Gut-microbiota-targeted diets modulate human immune status

Abstract

Diet modulates the gut microbiome, which in turn can impact the immune system. Here, we determined how two microbiota-targeted dietary interventions, plant-based fiber and fermented foods, influence the human microbiome and immune system in healthy adults. Using a 17-week randomized, prospective study (n = 18/arm) combined with -omics measurements of microbiome and host, including extensive immune profiling, we found diet-specific effects. The high-fiber diet increased microbiome-encoded glycan-degrading carbohydrate active enzymes (CAZymes) despite stable microbial community diversity. Although cytokine response score (primary outcome) was unchanged, three distinct immunological trajectories in high-fiber consumers corresponded to baseline microbiota diversity. Alternatively, the high-fermented-food diet steadily increased microbiota diversity and decreased inflammatory markers. The data highlight how coupling dietary interventions to deep and longitudinal immune and microbiome profiling can provide individualized and population-wide insight. Fermented foods may be valuable in countering the decreased microbiome diversity and increased inflammation pervasive in industrialized society.

Trial registration: ClinicalTrials.gov NCT03275662.

Keywords: CAZymes; CyTOF; fermented food; fiber diet; immune system; immune system profiling; inflammation; microbiome; nutrition; proteomics.

Conflict of interest statement

Declaration of interests H.C.W. is a founder and shareholder of Interface Biosciences. W.V.T. is a founder and shareholder of Interface Biosciences. J.L.S. is a founder, shareholder, and on the scientific advisory board of January AI and Novome Biotechnologies.

Copyright © 2021 Elsevier Inc. All rights reserved.

Figures

Figure 1.. Overview of Fiber Fermented Food Study.

(A) Consort flow diagram for participant enrollment, allocation, follow-up, and analysis. Side chart shows the number of participants in high-fiber (Fi) and high-fermented (Fe) diet arm collected for each platform. *2 participants were assigned to high fiber, not randomized, by special request. (B) The 14-week study overview timeline, sample types collection, and corresponding experimental platforms. (C) Fiber intake in the high-fiber diet arm shown in boxplots, fiber intake in the high-fermented food diet arm shown as dotted line. (D) Fermented food intake in the high-fermented food diet arm shown in boxplots, fermented food intake in high-fiber diet arm shown as dotted line. P-values ≤ 0.05 via t-test denoted by asterisks and calculated for each time point relative to baseline −2 week value.

Figure 2.. Diet-specific effects of a fiber vs. fermented food intervention on the host and microbiome.

(A) Accuracy of leave-one-out cross-validation (LOOCV) of random forest models predicting diet group; separate models using host-derived data (white bars) or microbe-derived data (black bars), using parameter changes from baseline to end of maintenance as model features. Recursive feature elimination chose the minimum number of parameters needed for maximum accuracy. (B) Differences in myosin-1, model feature selected for host proteomics model. (C) Differences in rank-order change of Lachnospira, model feature selected for 16S amplicon sequence variants (ASVs) model. Purple, fermented group; green, fiber group.

Figure 3.. Participants consuming fiber exhibit shifts in the functional profile of the microbiome.

(A) Observed number of amplicon sequence variants (ASVs) from 16S rRNA amplicon sequencing; no significant changes during any intervention time point compared to baseline (Week −2 or Week 0) (paired t-test). (B) Proteins measured using LC-MS (Gonzalez et al., 2020) were categorized as human or microbe derived using the HMP1 database (The Human Microbiome Project Consortium, 2012). Microbe proteins as a percent of total stool proteins increase from baseline to end of maintenance phase (Week 10, p-value=0.003 from Week −2, p-value=0.01 from Week 0, paired t-test). (C) CAZymes identified from metagenomic sequencing as significantly changing in relative abundance from baseline to end of maintenance phase (FDR ≤ 0.05, q-value ≤ 0.1, SAM two-class paired). CAZymes were annotated using dbCan and assigned to functional categories (Yin et al., 2012; Cantarel et al., 2012). (D) Significant decreases in two branched chain fatty acids and valeric acid in stool (p-value=0.044, 0.033, 0.033, paired t-test). Outliers not plotted but all values were included for statistical analysis testing (see Methods). (E) Total fiber intake (grams) correlated with percentage of carbohydrates in stool using linear mixed effects (LME) model (p-value=8e-4).

Figure 4.. Fiber-consuming participants exhibit varied immune responses that track with differences in microbiome composition and diversity.

(A) Immune features derived from immunophenotyping assays. (B) Heatmap depicting differences in immune features (grouped by feature type) from baseline (Week −3) to end of intervention (Week 10), rescaled from minimum change > −1 to maximum change < 1, no change=0. Each row is a participant, rows are clustered using hierarchical clustering by feature values in the fiber arm. (C) Counts of the mean positive (red) or mean negative (blue) changes in endogenous immune cell signaling from baseline (Week −3) to end of maintenance (Week 10) for the three clusters. Non-significant changes shown in light color, significant changes shown in dark color (SAM, two-class paired, q-value ≤ 0.1). (D) Average number of observed ASVs at baseline (Week −2 and Week 0) high-inflammation cluster (red), low-inflammation i cluster (gold), and low-inflammation ii cluster (blue) (unpaired t-test significant p-value = 0.037). (E) Significant taxa binned using tip_glom (phyloseq package in R) identified in pairwise comparisons using a zero-inflated beta regression, plotted over time. Percentage of participants with taxa present in high-inflammation (red) and low-inflammation i (gold) clusters shown in the first three panels (group logarithmic model adjusted p-value ≤ 0.05); abundance (percent of composition) of high-inflammation (red) and low-inflammation ii (blue) clusters shown in fourth panel (group beta regression model adjusted p-value ≤ .05); abundance (percent of composition) of low-inflammation i (gold) and ii (blue) clusters shown in fifth panel (group beta regression model adjusted p-value ≤ 0.05).

Figure 5.. High-fermented food diet increased microbiota diversity and altered composition.

(A, B) Observed ASVs (A) (p-values generated using paired t-test) and Shannon diversity (B) increased from baseline through choice phase. Observed ASVs significantly correlated with time using linear mixed effects (LME) model (p-value=2.3e-3 for Observed ASVs, p-value=1.4e-3 for Shannon). (C) Total fermented food intake, yogurt, and vegetable brine drinks positively correlated with observed ASVs using linear mixed effects (LME) model (p-value adjusted ≤ 0.05). (D) Rank normalized ASVs that were significantly correlated with fermented food consumption over time using an LME model (p-value adjusted ≤ 0.05). Graphs are colored by taxonomic family. (E) New ASVs (not present at baseline weeks −2 or 0 but detected at any other time during the intervention) that were detected in fermented foods were aggregated and summed for each participant and plotted as a percentage of all new ASVs by time point for the high-fermented food diet arm. Dotted line indicates trend for high-fiber diet arm.

Figure 6.. Fermented food consumption decreases levels of inflammation.

(A) Cytokines, chemokines, and other serum proteins plotted that change significantly from baseline (week −3) to end of intervention (week 10) (SAM two-class paired, FDR ≤ 0.05, q-value ≤ 0.1). Negative correlations for levels of each analyte across time calculated using LME. NPX refers to the normalized protein expression used by Olink Proteomics’ log2 scale. Fgf-21 also significantly decreased across time (data not shown). (B) Cell type-specific endogenous signaling proteins, measured using CyTOF, that change significantly from baseline (week −3) to end of intervention (week 10) (SAM two-class paired, FDR < 10%). Arcsinh ratio plotted from week −3 to week 10. (C) Fold change of cell frequencies (calculated as percentage of CD45+ cells) that change significantly from baseline (week −3) to end of intervention (week 10) (Wilcoxon paired test, adjusted p-value ≤ 0.05).

Figure 7.. Interaction between the host immune system and microbiota.

(A) Correlation of difference between baseline and end of maintenance was calculated for each parameter and percent of significant pairwise correlations between microbe and host assays were plotted. Light grey denotes correlations with a p-value adjust ≤ 0.05, dark grey shows p-value adjust ≤ 0.01 (corrected using Benjamini-Hochberg hypothesis correction). (B) Positive and negative correlations between host proteins annotated by disease or function (source: Ingenuity Pathway Analysis) and CAZymes. (C) Changes in stool butyrate levels vs. blood B-cell frequency changes from baseline (week −3 stool, week −2 blood) to end of maintenance (week 10) for both high-fiber (green) and high-fermented food (purple) arms. B cells are defined as CD45+CD66-CD3-CD19- (Figure S3); frequency quantified as B cell frequency as a fraction of CD45+CD66- cells. (D) Correlations between CAZymes (colored by CAZy family) and immune cells signaling capacity for various cell types and stimulatory cytokines.