO, USA), respectively. FD-4 (400 µg/mL) and HRP (200µg/mL) were added to the mucosal chamber and samples were collected from the serosal chamber every 30 min for 2 h. Concentration of FD-4 was measured via fluorescence at an excitation of 485 nm and an emission of 538 nm. O-dianisidine substrate was used to detect HRP at an absorbance of 450 nm. Flux was expressed as ng of FD4 or HRP transported per cm² of membrane per hour. Tissue samples with >1.2 × 10⁴ ng/cm²/h of FD4 were considered to be damaged and excluded for both HRP and FD4. In addition, tissue replicate samples that varied by more than five-fold or which became damaged while in the Ussing chambers were also excluded. Flux is calculated as the mean of tissues replicates (where available) 60 min after the addition of FD4 and HRP.

Fresh fecal samples were collected at necropsy and stored at −80 °C until analysis. Samples were extracted with nano-pure water (10 mg/mL) at 4 °C with gentle agitation overnight. The homogenates were centrifuged at 21,000× g for 5 min, and the supernatants were transferred and centrifuged at 21,000× g for additional 20 min. For each fecal sample, aliquots of 20 µL of supernatant were reacted with 20 µL of 200 mM N-(3-Dimethylaminopropyl)-N-ethylcarbodiimide hydrochloride in 5% pyridine and 40 µL of 100 mM 2-nitrophenylhydrazine in 80% acetonitrile /H₂O (v/v) with 50 mM HCl. Mixtures were incubated for 30 min at 40 °C and the reaction was terminated by the addition of 400 µL of 10% ACN/H₂O (v/v). Samples were centrifuged and transferred into a 96-well injection plate for analysis in an Agilent 6490 triple quadrupole mass spectrometer equipped with the Agilent 1290 infinity LC system (Agilent, Santa Clara, CA,). Chromatographic separations were carried out on an Agilent InfinityLab Poroshell 120 EC-C18 stationary phase (2.1 × 100 mm, 1.9 µm column). Mobile phases were 100% ACN (B) and water with 10% ACN (A). Flow rate was 0.3 mL/min and injection volume was 1 μL. Samples were held at 4 °C in the autosampler, and the column was operated at 40 °C. The mass spectrometer was operated in the multiple reaction monitoring method and positive ionization mode. Parameters were optimized with SCFA standards, and calibration curves were generated for all metabolites measured with corresponding true standards.

Stand-alone PICRUSt2 was installed and used to run the full pipeline with picrust2_pipeline.py and specifying–per_sequence_contrib as additional parameters using fasta sequences and biom table [24,25]. To examine differences between cohorts, ALDEx2 (ANOVA-like Differential Expression) was used to normalize the PICRUSt2 pathway data by performing log ratio transformation to normalize and Welch’s t and Wilcoxon rank sum tests to statistically test [26,27,28]. A cutoff of Benjamini–Hochberg-adjusted p < 0.05 for either was used to identify differentially regulated pathways. A similar approach using ALDEx2 was taken to identify taxa that showed statistically significant divergent populations between cohorts. Principal component and dendogram analyses were used to confirm accurate assignment of samples to the correct cohort. Different taxa (ASVs) contributing to the differentially regulated Metacyc pathways were extracted from PICRUSt2 and visualized using stacked bar charts. Heatmaps, generated with pheatmap in R, were used to visualize the normalized expression value of the predicted pathways for each sample.

Unless stated otherwise, statistical analysis was performed using Prism software (Prism 8.4; GraphPad Software, La Jolla, CA, USA). Data are presented as means ± standard error of the mean (SEM)s. The ROUT test was used to exclude outliers. Two-way analysis of variance (ANOVA) followed by Tukey’s (when the interaction was significant) or Sidak’s (for the main effects) post-hoc test was used to test for differences among groups. Repeated measures 2-way ANOVA was performed for body weight and body weight gain using group assignment (diet x treatment) and time with Sidaks’s post hoc multi-comparisons test. A paired Student’s t-test was used to determine statistical significance within groups for CCK sensitivity. Differences were considered significant if p < 0.05. Spearman correlations were run using sample metadata and microbiota data using packages Hmisc and corrplot for visualization in R.

In cohort 1 (Figure 1A), from the fourth week of diet treatment and throughout the remaining study period, HF feeding significantly increased body weight compared to LF-fed controls (LF/CON vs. HF/CON, p < 0.05 at weeks 5 and 6; LF/MB vs. HF/MB p < 0.05 at weeks 5 and 6). In cohort 2 (Figure 1B), HF feeding also resulted in significant increases in body weight compared to LF-fed groups (LF/CON vs. HF/CON and LF/MB vs. HF/MB, p < 0.05 at the fourth, fifth, and sixth weeks). MB supplementation had no significant effect on body weight in either LF- or HF-fed mice in either cohort 1 or 2. HF feeding significantly increased visceral adiposity compared to control groups (LF/CON vs. HF/CON, p < 0.05) in both cohorts. In cohort 2 (Figure 1F), visceral adiposity was significantly increased by HF feeding in both control and MB-treated mice (LF/CON vs. HF/CON, p < 0.05; LF/MB vs. HF/MB, p < 0.01).

Fecal lipocalin-2 (Lcn-2) was measured as a marker of intestinal inflammation [29]. In cohort 1 (Figure 2A), the fecal Lcn-2 level was significantly increased by HF feeding (LF/CON vs. HF/CON, p < 0.05). In cohort 2 (Figure 2B), there was a trend for an increase in fecal Lcn-2 in HF/CON mice, but this did not reach statistical significance (LF/CON vs. HF/CON, p = 0.0885). There was no significant effect of MB supplementation on fecal Lcn-2 level in either cohort. However, there was a trend for a decrease in fecal Lcn-2 in HF/MB compared to HF/CON mice in cohort 1, but this did not reach statistical significance.

Intestinal permeability, expressed as paracellular and transcellular permeability, was measured ex vivo in Ussing chambers in colonic tissue collected at necropsy. In cohort 1, the HF diet increased intestinal permeability and there was a trend for a decrease in paracellular permeability in response to MB (Figure 3). In cohort 2, there was no significant effect of diet or treatment on permeability.

The ability of exogenous CCK to suppress energy intake was determined to measure the effect of MB supplementation on intestinal satiety signaling. In both cohorts 1 and 2 (Figure 4A,B), CCK administration significantly reduced food intake compared to the saline control in both control and MB-treated groups fed an LF diet (cohort 1: CCK vs. saline, p < 0.01 LF/MB and p < 0.05, LF/CON; cohort 2: CCK vs. saline, p = 0.01, LF/CON and p < 0.05 LF/MB). CCK had no significant effect on food intake in HF-fed mice (CCK vs. saline, p > 0.05, HF/CON) in both cohorts. However, in HF-fed mice given MB, CCK significantly decreased food intake (CCK vs. saline, p < 0.05, HF/MB) in both cohorts.

The fecal microbiota was taxonomically classified to profile the bacterial community. Measures of alpha and beta diversity were evaluated (Figure 5). Alpha diversity metrics showed significantly lower biodiversity in cohort 2 for both observed species (Kruskal–Wallis pairwise p = 0.002) and Faith’s phylogenetic diversity (p < 0.001) (Figure 5A,B). A clear difference between cohorts was also seen in beta diversity for both phylogenetic (UniFrac) and non–phylogenetic (Bray-Curtis or Jaccard) measures of beta diversity using ANOSIM (p ≤ 0.002) (Supplemental Table S1). At the genus level, the partial least squares discriminant analysis (PLS-DA) was implemented to characterize the effect of each primary variable: diet (Figure 5C), treatment (Figure 5D), or cohort (Figure 5E,F). Consistent with diversity measures, the separation in the bacterial communities between cohorts was greater than the effect of either diet (LF or HF) or treatment (CON or MB).

To identify differentially abundant microbiota taxa characterizing the cohort effect, several complementary methods, Analysis of the Composition of Microbiomes (ANCOM), ALDEx2, and linear discriminant effect size (LEfSe) were used. There was a strong consensus between the methods.lists the genera identified by ALDEx2 that are differentially abundant between cohorts, which include members from 2 families within Actinobacteria, 3 families within the Bacteroidetes, 23 taxa within the Firmicutes, 1 genus of Proteobacteria, and 1 class from Tenericutes. The cladogram showing the taxonomical structure between the differential taxa identified by LEfSe is shown in. Supplemental Table S2 Supplemental Figure S1

In cohort 1 (Figure 6A,B), HF feeding led to significant changes in microbial metabolites compared to the LF control group. In particular, concentrations of butyric, valeric, and glycolic acids were significantly elevated in the HF/MB group compared to the LF/CON group (LF/MB vs. HF/MB, p < 0.05 for butyric and valeric acids and p < 0.01 for glycolic acid). MB supplementation had no effect on the metabolite profile. In cohort 2, there were significant differences in levels of butyric, valeric, and indole-3-lactic acids between LF- and HF-fed mice (LF/MB vs. HF/MB, p < 0.01, <0.05, and <0.01 for butyric, valeric, and indole-3-lactic acids, respectively). There was no significant effect of MB supplementation on most microbial metabolites; however, there was a significant increase in the level of butyric acid in HF-fed groups (HF/CON vs. HF/MB, p < 0.05) in cohort 2. In cohort 1, MB treatment significantly reduced hexanoic acid in the HF-fed mice, but this effect was not observed in cohort 2.

The complete list of metabolites measured and the results of 3-way ANOVA showing effect of cohort, treatment and diet is presented in. There were significant cohort differences in isobutyric, valeric, isovaleric, and glycolic acids, and in indole-3-acetic acid. Supplemental Table S3

To understand the potential functional changes in the microbiota between cohorts, we employed PICRUSt2 for metagenome inference using the 16S amplicon sequencing data and ALDEx2 to identify the differentially abundant pathways. We identified 51 Metacyc pathways that differed between cohorts and the contribution of the different taxa to those pathways. The top categories of microbial functions that demonstrated cohort specific regulation included amine and polyamine degradation, carbohydrate biosynthesis and degradation, cofactor, prosthetic group, electron carrier or vitamin synthesis, fermentation, nucleoside and nucleotide biosynthesis, polymeric compound degradation, secondary metabolite degradation, and the tricarboxylic acid cycle (Supplemental Table S4). A higher cutoff was used to identify those pathways with the most robust differential expression based on the metagenome inference (Figure 7). In this set of the 14 most highly regulated predicted pathways, 11 Metacyc pathways (P23-PWY, PWY-6749, FUCCAT-PWY, P441-PWY, PWY- 6263, PWY-7371, P461-PWY, PWY-7003, PWY-7198, PWY-7210, and GLYCOLYSIS-E-D) were decreased and only 3 pathways were increased (LACTOSECAT-PWY, PWY-5304, and PWY-6572) in cohort 2 compared to cohort 1. Important functions for replication and growth, such as pyrimidine deoxyribonucleotide biosynthesis (PWY-7198, PWY-7210), fermentation (PWY-7003), glycolysis (GLYCOLYSIS-E-D), and menaquinone biosynthesis (redox component of electron transfer chain- PWY-6263, PWY-7371)) were reduced in cohort 2. Different populations of bacteria were predicted to contribute to these pathways. In cohort 1, the increased abundance of the genus Alistipes in the family Rikenellaceae of the Bacteroidetes phylum contributed to the differential abundance of PWY-7371 (Figure 7B, left panel). In cohort 2, abundance of the genus Bacteroides specifically contributed to PWY-6572, demonstrating the unique functional potential of the microbiota in the two cohorts (Figure 7B, right panel).

We performed Spearman correlational analyses using a heatmap for visualization to elucidate potential relationships between physiological phenotypes, microbiota, and other metadata in the combined data from both cohorts (Figure 8). As expected, the magnitude of CCK-induced inhibition of food intake was correlated to diet, body weight, and fat pad weight. However, somewhat unexpectedly, the marker of intestinal inflammation, lipocalin-2, did not correlate to any variables. Other intestinal measures, such as permeability in the ileum, similarly did not show strong correlations to diet or treatment. Of all the metadata, only one parameter, the family Streptococcaceae, correlated to treatment. The metabolites lactic acid and valeric acid had opposing correlations to diet. Glycolic, isobutyric, isovaleric, and indole 3 lactic acids demonstrated strong correlations to cohort. However, the majority of the correlations were in the microbiota. Unsurprisingly, many microbiota taxa were correlated to diet, but there was a remarkable number of correlations to cohort. When PLS-DA was used to identify the signatures that differentiated the cohorts using all the project metadata, the microbiota dominated the loadings plot and variable importance in projection (VIP) list of the most relevant variables. These approaches highlight the variability in microbiota, even under conditions when mice were housed concurrently in the same room in the vivarium, fed an identical batch/lot of diet, and provided with same preparation of treatment.